CN117994255B - Anal fissure detecting system based on deep learning - Google Patents
Anal fissure detecting system based on deep learning Download PDFInfo
- Publication number
- CN117994255B CN117994255B CN202410402866.9A CN202410402866A CN117994255B CN 117994255 B CN117994255 B CN 117994255B CN 202410402866 A CN202410402866 A CN 202410402866A CN 117994255 B CN117994255 B CN 117994255B
- Authority
- CN
- China
- Prior art keywords
- module
- feature
- probability
- image
- diagnosis
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 206010002153 Anal fissure Diseases 0.000 title claims abstract description 80
- 208000016583 Anus disease Diseases 0.000 title claims abstract description 80
- 208000009531 Fissure in Ano Diseases 0.000 title claims abstract description 80
- 238000013135 deep learning Methods 0.000 title claims abstract description 27
- 238000004458 analytical method Methods 0.000 claims abstract description 116
- 238000003745 diagnosis Methods 0.000 claims abstract description 107
- 238000005457 optimization Methods 0.000 claims abstract description 79
- 238000011156 evaluation Methods 0.000 claims abstract description 75
- 238000010276 construction Methods 0.000 claims abstract description 40
- 238000013507 mapping Methods 0.000 claims abstract description 32
- 238000001514 detection method Methods 0.000 claims abstract description 30
- 238000000034 method Methods 0.000 claims abstract description 30
- 230000008569 process Effects 0.000 claims abstract description 16
- 238000013398 bayesian method Methods 0.000 claims abstract description 5
- 230000002708 enhancing effect Effects 0.000 claims abstract description 5
- 239000013598 vector Substances 0.000 claims description 101
- 238000004422 calculation algorithm Methods 0.000 claims description 21
- 238000010586 diagram Methods 0.000 claims description 19
- 238000004364 calculation method Methods 0.000 claims description 14
- 238000012549 training Methods 0.000 claims description 14
- 230000006870 function Effects 0.000 claims description 12
- 230000006872 improvement Effects 0.000 claims description 11
- 230000010354 integration Effects 0.000 claims description 10
- 230000008859 change Effects 0.000 claims description 9
- 238000003708 edge detection Methods 0.000 claims description 9
- 230000003993 interaction Effects 0.000 claims description 9
- 238000012545 processing Methods 0.000 claims description 9
- 230000000007 visual effect Effects 0.000 claims description 8
- 230000000877 morphologic effect Effects 0.000 claims description 7
- 239000003623 enhancer Substances 0.000 claims description 6
- 230000007797 corrosion Effects 0.000 claims description 5
- 238000005260 corrosion Methods 0.000 claims description 5
- 238000004088 simulation Methods 0.000 claims description 5
- 230000002159 abnormal effect Effects 0.000 claims description 4
- 230000004913 activation Effects 0.000 claims description 4
- 239000002131 composite material Substances 0.000 claims description 4
- 230000008030 elimination Effects 0.000 claims description 4
- 238000003379 elimination reaction Methods 0.000 claims description 4
- 238000001914 filtration Methods 0.000 claims description 4
- 230000009466 transformation Effects 0.000 claims description 4
- 238000013139 quantization Methods 0.000 claims description 3
- 238000000605 extraction Methods 0.000 abstract description 4
- 238000007781 pre-processing Methods 0.000 abstract description 3
- 238000005516 engineering process Methods 0.000 description 5
- 208000025174 PANDAS Diseases 0.000 description 4
- 208000021155 Paediatric autoimmune neuropsychiatric disorders associated with streptococcal infection Diseases 0.000 description 4
- 240000004718 Panda Species 0.000 description 4
- 235000016496 Panda oleosa Nutrition 0.000 description 4
- 238000003384 imaging method Methods 0.000 description 3
- 238000000540 analysis of variance Methods 0.000 description 2
- 238000003066 decision tree Methods 0.000 description 2
- 238000013136 deep learning model Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 230000008451 emotion Effects 0.000 description 2
- 239000000463 material Substances 0.000 description 2
- 238000003058 natural language processing Methods 0.000 description 2
- 238000007619 statistical method Methods 0.000 description 2
- 238000012360 testing method Methods 0.000 description 2
- 238000013076 uncertainty analysis Methods 0.000 description 2
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 description 1
- 238000013473 artificial intelligence Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000004140 cleaning Methods 0.000 description 1
- 239000003086 colorant Substances 0.000 description 1
- 238000012937 correction Methods 0.000 description 1
- 230000007812 deficiency Effects 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000002405 diagnostic procedure Methods 0.000 description 1
- 201000010099 disease Diseases 0.000 description 1
- 208000037265 diseases, disorders, signs and symptoms Diseases 0.000 description 1
- 230000004927 fusion Effects 0.000 description 1
- 238000011478 gradient descent method Methods 0.000 description 1
- 238000011835 investigation Methods 0.000 description 1
- 238000003064 k means clustering Methods 0.000 description 1
- 238000007477 logistic regression Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000011056 performance test Methods 0.000 description 1
- 238000000611 regression analysis Methods 0.000 description 1
- 230000004044 response Effects 0.000 description 1
- 238000012216 screening Methods 0.000 description 1
- 230000001629 suppression Effects 0.000 description 1
- 238000011477 surgical intervention Methods 0.000 description 1
- 238000002604 ultrasonography Methods 0.000 description 1
Landscapes
- Image Analysis (AREA)
- Measuring And Recording Apparatus For Diagnosis (AREA)
Abstract
The invention discloses an anal fissure detection system based on deep learning in the technical field of surgical detection, which comprises an image quality enhancement module, a feature mapping construction module, a logic relation analysis module, a model parameter optimization module, a probability evaluation construction module, a comprehensive decision support module and a performance feedback optimization module. According to the invention, through the image quality enhancement module and the feature mapping construction module, the accuracy of image preprocessing and feature extraction is improved, so that the anal fissure is more accurately and efficiently identified. In addition, the logic relation analysis module and the probability evaluation construction module not only strengthen the connection between the image features and the medical knowledge base, but also introduce a Bayesian method to quantitatively analyze the uncertainty of the diagnosis result, thereby enhancing the information support strength of the diagnosis decision. The comprehensive decision support module enables the image recognition result, the logic reasoning analysis and the probability evaluation to be comprehensively considered in the diagnosis process, and provides a comprehensive decision basis.
Description
Technical Field
The invention relates to the technical field of surgical detection, in particular to an anal fissure detection system based on deep learning.
Background
Anal fissure detection systems belong to the field of surgical detection technology, focusing on the diagnosis and assessment of conditions requiring surgical intervention using various devices and methods. The physician performs a more accurate and detailed examination of the patient by non-invasive or minimally invasive means to determine the nature and severity of the disease. Surgical detection techniques cover imaging examination (e.g., X-ray, MRI, ultrasound) to molecular diagnostic methods and computer-based analytical processing such as deep learning and artificial intelligence aided image recognition. Thereby providing faster and more accurate diagnostic information so that a treatment plan can be rapidly formulated, and the treatment result and quality of life of the patient can be improved.
Among them, the anal fissure detection system is a specific application in the field of surgical detection, and automatically recognizes and evaluates the existence of anal fissure and its severity through a deep learning algorithm. The system identifies the characteristics of anal fissure, such as the length, depth, position and other information of the fissure, by analyzing the image of the anal region, thereby providing accurate diagnosis basis for doctors. The accuracy and efficiency of anal fissure diagnosis are improved, and the dependence on traditional physical examination is reduced, so that discomfort of patients and invasiveness in the diagnosis process are reduced.
Although the prior art has made a certain progress in anal fissure detection, the treatment result and the quality of life of a patient can be improved by a deep learning algorithm assisted by an imaging examination, the uncertainty analysis in the process of diagnosis treatment cannot sufficiently quantify and explain the confidence and probability distribution of different diagnosis results, so that a doctor lacks sufficient information support in making a diagnosis decision. In addition, the difficulty of the system to translate the image features identified by the deep learning model into data usable by the logic analysis system limits the diagnostic system's ability to provide comprehensive, logic-based diagnostic advice.
Based on the above, the invention designs an anal fissure detection system based on deep learning to solve the above problems.
Disclosure of Invention
The invention aims to provide an anal fissure detection system based on deep learning, which solves the problems that although the prior art in the background technology has made a certain progress in anal fissure detection, the treatment result and the life quality of a patient can be improved by a deep learning algorithm assisted by an imaging examination mode, but the confidence and probability distribution after different diagnosis results are not fully quantized and interpreted in the aspect of uncertainty analysis in the process of diagnosis treatment, so that doctors lack sufficient information support when making diagnosis decisions. In addition, the difficulty of the system to translate the image features identified by the deep learning model into data usable by the logic analysis system limits the ability of the diagnostic system to provide comprehensive, logic-based diagnostic advice.
In order to achieve the above purpose, the present invention provides the following technical solutions: the anal fissure detection system based on deep learning comprises an image quality enhancement module, a feature mapping construction module, a logic relation analysis module, a model parameter optimization module, a probability evaluation construction module, a comprehensive decision support module and a performance feedback optimization module;
the image quality enhancement module carries out pretreatment on an input anal fissure image, and comprises the steps of removing noise, executing corrosion and expansion operation, highlighting anal fissure characteristics in the image, enhancing image contrast and edge definition through open operation and closed operation, and generating an optimized image set;
the feature mapping construction module is used for extracting key visual features in the image based on the optimized image set, including edge detection, color intensity analysis and morphological features, constructing feature vectors describing anal fissure and surrounding tissues thereof, and generating a feature vector set;
the logic relationship analysis module utilizes the feature vector set to match the image features with anal fissure features in a predefined medical knowledge base, analyzes the logic relationship among the features, constructs a corresponding logic relationship graph and generates a feature logic analysis result;
The model parameter optimization module takes the feature vector set as input, performs parameter optimization operation, and comprises the steps of adjusting network hierarchical structure, learning rate and regularization parameters, optimizing the ability of the model in identifying and classifying anal fissure images, and generating optimized model parameters;
The probability evaluation construction module performs probability analysis based on the feature logic analysis result and the optimized model parameters, quantifies the uncertainty of the diagnosis result by using a Bayesian method, and generates a probability evaluation result by calculating probability distribution under various diagnosis hypotheses;
The comprehensive decision support module is used for analyzing the influence of the change source of the diagnosis result and the probability evaluation on the diagnosis decision by combining the feature logic analysis result and the probability evaluation result, integrating all analysis contents and recommended diagnosis directions, and generating a comprehensive analysis information set;
The performance feedback optimization module collects feedback of doctors and patients based on the comprehensive analysis information set, and performs iterative optimization of the model and the flow according to the collected feedback to generate a performance optimization model.
Preferably, the optimized image set includes an image with improved edge definition, an image with enhanced contrast, and an image after noise processing, the feature vector set includes a vector of an edge detection result, a numerical vector of color intensity, and a description vector of morphological features, the feature logic analysis result includes a logic mapping of feature matching, a logic atlas of a relationship between features, and a result set of logic reasoning, the optimized model parameters include a network hierarchical depth set value, a learning rate adjustment value, and a regularization parameter optimization value, the probability evaluation result includes a probability value, an uncertainty quantization index, and a confidence evaluation range of each diagnosis hypothesis, the comprehensive analysis information set includes a source analysis of diagnosis uncertainty, a probability evaluation comprehensive interpretation, and a decision basis recommending a diagnosis direction, and the performance optimization model includes an improvement measure of a performance index, an integration suggestion of user feedback, and an iterative optimization parameter adjustment.
Preferably, the image quality enhancement module comprises an image denoising processing sub-module, an image characteristic enhancement sub-module and an image contrast and definition optimization sub-module;
the image denoising processing submodule carries out frequency filtering on the input anal fissure image, eliminates frequency abnormal parts, adjusts brightness and contrast parameters and generates a denoised image;
the image characteristic enhancer module performs edge enhancement on the denoised image, adjusts pixel value distribution, highlights an anal fissure region at surrounding tissues, and improves detail visibility by applying local contrast adjustment to obtain a characteristic enhanced image;
The image contrast and definition optimizing submodule adjusts the image contrast and the edge definition through open operation and close operation based on the characteristic-enhanced image, optimizes the overall and local brightness balance and constructs an optimized image set.
Preferably, the feature map construction module comprises an edge definition analysis sub-module, a color intensity analysis sub-module and a shape and structure identification sub-module;
The edge definition analysis submodule carries out difference calculation on pixel intensity around each pixel based on the optimized image set, highlights the edge part of the image, applies nonlinear transformation to adjust the image contrast, strengthens the visibility of key edge lines, refines edge representation by utilizing edge tracking, and generates an edge feature mapping set;
The color intensity analysis submodule receives the edge feature mapping set, uniformly samples the image in a color space, analyzes the difference of pixel distribution in each color channel, adjusts the color distribution, enhances the color contrast between anal fissure and surrounding tissues, and obtains a color difference mapping set;
The shape and structure recognition submodule is used for recognizing shape features associated with anal fissure by adjusting parameters based on a color difference mapping set through analyzing geometric figures in an image, and constructing a composite visual feature expression through a layer-by-layer superposition analysis method by combining the edge feature mapping set to obtain a feature vector set.
Preferably, the logic relationship analysis module comprises a feature and medical standard comparison sub-module, a feature logic relationship analysis sub-module and a medical logic relationship diagram construction sub-module;
Each feature in the feature vector set is scanned item by the feature and medical standard comparison submodule, the feature and the feature are compared with anal fissure features recorded in a medical knowledge base, the feature similarity is scored by adopting a cosine similarity algorithm, and a key index is screened by utilizing a recursive feature elimination method to obtain a key feature matching result;
The cosine similarity algorithm is according to formula I:
I is a kind of
Calculating cosine value evaluation similarity between the feature vector and the anal fissure feature vector recorded in the medical knowledge base, and generating a key feature matching result;
Wherein, Is one feature vector in the feature vector set,/>Is anal fissure characteristic vector in the medical knowledge base,/>Representative AND vector/>Associated context information vector,/>Representative AND vector/>Associated context information vector,/>Is a weight coefficient,/>And/>Respectively represent vector/>And/>Is provided for the mold or length of (a),And/>Respectively represent context information vector/>And/>Modulus or length,/>Is a cosine similarity value;
The feature logic relationship analysis submodule receives the key feature matching result, evaluates the spatial arrangement and interaction between features, identifies feature combinations with diagnostic value by using an arrangement combination principle, and obtains a logic relationship model by qualitatively analyzing the logic relationship of the selected feature combinations;
The medical logic relation diagram construction submodule displays the logic relation between feature combinations and the corresponding relation diagram with anal fissure diagnosis standard based on the logic relation model to obtain feature logic analysis results.
Preferably, the model parameter optimization module comprises a network architecture optimization sub-module, a learning efficiency adjustment sub-module and a generalization capability enhancer module;
The network architecture optimization submodule analyzes the complexity of the data in the feature vector set, adjusts the number and the type of layers, and comprises the steps of adding a convolution layer or adjusting an activation function to generate an adjusted network architecture;
the learning efficiency adjustment sub-module refines the learning rate adjustment strategy by using the adjusted network architecture, and optimizes details in the training process by gradually reducing the learning rate so as to obtain the refined learning strategy;
And the generalization capability enhancement submodule applies a refinement learning strategy, adjusts batch size and regularization coefficient when data is input, and obtains optimized model parameters by adjusting data diversity and avoiding fitting of a loss function.
Preferably, the probability evaluation construction module comprises a probability model construction sub-module, a diagnosis result variability analysis sub-module and a diagnosis hypothesis probability calculation sub-module;
The probability model construction submodule analyzes the feature logic analysis result and the optimized data of the model parameters, sets the initial probability of each diagnosis hypothesis, adopts a conditional probability table and a Bayesian network, adjusts the conditional probability according to the interaction between the feature and the hypothesis, and generates a basic probability framework;
the bayesian network is according to formula II:
II (II)
Calculating a conditional probability value after referring to the context information to generate a basic probability frame;
Wherein, Is a conditional probability,/>To be on assumption/>Evidence of true timeProbability of/>To assume/>Prior probability of/>For evidence/>The probability of an unconditional occurrence,For and hypothesis/>Probability of associated context information,/>Is a weight coefficient;
the diagnosis result variability analysis submodule adopts a basic probability framework, identifies key factors causing the variation of the diagnosis result through analyzing probability distribution of each hypothesis in the framework, and carries out probability adjustment on the factors to obtain an optimized probability framework;
the diagnosis hypothesis probability calculation submodule calculates a comprehensive probability value for each diagnosis hypothesis by using the optimized probability framework, adjusts parameters in the framework, reflects current diagnosis information and obtains a probability evaluation result.
Preferably, the comprehensive decision support module comprises a variability factor source identification sub-module, a probability evaluation influence analysis sub-module and a diagnosis suggestion comprehensive sub-module;
The variability factor source identification sub-module checks the probability evaluation result, identifies key variables causing diagnosis change, including data sparsity or feature ambiguity, locates factors affecting diagnosis accuracy, and generates a key variable identification result;
The probability evaluation influence analysis submodule utilizes the key variable identification result to quantify the influence degree of the variable on the diagnosis probability, calculates the influence of a plurality of variables on the result when the variables are changed through simplifying a simulation experiment, and establishes influence weights to obtain an influence weight evaluation result;
And the diagnosis suggestion comprehensive submodule collects information of the feature logic analysis result and the probability evaluation result by depending on the influence weight evaluation result, compares comprehensive support strength of each diagnosis hypothesis, and forms a comprehensive analysis information set.
Preferably, the performance feedback optimization module comprises a user feedback integration sub-module, a system performance evaluation sub-module and an optimization strategy implementation sub-module;
The user feedback integration submodule collects feedback of doctors and patients on system use, analyzes and summarizes information, and generates user demand overview;
The system performance evaluation submodule is used for identifying performance bottlenecks and optimization points in various aspects of the evaluation system based on user demand overview to obtain performance improvement points;
And the optimization strategy implementation submodule formulates optimization measures based on the comprehensive analysis information set and the performance improvement points to form a performance optimization model.
Compared with the prior art, the invention has the beneficial effects that: through the image quality enhancement module and the feature mapping construction module, the accuracy of image preprocessing and feature extraction is improved, and anal fissure identification is more accurate and efficient. In addition, the logic relation analysis module and the probability evaluation construction module not only strengthen the connection between the image features and the medical knowledge base, but also introduce a Bayesian method to quantitatively analyze the uncertainty of the diagnosis result, thereby enhancing the information support strength of the diagnosis decision. The comprehensive decision support module enables the image recognition result, the logic reasoning analysis and the probability evaluation to be comprehensively considered in the diagnosis process, provides comprehensive decision basis, and increases the transparency and the credibility of the diagnosis suggestion.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings that are needed for the description of the embodiments will be briefly described below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and that other drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.
FIG. 1 is a block diagram of an anal fissure detection system based on deep learning according to the present invention;
FIG. 2 is a system frame diagram of the deep learning-based anal fissure detection system according to the present invention;
FIG. 3 is a schematic diagram of an image quality enhancement module in the deep learning-based anal fissure detection system according to the present invention;
FIG. 4 is a schematic diagram of a feature map construction module in the deep learning-based anal fissure detection system according to the present invention;
FIG. 5 is a schematic diagram showing a logic relationship analysis module in the deep learning-based anal fissure detection system according to the present invention;
FIG. 6 is a schematic diagram of a model parameter optimization module in the deep learning-based anal fissure detection system according to the present invention;
FIG. 7 is a schematic diagram of a probability evaluation construction module in the deep learning-based anal fissure detection system according to the present invention;
FIG. 8 is a schematic diagram of a comprehensive decision support module in an anal fissure detection system based on deep learning according to the present invention;
fig. 9 is a schematic diagram of a performance feedback optimization module in the anal fissure detection system based on deep learning according to the present invention.
Detailed Description
The following description of the embodiments of the present invention will be made clearly and completely with reference to the accompanying drawings, in which it is apparent that the embodiments described are only some embodiments of the present invention, but not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
Referring to fig. 1, the present invention provides a technical solution: the anal fissure detection system based on deep learning comprises an image quality enhancement module, a feature mapping construction module, a logic relation analysis module, a model parameter optimization module, a probability evaluation construction module, a comprehensive decision support module and a performance feedback optimization module;
The image quality enhancement module carries out preprocessing on an input anal fissure image, and comprises the steps of removing noise, executing corrosion and expansion operation, highlighting anal fissure characteristics in the image, enhancing image contrast and edge definition through open operation and closed operation, and generating an optimized image set;
the feature map construction module is used for extracting key visual features in the image based on the optimized image set, including edge detection, color intensity analysis and morphological features, constructing feature vectors describing anal fissure and surrounding tissues thereof, and generating a feature vector set;
The logic relationship analysis module utilizes the feature vector set to match the image features with anal fissure features in a predefined medical knowledge base, analyzes the logic relationship among the features, constructs a corresponding logic relationship graph and generates a feature logic analysis result;
the model parameter optimization module takes the feature vector set as input, performs parameter optimization operation, and comprises the steps of adjusting network hierarchical structure, learning rate and regularization parameters, optimizing the ability of the model in identifying and classifying anal fissure images, and generating optimized model parameters;
The probability evaluation construction module performs probability analysis based on the feature logic analysis result and the optimized model parameters, quantifies the uncertainty of the diagnosis result by using a Bayesian method, and generates a probability evaluation result by calculating probability distribution under various diagnosis hypotheses;
the comprehensive decision support module is used for analyzing the influence of the change source of the diagnosis result and the probability evaluation on the diagnosis decision by combining the feature logic analysis result and the probability evaluation result, integrating all analysis contents and recommended diagnosis directions, and generating a comprehensive analysis information set;
The performance feedback optimization module collects feedback of doctors and patients based on the comprehensive analysis information set, and performs iterative optimization of the model and the flow according to the collected feedback to generate a performance optimization model.
The optimized image set comprises an image with improved edge definition, a contrast enhanced image and a noise processed image, the feature vector set comprises a vector of an edge detection result, a numerical vector of color intensity and a description vector of morphological features, the feature logic analysis result comprises a logic mapping of feature matching, a logic map of relationship among features and a result set of logic reasoning, the optimized model parameters comprise a network hierarchy depth set value, a learning rate adjustment value and a regularization parameter optimization value, the probability evaluation result comprises a probability value of each diagnosis hypothesis, an uncertainty quantization index and a confidence evaluation range, the comprehensive analysis information set comprises source analysis of diagnosis uncertainty, probability evaluation comprehensive interpretation and decision basis of recommending a diagnosis direction, and the performance optimization model comprises improvement measures of performance indexes, integration suggestions of user feedback and parameter adjustment of iterative optimization.
Referring to fig. 2 and 3, the image quality enhancement module includes an image denoising processing sub-module, an image feature enhancement sub-module, and an image contrast and sharpness optimization sub-module;
the image denoising processing submodule carries out frequency filtering on an input anal fissure image, eliminates a frequency abnormal part, adjusts brightness and contrast parameters, and generates a denoised image, wherein the specific flow is as follows:
the image denoising processing submodule adopts a Fourier transform algorithm to carry out Fourier transform on an input anal fissure image based on frequency domain filtering, converts the image from a space domain to a frequency domain, eliminates a low-frequency noise part by using a high-pass filter, retains high-frequency detail information, converts the processed frequency domain image back to the space domain by inverse Fourier transform, adjusts brightness and contrast parameters, adjusts brightness distribution of the image by a histogram equalization method, enables the brightness distribution to be more similar to the observation range of human eyes, and generates a denoised image.
The image characteristic enhancer module carries out edge enhancement on the denoised image, adjusts pixel value distribution, highlights anal fissure areas at surrounding tissues, applies local contrast adjustment, improves detail visibility, and obtains a specific flow of the characteristic enhanced image as follows:
The image feature enhancement submodule carries out edge detection on the denoised image based on edge detection by adopting a Canny edge detection algorithm, smoothes the image by a Gaussian filter to reduce noise influence, calculates gradient strength and direction of each pixel point in the image, eliminates non-edge pixels by using a non-maximum suppression technology, determines real and potential edges by adopting a double-threshold algorithm, adjusts pixel value distribution, adjusts pixel values by adopting a local self-adaptive contrast enhancement algorithm to highlight anal fissure areas at surrounding tissues, and generates a feature enhanced image.
The image contrast and definition optimization submodule adjusts the image contrast and the edge definition through open operation and close operation based on the characteristic-enhanced image, optimizes the overall and local brightness balance, and constructs the specific flow of the optimized image set as follows:
The image contrast and definition optimization submodule is based on mathematical morphology operation, adopts morphological open operation and close operation, performs open operation of corrosion after expanding the image with enhanced characteristics so as to smooth noise outside the image, performs close operation of expansion after corrosion so as to fill small black holes inside the image, adjusts the image contrast and edge definition through operation, optimizes the overall and local brightness balance, and constructs an optimized image set.
Referring to fig. 2 and fig. 4, the feature map construction module includes an edge definition analysis sub-module, a color intensity analysis sub-module, and a shape and structure recognition sub-module;
The edge definition analysis submodule carries out difference calculation on pixel intensity around each pixel based on the optimized image set, highlights the edge part of the image, applies nonlinear transformation to adjust the image contrast, strengthens the visibility of key edge lines, refines edge representation by utilizing edge tracking, and generates the specific flow of an edge feature mapping set as follows:
The edge definition analysis submodule carries out gradient calculation on pixel intensity around each pixel based on an optimized image set by adopting a Sobel algorithm, identifies horizontal edges and vertical edges of the image by using filters in horizontal and vertical directions, merges results to highlight edge parts of the image, then adopts nonlinear transformation, specifically a Gamma correction method, increases image contrast by adjusting parameter values, strengthens the visibility of key edge lines, and then utilizes an edge tracking algorithm to track edge connection based on gradient directions, refines edge representation and generates an edge feature mapping set.
The color intensity analysis submodule receives the edge feature mapping set, uniformly samples the image in a color space, analyzes the difference of pixel distribution in each color channel, adjusts the color distribution, enhances the color contrast between the anal fissure and surrounding tissues, and obtains the specific flow of the color difference mapping set as follows:
The color intensity analysis submodule receives the edge feature mapping set, adopts a k-means clustering algorithm to uniformly sample the color space of the image, groups pixels in the image by determining the number of clusters, analyzes the difference of pixel distribution in each color channel, adjusts the color distribution in each cluster by utilizing a color propagation technology, enhances the color comparison of anal fissure and surrounding tissues, and obtains a color difference mapping set.
The shape and structure recognition submodule is based on a color difference mapping set, adjusts parameters to recognize shape features related to anal fissure by analyzing geometric figures in an image, combines an edge feature mapping set, and constructs a composite visual feature expression by a layer-by-layer stacking analysis method, and the specific flow of obtaining a feature vector set is as follows:
the shape and structure recognition submodule is based on a color difference mapping set, adopts a geometric feature extraction method, can uniquely describe the shape by calculating the Hu invariant moment of geometric figures in the image, is not influenced by the size, rotation and reflection of the image, combines an edge feature mapping set, and constructs a composite visual feature expression by a layer-by-layer superposition analysis method, namely, the edge and color features are combined through gradual comprehensive analysis and fusion, so as to obtain a feature vector set.
Referring to fig. 2 and 5, the logic relationship analysis module includes a feature and medical standard comparison sub-module, a feature logic relationship analysis sub-module, and a medical logic relationship diagram construction sub-module;
Each feature in the feature vector set is scanned item by a feature and medical standard comparison submodule, the feature is compared with anal fissure features recorded in a medical knowledge base, the feature similarity is scored by adopting a cosine similarity algorithm, and a recursive feature elimination method is utilized for screening key indexes, so that a specific flow for obtaining a key feature matching result is as follows:
The feature and medical standard comparison submodule carries out cosine similarity algorithm application based on a feature vector set, evaluates similarity by calculating cosine values between feature vectors and anal fissure feature vectors recorded in a medical knowledge base, calculates similarity by means of product of vector point multiplication results and vector length, applies recursive feature elimination method, and uses model accuracy as an index to evaluate importance of features by constructing a model and gradually removing features with lowest contribution degree until a preset feature number is reached, and generates a key feature matching result.
The cosine similarity algorithm is according to formula I:
I is a kind of
Calculating cosine value evaluation similarity between the feature vector and the anal fissure feature vector recorded in the medical knowledge base, and generating a key feature matching result;
Wherein, Is one feature vector in the feature vector set,/>Is the anal fissure feature vector in the medical knowledge base,Representative AND vector/>The associated context information vector may contain additional medical parameters or statistical data associated with the feature. /(I)Representative AND vector/>The associated context information vector also contains additional medical parameters or statistics associated with the anal fissure feature. /(I)And the vector length is used for adjusting the dot product of the original feature vector, the vector length and the contribution rate of the context information in the similarity evaluation. The coefficients are determined based on performance optimization on the training dataset of the model. The dataset comprises medical image data, associated label data, context information associated with the image and feature vectors extracted from the original image, and the model is trained and evaluated by means of a supervised learning method and performance indexes through the steps of feature extraction and vectorization, model training, performance evaluation and optimization, and weight adjustment, so that the optimal value of the weight coefficient is determined, prediction errors are minimized, and the overall performance of the model is optimized. /(I)And/>Respectively represent vector/>And/>Is a function of the calculated vector size. /(I)And/>Respectively represent context information vector/>And/>Is also used to represent the size of the vector. /(I)For cosine similarity values, two vectors/>And(And associated contextual information/>And/>) Similarity between them.
The execution process is as follows:
Selecting a vector from a set of feature vectors And an anal fissure feature vector/>, in a medical knowledge basePreparing corresponding context information vector/>And/>Wherein the context information may also contain additional medical parameters or statistical data associated with the feature.
Determining weight coefficients through a performance optimization process on a training datasetThe dot product of the original feature vector, the vector length and the contribution of the context information in similarity evaluation are balanced, and the optimal model performance is realized.
Calculating vectors using dot product operationsAnd/>Between and context information vector,/>And/>Similarity between them. And calculates the vector/>、/>、/>And/>Is obtained by euclidean norms.
Calculating final similarity score using formula I, according to predetermined weight coefficientCombining the vector point and the length, and calculating to obtain a similarity value.
Evaluating the feature vector according to the calculated cosine similarity valueAnal fissure feature vector/>, in a medical knowledge base(And its associated context information). And if the feature vectors are highly similar, matching, and generating a key feature matching result.
The feature logic relationship analysis submodule receives the key feature matching result, evaluates the space arrangement and interaction between the features, identifies the feature combination with diagnostic value by using the arrangement combination principle, and obtains a logic relationship model by qualitatively analyzing the logic relationship of the selected feature combination, wherein the specific flow is as follows:
The feature logic relation analysis submodule receives key feature matching results, performs an arrangement and combination principle, generates all possible feature combinations according to set feature combination length parameters by utilizing a combinations method in a itertools library of Python, applies a logic regression model to each combination, evaluates the spatial arrangement and interaction between the feature combinations by adjusting weight parameters and intercept of the logic regression, determines feature combinations with diagnostic value according to the output of the model, and obtains a logic relation model.
The medical logic relation diagram construction submodule displays the logic relation between feature combinations and the corresponding relation diagram with anal fissure diagnosis standard based on the logic relation model, and the specific flow for obtaining the feature logic analysis result is as follows:
The medical logic relation diagram construction submodule carries out graphical representation based on a logic relation model, a Graphviz library is adopted, graphic parameters including node patterns, colors, shapes and types of edges are set, nodes and edges in the logic relation model are described by using a dot language, the nodes represent feature combinations, the edges represent logic relations among the feature combinations, and logic relations among the feature combinations and the corresponding relations between the feature combinations and anal fissure diagnosis standards are displayed in a visual mode through adjusting Graphviz drawing engine parameters such as typesetting directions and graphic sizes, so that feature logic analysis results are obtained.
Referring to fig. 2 and 6, the model parameter optimization module includes a network architecture optimization sub-module, a learning efficiency adjustment sub-module, and a generalization capability enhancer module;
The network architecture optimization submodule analyzes the complexity of the feature vector concentrated data, adjusts the number and the type of layers, comprises adding a convolution layer or adjusting an activation function, and generates the specific flow of the adjusted network architecture as follows:
The network architecture optimization submodule adjusts the number and the type of layers based on the complexity of the data in the feature vector set, firstly analyzes the complexity of the data, increases a convolution layer according to the analysis result to improve the capturing capability of the model to details by calculating the dimension and the distribution of the feature vector, replaces the traditional Sigmoid function by adopting a ReLU activation function to solve the gradient vanishing problem, adjusts the number and the size of the filter of the convolution layer to match the complexity of the data, and generates an adjusted network architecture.
The learning efficiency adjustment sub-module refines the learning rate adjustment strategy by using the adjusted network architecture, and optimizes details in the training process by gradually reducing the learning rate, so that the detailed learning strategy is obtained as follows:
The learning efficiency adjustment submodule refines the learning rate adjustment strategy by using an adjusted network architecture, adopts a learning rate attenuation method, sets an initial learning rate and an attenuation coefficient, optimizes details in the training process by gradually reducing the learning rate after each training period, adjusts the learning rate by using an Adam optimizer, ensures rapid convergence in the initial stage of training, refines model parameters by reducing the learning rate in the later stage of training, and obtains the refinement learning strategy.
The generalization capability enhancement submodule applies a refinement learning strategy, adjusts batch size and regularization coefficient when data is input, and avoids fitting by adjusting data diversity and a loss function, so that the specific flow of obtaining optimized model parameters is as follows:
The generalization capability enhancement submodule applies a refined learning strategy, adjusts the batch size and regularization coefficient when data is input, firstly sets the batch size to be small batch to improve the stability of model training and reduce the memory requirement, adopts an L2 regularization method to adjust the regularization coefficient to inhibit model complexity, avoids the problem of over-fitting, and adds a regularization term to the loss function to force model weight to incline to a smaller value, thereby increasing the generalization capability of the model to unseen data and obtaining optimized model parameters.
Referring to fig. 2 and 7, the probability evaluation construction module includes a probability model construction sub-module, a diagnosis result variability analysis sub-module, and a diagnosis hypothesis probability calculation sub-module;
The probability model construction submodule analyzes the feature logic analysis result and the optimized model parameter data, sets the initial probability of each diagnosis hypothesis, adopts a conditional probability table and a Bayesian network, adjusts the conditional probability according to the interaction between the feature and the hypothesis, and generates a specific flow of a basic probability framework as follows:
The probability model construction submodule builds a conditional probability table based on the feature logic analysis result and the optimized model parameter data, creates DATAFRAME by adopting a Pandas library of Python to store an initial probability value of each diagnosis hypothesis, dynamically adjusts the conditional probability according to the interaction between the feature and the hypothesis by using a Bayesian formula function in a SciPy library, allocates a unique identifier for each diagnosis hypothesis, and updates the conditional probability value by an iterative algorithm to reflect the latest diagnosis information to generate a basic probability framework.
Bayesian networks follow formula II:
II (II)
Calculating a conditional probability value after referring to the context information to generate a basic probability frame;
Wherein, Is a conditional probability,/>To be on assumption/>Evidence of true timeProbability of/>To assume/>I.e. before any evidence is observed, assume/>Probability of true. Preset based on past experience or knowledge. /(I)For evidence/>The unconditional probability of occurrence provides a baseline for assessing evidence for prevalence. /(I)For and hypothesis/>The probability of the associated context information, which introduces additional information directly associated with the hypothesis, such as severity of the condition, age and sex of the patient, etc., to improve the accuracy and relevance of the conditional probability calculation. /(I)And the weight coefficient is used for adjusting the contribution proportion of each item in the formula. The feature vectors, anal fissure feature vectors in the medical knowledge base, the context information associated with the vectors, and the prior probabilities of the hypotheses are determined by performance optimization on a training dataset.
The execution process is as follows:
data for calculating conditional probabilities is collected and prepared, including vectors in the feature vector set and anal fissure feature vectors in the medical knowledge base, and contextual information associated with the vectors.
Determining weight coefficients based on optimization of model performance by applying gradient descent method on training dataset。
The Pandas library of Python is used to construct DATAFRAME, store the initial probability values for each diagnostic hypothesis, and adjust the probability values based on the interactions between features and hypotheses and the context information.
The conditional probability values for each diagnostic hypothesis after observation of a given feature are calculated using formula II in combination with the collected data and the determined weight coefficients.
And continuously updating the conditional probability value through an iterative algorithm to reflect the latest diagnosis information. This process ensures that the model adapts and updates over time and data.
And finally, generating a basic probability framework, and providing a basis for subsequent variability analysis of diagnosis results and calculation of diagnosis hypothesis probabilities.
The diagnosis result variability analysis submodule adopts a basic probability framework, identifies key factors causing diagnosis result variation through analyzing probability distribution of each hypothesis in the framework, and carries out probability adjustment on the factors, and the specific flow of obtaining an optimized probability framework is as follows:
the diagnosis result variability analysis submodule adopts a basic probability framework to analyze probability distribution, uses NumPy library to generate random probability values to simulate probability distribution of different diagnosis hypotheses, recognizes deviation and abnormal values in the probability distribution as key factors causing variation of the diagnosis result, uses Matplotlib library to draw the probability distribution so as to intuitively display each hypothesis probability, uses SciPy to carry out hypothesis testing, and finely adjusts the probability values according to the test result to obtain an optimized probability framework.
The diagnosis hypothesis probability calculation submodule calculates a comprehensive probability value for each diagnosis hypothesis by using an optimized probability framework, adjusts parameters in the framework, reflects current diagnosis information and obtains a probability evaluation result, wherein the concrete flow is as follows:
The diagnosis hypothesis probability calculation submodule calculates a comprehensive probability value for each diagnosis hypothesis by using an optimized probability framework, adopts a LogisticRegression model in a Sklearn library, takes the optimized probability framework as input data, sets a solver parameter as 'liblinear' to adapt to the optimization calculation of a small dataset, adjusts the value of regularization strength C to control the complexity of the model, and adjusts parameters in the framework through a model training process so as to enable the parameters to reflect current diagnosis information more accurately and obtain a probability evaluation result.
Referring to fig. 2 and 8, the comprehensive decision support module includes a variability factor source identification sub-module, a probability evaluation influence analysis sub-module, and a diagnosis suggestion comprehensive sub-module;
The variability factor source identification submodule checks the probability evaluation result, identifies key variables causing diagnosis change, including data sparsity or feature ambiguity, locates factors influencing diagnosis accuracy, and generates a specific flow of key variable identification result as follows:
The variability factor source identification submodule carries out identification of key variables based on probability evaluation results, adopts a Pandas library of Python to carry out data exploration, screens out features with high fluctuation or data sparsity through DATAFRAME operation, uses an information gain ratio to quantify the mutual information quantity between the features and a diagnosis result, determines factors such as data sparsity or feature ambiguity which influence diagnosis accuracy, and generates a key variable identification result.
The probability evaluation influence analysis submodule utilizes the key variable identification result to quantify the influence degree of the variable on the diagnosis probability, and calculates the influence size of a plurality of variables on the result by simplifying a simulation experiment, and establishes influence weights, so that the specific flow of obtaining the influence weight evaluation result is as follows:
The probability evaluation influence analysis submodule utilizes key variable identification results to quantify the influence degree of variables, adopts NumPy libraries to carry out numerical simulation, calculates the influence on the probability distribution of the diagnosis result when variables such as data sparsity or feature ambiguity change by generating random data change simulation experiments, quantitatively evaluates the specific influence of each variable change on the diagnosis probability by using a Sklearn regression analysis method, and accordingly establishes the influence weight of each variable to obtain an influence weight evaluation result.
The diagnosis suggestion comprehensive submodule collects information of a feature logic analysis result and a probability evaluation result by depending on an influence weight evaluation result, compares comprehensive support strength of each diagnosis hypothesis, and forms a comprehensive analysis information set as follows:
The diagnosis suggestion comprehensive submodule collects information of feature logic analysis results and probability analysis results according to influence weight assessment results, draws a comparison chart of support strength of each diagnosis hypothesis by using Matplotlib library, compares comprehensive support strength of each diagnosis hypothesis by a graphical method, determines a final diagnosis suggestion direction according to the influence weight assessment results and the feature logic analysis results, and forms a comprehensive analysis information set by applying a decision tree algorithm.
Referring to fig. 2 and 9, the performance feedback optimization module includes a user feedback integration sub-module, a system performance evaluation sub-module, and an optimization strategy implementation sub-module;
The user feedback integration submodule collects feedback from doctors and patients on system use, analyzes and summarizes information, and the specific flow of generating user demand overviews is as follows:
The user feedback integration submodule collects data based on system usage feedback from doctors and patients, collects qualitative and quantitative feedback by adopting an electronic questionnaire tool, pre-processes the collected data by utilizing a Pandas library of Python, including data cleaning and duplicate term removal, then analyzes emotion tendencies of user feedback by using text analysis technology, particularly emotion analysis of Natural Language Processing (NLP), and identifies commonality and specificity of user requirements by using a NLTK library to generate a user requirement overview.
The system performance evaluation submodule is based on user demand overview, the evaluation system is characterized in that performance bottlenecks and optimization points are identified in multiple aspects, and the specific flow for obtaining performance improvement points is as follows:
The system performance evaluation submodule carries out multi-aspect evaluation of system performance based on user demand overview, uses SciPy library of Python to carry out statistical analysis, determines the average performance level of system performance through descriptive statistical analysis, uses analysis of variance (ANOVA) to detect performance differences among different user group feedback, identifies performance bottlenecks such as response time delay or user interface unfriendly, and identifies main factors causing performance deficiency through decision tree algorithm, particularly DecisionTreeClassifier of Sklearn library, so as to obtain performance improvement points.
The optimization strategy implementation submodule formulates optimization measures based on the comprehensive analysis information set and the performance improvement points, and the specific flow for forming the performance optimization model is as follows:
The optimization strategy implementation submodule establishes optimization measures based on comprehensive analysis information sets and performance improvement points, adopts an iteration model in an agile development method, manages optimization tasks through JIRA software, sets priority and iteration period, adjusts model parameters by utilizing GRIDSEARCHCV functions of a Sklearn library, selects optimal parameter combinations to enhance system performance, and continuously tracks optimization effects in the implementation process, and verifies the optimization effects through user satisfaction investigation and system performance test to form a performance optimization model.
In the description of the present specification, the descriptions of the terms "one embodiment," "example," "specific example," and the like, mean that a particular feature, structure, material, or characteristic described in connection with the embodiment or example is included in at least one embodiment or example of the present invention. In this specification, schematic representations of the above terms do not necessarily refer to the same embodiments or examples. Furthermore, the particular features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples.
The preferred embodiments of the invention disclosed above are intended only to assist in the explanation of the invention. The preferred embodiments are not exhaustive or to limit the invention to the precise form disclosed. Obviously, many modifications and variations are possible in light of the above teaching. The embodiments were chosen and described in order to best explain the principles of the invention and the practical application, to thereby enable others skilled in the art to best understand and utilize the invention. The invention is limited only by the claims and the full scope and equivalents thereof.
Claims (6)
1. The anal fissure detection system based on deep learning is characterized by comprising an image quality enhancement module, a feature mapping construction module, a logic relation analysis module, a model parameter optimization module, a probability evaluation construction module, a comprehensive decision support module and a performance feedback optimization module;
the image quality enhancement module carries out pretreatment on an input anal fissure image, and comprises the steps of removing noise, executing corrosion and expansion operation, highlighting anal fissure characteristics in the image, enhancing image contrast and edge definition through open operation and closed operation, and generating an optimized image set;
the feature mapping construction module is used for extracting key visual features in the image based on the optimized image set, including edge detection, color intensity analysis and morphological features, constructing feature vectors describing anal fissure and surrounding tissues thereof, and generating a feature vector set;
the logic relationship analysis module utilizes the feature vector set to match the image features with anal fissure features in a predefined medical knowledge base, analyzes the logic relationship among the features, constructs a corresponding logic relationship graph and generates a feature logic analysis result;
The logic relationship analysis module comprises a feature and medical standard comparison sub-module, a feature logic relationship analysis sub-module and a medical logic relationship diagram construction sub-module;
Each feature in the feature vector set is scanned item by the feature and medical standard comparison submodule, the feature and the feature are compared with anal fissure features recorded in a medical knowledge base, the feature similarity is scored by adopting a cosine similarity algorithm, and a key index is screened by utilizing a recursive feature elimination method to obtain a key feature matching result;
The cosine similarity algorithm is according to formula I:
，
calculating cosine value evaluation similarity between the feature vector and the anal fissure feature vector recorded in the medical knowledge base, and generating a key feature matching result;
Wherein, Is one feature vector in the feature vector set,/>Is anal fissure characteristic vector in the medical knowledge base,/>Representative AND vector/>Associated context information vector,/>Representative AND vector/>The associated context information vector is used to determine,Is a weight coefficient,/>And/>Respectively represent vector/>And/>Modulus or length,/>And/>Respectively represent context information vector/>And/>Modulus or length,/>Is a cosine similarity value;
The feature logic relationship analysis submodule receives the key feature matching result, evaluates the spatial arrangement and interaction between features, identifies feature combinations with diagnostic value by using an arrangement combination principle, and obtains a logic relationship model by qualitatively analyzing the logic relationship of the selected feature combinations;
the medical logic relation diagram construction submodule displays the logic relation between feature combinations and a corresponding relation diagram with anal fissure diagnosis standard based on a logic relation model to obtain a feature logic analysis result;
The model parameter optimization module takes the feature vector set as input, performs parameter optimization operation, and comprises the steps of adjusting network hierarchical structure, learning rate and regularization parameters, optimizing the ability of the model in identifying and classifying anal fissure images, and generating optimized model parameters;
The probability evaluation construction module performs probability analysis based on the feature logic analysis result and the optimized model parameters, quantifies the uncertainty of the diagnosis result by using a Bayesian method, and generates a probability evaluation result by calculating probability distribution under various diagnosis hypotheses;
The probability evaluation construction module comprises a probability model construction sub-module, a diagnosis result variability analysis sub-module and a diagnosis hypothesis probability calculation sub-module;
The probability model construction submodule analyzes the feature logic analysis result and the optimized data of the model parameters, sets the initial probability of each diagnosis hypothesis, adopts a conditional probability table and a Bayesian network, adjusts the conditional probability according to the interaction between the feature and the hypothesis, and generates a basic probability framework;
the bayesian network is according to formula II:
the formula II calculates a conditional probability value after the context information is referred to, and a basic probability frame is generated;
Wherein, Is a conditional probability,/>To be on assumption/>Is true evidence/>Probability of/>To assume/>Prior probability of/>For evidence/>The probability of an unconditional occurrence,For and hypothesis/>Probability of associated context information,/>Is a weight coefficient;
the diagnosis result variability analysis submodule adopts a basic probability framework, identifies key factors causing the variation of the diagnosis result through analyzing probability distribution of each hypothesis in the framework, and carries out probability adjustment on the factors to obtain an optimized probability framework;
the diagnosis hypothesis probability calculation submodule calculates a comprehensive probability value for each diagnosis hypothesis by using an optimized probability framework, adjusts parameters in the framework, reflects current diagnosis information and obtains a probability evaluation result;
The comprehensive decision support module is used for analyzing the influence of the change source of the diagnosis result and the probability evaluation on the diagnosis decision by combining the feature logic analysis result and the probability evaluation result, integrating all analysis contents and recommended diagnosis directions, and generating a comprehensive analysis information set;
the comprehensive decision support module comprises a variability factor source identification sub-module, a probability evaluation influence analysis sub-module and a diagnosis suggestion comprehensive sub-module;
The variability factor source identification sub-module checks the probability evaluation result, identifies key variables causing diagnosis change, including data sparsity or feature ambiguity, locates factors affecting diagnosis accuracy, and generates a key variable identification result;
The probability evaluation influence analysis submodule utilizes the key variable identification result to quantify the influence degree of the variable on the diagnosis probability, calculates the influence of a plurality of variables on the result when the variables are changed through simplifying a simulation experiment, and establishes influence weights to obtain an influence weight evaluation result;
The diagnosis suggestion comprehensive submodule relies on the influence weight evaluation result, gathers the information of the feature logic analysis result and the probability evaluation result, compares the comprehensive support strength of each diagnosis hypothesis, and forms a comprehensive analysis information set;
The performance feedback optimization module collects feedback of doctors and patients based on the comprehensive analysis information set, and performs iterative optimization of the model and the flow according to the collected feedback to generate a performance optimization model.
2. The deep learning based anal fissure detection system according to claim 1 wherein: the optimized image set comprises an image with improved edge definition, a contrast enhanced image and a noise processed image, the feature vector set comprises a vector of an edge detection result, a numerical vector of color intensity and a description vector of morphological features, the feature logic analysis result comprises a logic mapping of feature matching, a logic atlas of a relationship among features and a result set of logic reasoning, the optimized model parameters comprise a network hierarchy depth set value, a learning rate adjustment value and a regularization parameter optimization value, the probability evaluation result comprises a probability value of each diagnosis hypothesis, an uncertainty quantization index and a confidence degree evaluation range, the comprehensive analysis information set comprises source analysis of diagnosis uncertainty, probability evaluation comprehensive interpretation and decision basis of recommending a diagnosis direction, and the performance optimization model comprises improvement measures of performance indexes, integration suggestions of user feedback and parameter adjustment of iterative optimization.
3. The deep learning based anal fissure detection system according to claim 1 wherein: the image quality enhancement module comprises an image denoising processing sub-module, an image characteristic enhancement sub-module and an image contrast and definition optimization sub-module;
the image denoising processing submodule carries out frequency filtering on the input anal fissure image, eliminates frequency abnormal parts, adjusts brightness and contrast parameters and generates a denoised image;
the image characteristic enhancer module performs edge enhancement on the denoised image, adjusts pixel value distribution, highlights an anal fissure region at surrounding tissues, and improves detail visibility by applying local contrast adjustment to obtain a characteristic enhanced image;
The image contrast and definition optimizing submodule adjusts the image contrast and the edge definition through open operation and close operation based on the characteristic-enhanced image, optimizes the overall and local brightness balance and constructs an optimized image set.
4. The deep learning based anal fissure detection system according to claim 1 wherein: the feature mapping construction module comprises an edge definition analysis sub-module, a color intensity analysis sub-module and a shape and structure identification sub-module;
The edge definition analysis submodule carries out difference calculation on pixel intensity around each pixel based on the optimized image set, highlights the edge part of the image, applies nonlinear transformation to adjust the image contrast, strengthens the visibility of key edge lines, refines edge representation by utilizing edge tracking, and generates an edge feature mapping set;
The color intensity analysis submodule receives the edge feature mapping set, uniformly samples the image in a color space, analyzes the difference of pixel distribution in each color channel, adjusts the color distribution, enhances the color contrast between anal fissure and surrounding tissues, and obtains a color difference mapping set;
The shape and structure recognition submodule is used for recognizing shape features associated with anal fissure by adjusting parameters based on a color difference mapping set through analyzing geometric figures in an image, and constructing a composite visual feature expression through a layer-by-layer superposition analysis method by combining the edge feature mapping set to obtain a feature vector set.
5. The deep learning based anal fissure detection system according to claim 1 wherein: the model parameter optimization module comprises a network architecture optimization sub-module, a learning efficiency adjustment sub-module and a generalization capability enhancer module;
The network architecture optimization submodule analyzes the complexity of the data in the feature vector set, adjusts the number and the type of layers, and comprises the steps of adding a convolution layer or adjusting an activation function to generate an adjusted network architecture;
the learning efficiency adjustment sub-module refines the learning rate adjustment strategy by using the adjusted network architecture, and optimizes details in the training process by gradually reducing the learning rate so as to obtain the refined learning strategy;
And the generalization capability enhancement submodule applies a refinement learning strategy, adjusts batch size and regularization coefficient when data is input, and obtains optimized model parameters by adjusting data diversity and avoiding fitting of a loss function.
6. The deep learning based anal fissure detection system according to claim 1 wherein: the performance feedback optimization module comprises a user feedback integration sub-module, a system performance evaluation sub-module and an optimization strategy implementation sub-module;
The user feedback integration submodule collects feedback of doctors and patients on system use, analyzes and summarizes information, and generates user demand overview;
The system performance evaluation submodule is used for identifying performance bottlenecks and optimization points in various aspects of the evaluation system based on user demand overview to obtain performance improvement points;
And the optimization strategy implementation submodule formulates optimization measures based on the comprehensive analysis information set and the performance improvement points to form a performance optimization model.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410402866.9A CN117994255B (en) | 2024-04-03 | 2024-04-03 | Anal fissure detecting system based on deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410402866.9A CN117994255B (en) | 2024-04-03 | 2024-04-03 | Anal fissure detecting system based on deep learning |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN117994255A CN117994255A (en) | 2024-05-07 |
| CN117994255B true CN117994255B (en) | 2024-06-07 |
Family
ID=90896384
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202410402866.9A Active CN117994255B (en) | 2024-04-03 | 2024-04-03 | Anal fissure detecting system based on deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117994255B (en) |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN1839391A (en) * | 2003-06-25 | 2006-09-27 | 美国西门子医疗解决公司 | Systems and methods for automated diagnosis and decision support for breast imaging |
| CN110097093A (en) * | 2019-04-15 | 2019-08-06 | 河海大学 | A kind of heterologous accurate matching of image method |
| KR102403463B1 (en) * | 2021-06-04 | 2022-05-31 | 가천대학교 산학협력단 | Method and System for Recognizing The Personal Health Information using Machine Learning Model |
| CN114782307A (en) * | 2022-02-11 | 2022-07-22 | 安徽医科大学第一附属医院 | Enhanced CT image colorectal cancer staging auxiliary diagnosis system based on deep learning |
| CN116341545A (en) * | 2022-12-26 | 2023-06-27 | 广州启生信息技术有限公司 | Neural network and vector similarity matching method and device based on medical treatment |
| CN117011892A (en) * | 2023-08-30 | 2023-11-07 | 中国计量大学 | Ultrasonic image automatic analysis and diagnosis system based on deep learning |
| CN117474876A (en) * | 2023-11-06 | 2024-01-30 | 郑州轻工业大学 | Deep learning-based kidney cancer subtype auxiliary diagnosis and uncertainty evaluation method |
Family Cites Families (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20190122073A1 (en) * | 2017-10-23 | 2019-04-25 | The Charles Stark Draper Laboratory, Inc. | System and method for quantifying uncertainty in reasoning about 2d and 3d spatial features with a computer machine learning architecture |
| US20220392065A1 (en) * | 2020-01-07 | 2022-12-08 | Cleerly, Inc. | Systems, methods, and devices for medical image analysis, diagnosis, risk stratification, decision making and/or disease tracking |
| US20220230759A1 (en) * | 2020-09-09 | 2022-07-21 | X- Act Science, Inc. | Predictive risk assessment in patient and health modeling |
-
2024
- 2024-04-03 CN CN202410402866.9A patent/CN117994255B/en active Active
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN1839391A (en) * | 2003-06-25 | 2006-09-27 | 美国西门子医疗解决公司 | Systems and methods for automated diagnosis and decision support for breast imaging |
| CN110097093A (en) * | 2019-04-15 | 2019-08-06 | 河海大学 | A kind of heterologous accurate matching of image method |
| KR102403463B1 (en) * | 2021-06-04 | 2022-05-31 | 가천대학교 산학협력단 | Method and System for Recognizing The Personal Health Information using Machine Learning Model |
| CN114782307A (en) * | 2022-02-11 | 2022-07-22 | 安徽医科大学第一附属医院 | Enhanced CT image colorectal cancer staging auxiliary diagnosis system based on deep learning |
| CN116341545A (en) * | 2022-12-26 | 2023-06-27 | 广州启生信息技术有限公司 | Neural network and vector similarity matching method and device based on medical treatment |
| CN117011892A (en) * | 2023-08-30 | 2023-11-07 | 中国计量大学 | Ultrasonic image automatic analysis and diagnosis system based on deep learning |
| CN117474876A (en) * | 2023-11-06 | 2024-01-30 | 郑州轻工业大学 | Deep learning-based kidney cancer subtype auxiliary diagnosis and uncertainty evaluation method |
Non-Patent Citations (2)
| Title |
|---|
| 医学磁共振影像及光学显微成像设备智能诊断关键技术研究;郝如茜;《中国博士学位论文全文数据库 医药卫生科技辑》;20240315(第03期);E060-14 * |
| 基于深度学习的慢性肝病CT报告相似度分析;常炳国;刘清星;;《计算机应用与软件》;20180812(08);全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN117994255A (en) | 2024-05-07 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Kavitha et al. | Deep learning based capsule neural network model for breast cancer diagnosis using mammogram images | |
| Feng et al. | CcNet: A cross-connected convolutional network for segmenting retinal vessels using multi-scale features | |
| Piccialli et al. | A survey on deep learning in medicine: Why, how and when? | |
| EP3770850B1 (en) | Medical image identifying method, model training method, and computer device | |
| CN112102266B (en) | Attention mechanism-based cerebral infarction medical image classification model training method | |
| Uysal et al. | Computer-aided retinal vessel segmentation in retinal images: convolutional neural networks | |
| CN111667490B (en) | Fundus picture cup optic disc segmentation method | |
| WO2021136368A1 (en) | Method and apparatus for automatically detecting pectoralis major region in molybdenum target image | |
| Dayana et al. | An enhanced swarm optimization-based deep neural network for diabetic retinopathy classification in fundus images | |
| Osadebey et al. | Three-stage segmentation of lung region from CT images using deep neural networks | |
| Wang et al. | Segmenting retinal vessels with revised top-bottom-hat transformation and flattening of minimum circumscribed ellipse | |
| Wang et al. | Retinal vessel segmentation approach based on corrected morphological transformation and fractal dimension | |
| Renukadevi et al. | Brain Image Classification Using Time Frequency Extraction with Histogram Intensity Similarity. | |
| Kumar et al. | A novel glaucoma detection model using Unet++-based segmentation and ResNet with GRU-based optimized deep learning | |
| Singh et al. | Deep attention network for pneumonia detection using chest X-ray images | |
| CN113362360B (en) | Ultrasonic carotid plaque segmentation method based on fluid velocity field | |
| CN113011340B (en) | Cardiovascular operation index risk classification method and system based on retina image | |
| Liu et al. | ESDiff: a joint model for low-quality retinal image enhancement and vessel segmentation using a diffusion model | |
| Bhuvaneswari et al. | Contrast enhancement of retinal images using green plan masking and whale optimization algorithm | |
| CN117994255B (en) | Anal fissure detecting system based on deep learning | |
| Al-Mukhtar et al. | Weakly Supervised Sensitive Heatmap framework to classify and localize diabetic retinopathy lesions | |
| Upadhyay et al. | Learning multi-scale deep fusion for retinal blood vessel extraction in fundus images | |
| Thomas et al. | Diabetic retinopathy detection using ensembled transfer learning based thrice CNN with SVM classifier | |
| Mohammad et al. | Predicting Retinal Diseases using Efficient Image Processing and Convolutional Neural Network (CNN) | |
| Shi et al. | Choice over effort: Mapping and diagnosing augmented whole slide image datasets with training dynamics |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant |