CN110766659A - Medical image recognition method, apparatus, device and medium - Google Patents
Medical image recognition method, apparatus, device and medium Download PDFInfo
- Publication number
- CN110766659A CN110766659A CN201910905767.1A CN201910905767A CN110766659A CN 110766659 A CN110766659 A CN 110766659A CN 201910905767 A CN201910905767 A CN 201910905767A CN 110766659 A CN110766659 A CN 110766659A
- Authority
- CN
- China
- Prior art keywords
- medical image
- nodule
- suspected
- unit
- image
- 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.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 90
- 206010028980 Neoplasm Diseases 0.000 claims abstract description 30
- 230000003211 malignant effect Effects 0.000 claims description 24
- 238000004364 calculation method Methods 0.000 claims description 12
- 238000012545 processing Methods 0.000 claims description 12
- 238000000605 extraction Methods 0.000 claims description 11
- 238000003709 image segmentation Methods 0.000 claims description 11
- 238000003325 tomography Methods 0.000 claims description 10
- 230000011218 segmentation Effects 0.000 claims description 9
- 238000012952 Resampling Methods 0.000 claims description 8
- 238000012549 training Methods 0.000 claims description 8
- 238000007781 pre-processing Methods 0.000 claims description 7
- 238000012216 screening Methods 0.000 claims description 6
- 238000004590 computer program Methods 0.000 claims description 4
- 206010054107 Nodule Diseases 0.000 description 23
- 230000008569 process Effects 0.000 description 17
- 238000010586 diagram Methods 0.000 description 12
- 201000011510 cancer Diseases 0.000 description 11
- 241000282414 Homo sapiens Species 0.000 description 8
- 210000004072 lung Anatomy 0.000 description 8
- 210000001519 tissue Anatomy 0.000 description 7
- 230000009471 action Effects 0.000 description 3
- 238000004422 calculation algorithm Methods 0.000 description 3
- 230000006870 function Effects 0.000 description 3
- 206010058467 Lung neoplasm malignant Diseases 0.000 description 2
- 229910000831 Steel Inorganic materials 0.000 description 2
- 210000000988 bone and bone Anatomy 0.000 description 2
- 238000013527 convolutional neural network Methods 0.000 description 2
- 201000005202 lung cancer Diseases 0.000 description 2
- 208000020816 lung neoplasm Diseases 0.000 description 2
- 238000013507 mapping Methods 0.000 description 2
- 238000003062 neural network model Methods 0.000 description 2
- 239000010959 steel Substances 0.000 description 2
- 206010056342 Pulmonary mass Diseases 0.000 description 1
- 238000010521 absorption reaction Methods 0.000 description 1
- 238000007792 addition Methods 0.000 description 1
- 238000003915 air pollution Methods 0.000 description 1
- 238000013459 approach Methods 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000004891 communication Methods 0.000 description 1
- 238000002591 computed tomography Methods 0.000 description 1
- 201000010099 disease Diseases 0.000 description 1
- 208000037265 diseases, disorders, signs and symptoms Diseases 0.000 description 1
- 238000003912 environmental pollution Methods 0.000 description 1
- 239000000835 fiber Substances 0.000 description 1
- 238000001914 filtration Methods 0.000 description 1
- 230000036541 health Effects 0.000 description 1
- 238000003905 indoor air pollution Methods 0.000 description 1
- 230000003902 lesion Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 210000000056 organ Anatomy 0.000 description 1
- 238000002203 pretreatment Methods 0.000 description 1
- 239000004065 semiconductor Substances 0.000 description 1
- 230000000391 smoking effect Effects 0.000 description 1
- 210000001685 thyroid gland Anatomy 0.000 description 1
- 238000012546 transfer Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
- G06T7/0012—Biomedical image inspection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
- G06F18/232—Non-hierarchical techniques
- G06F18/2321—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/11—Region-based segmentation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/136—Segmentation; Edge detection involving thresholding
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10072—Tomographic images
- G06T2207/10081—Computed x-ray tomography [CT]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20076—Probabilistic image processing
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30004—Biomedical image processing
- G06T2207/30061—Lung
- G06T2207/30064—Lung nodule
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30004—Biomedical image processing
- G06T2207/30096—Tumor; Lesion
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Quality & Reliability (AREA)
- Life Sciences & Earth Sciences (AREA)
- Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
- Radiology & Medical Imaging (AREA)
- General Health & Medical Sciences (AREA)
- Probability & Statistics with Applications (AREA)
- Health & Medical Sciences (AREA)
- Medical Informatics (AREA)
- Artificial Intelligence (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a medical image recognition method, a medical image recognition device, medical image recognition equipment and a medical image recognition medium. The method comprises the following steps: obtaining the position information of a suspected nodule unit in an original medical image of a target structure by utilizing a pre-trained nodule identification model; according to the position information of the suspected nodule unit in the original medical image, segmenting the medical image of the suspected nodule unit in the medical image; and recognizing the medical image of the suspected nodule unit by using a pre-trained tumor recognition model to obtain a recognition result score of the suspected nodule unit. According to the medical image identification method, the medical image identification device, the medical image identification equipment and the medical image identification medium, the identification accuracy of the medical image can be guaranteed.
Description
Technical Field
The present invention relates to the field of image processing, and in particular, to a medical image recognition method, apparatus, device, and medium.
Background
Due to the influence of factors such as air pollution, environmental pollution, indoor air pollution and smoking, malignant tumors become one of the diseases with the highest global morbidity and mortality, and seriously threaten the health and the life of human beings. Taking lung cancer as an example, in 2015, 73 thousands of new lung cancer cases in China account for 35.8% of new cases worldwide and are growing at a rate of 80 thousands per year, wherein the death rate is higher at the middle and late stages by 70%.
Some early diseased tissues of cancer are called nodules, which are a descriptive term in the field of medical image processing, and the nodules have the same amount of x-ray absorption as normal tissues of the body. Lung nodules generally have low contrast, high degree of shape dissimilarity, and relatively small size in CT scan images. At this stage, observation is mainly performed by human eyes.
Disclosure of Invention
The medical image identification method, the medical image identification device, the medical image identification equipment and the medical image identification medium can ensure the identification accuracy of the medical image.
According to an aspect of an embodiment of the present invention, there is provided a medical image recognition method including: obtaining the position information of a suspected nodule unit in an original medical image of a target structure by utilizing a pre-trained nodule identification model; according to the position information of the suspected nodule unit in the original medical image, segmenting the medical image of the suspected nodule unit in the medical image; and identifying the medical image of the suspected nodule unit by utilizing a pre-trained tumor identification model to obtain an identification result of the suspected nodule unit, wherein the identification result comprises a benign nodule or a malignant nodule.
According to the medical image identification method in one aspect of the embodiment of the invention, the medical image of the suspected nodule unit can be segmented from the original medical image of the target structure by using the nodule identification model. And inputting the medical image of the suspected nodule unit into a pre-trained tumor recognition model to obtain a recognition result of the suspected nodule unit. Wherein, the identification result of the suspected nodule unit comprises a benign nodule or a malignant nodule. Compared with the subjective identification medical image of human eyes, the method can be used for objectively and quantitatively identifying benign nodules and malignant nodules in the medical image through the two models. Therefore, the recognition accuracy of the medical image is ensured.
In an optional embodiment, the identifying the medical image of the suspected nodule unit by using a pre-trained tumor identification model to obtain an identification result of the suspected nodule unit includes: recognizing the medical image of the suspected nodule unit by utilizing a pre-trained tumor recognition model to obtain a recognition score of the suspected nodule unit; and determining the identification result of the suspected nodule unit according to the identification score of the suspected nodule unit.
By the medical image identification method provided by the embodiment, the suspected nodule unit can be objectively and quantitatively scored, and the identification result of the suspected nodule unit is determined according to the objective and quantitative scoring, so that the identification precision is improved.
In an optional embodiment, after obtaining the identification score of the suspected nodule unit, the method further comprises: and determining the identification scores of one or more suspected nodule units of the target structure to be input into a decision-making model, so as to obtain the identification score of the target structure.
With the medical image recognition method provided by the embodiment, since the suspected nodule units between the target structures are related to each other, for example, if there is a malignant tumor in the lung, the malignant tumor may spread into multiple malignant tumors. The probability of a malignant tumor in the lung is high if the lung identifies multiple units of malignant tumor. Therefore, the identification score of the target structure is calculated by using the identification scores of one or more suspected nodule units of one or more target structures, so that the target structure can be accurately identified.
In an alternative embodiment, before obtaining the position information of the suspected nodule unit in the original medical image of the target structure by using the pre-trained nodule recognition model, the method further comprises: acquiring an original medical image of a target structure; segmenting a substantial image of a target structure from an original medical image; the parenchymal image is input to a nodule recognition model.
Compared to the original medical image, the virtual image does not contain other tissues than the target structure, such as a steel plate, clothes, bones, and the like. Therefore, the parenchymal image is input into the nodule identification model to be processed, the data processing amount is reduced, and the interference of other tissues on the identification result is also reduced.
In an alternative embodiment, segmenting a parenchymal image of the target structure from the medical image comprises: carrying out image segmentation processing on the original medical image by using a pre-trained mask segmentation model to obtain a mask image of a target structure; using the mask image, a substantial image is extracted from the original medical image.
By using the mask image, a substantial image can be accurately extracted from the original medical image. In addition, in the embodiment of the invention, the mask image is obtained by utilizing the mask training model, so that the matching degree of the mask image and the original medical image is extremely high. Therefore, the accuracy of dividing the virtual image can be improved, and the recognition accuracy can be improved.
In an optional embodiment, the method further comprises: marking the medical image samples of the multiple target structures, wherein suspected nodule units in the medical images of the multiple target structures are marked as positive samples, and other units except the suspected nodule units are marked as negative samples; and training a nodule identification model by using the marked medical image sample.
By marking the suspected nodule unit as a positive sample and marking other units except the suspected nodule unit as negative samples, the nodule identification model can be trained according to the difference between the suspected nodule unit and other units, and the accuracy of the nodule identification model is improved.
In an alternative embodiment, the medical image samples of the target structure comprise N sets of medical images, each set of medical images comprising a plurality of consecutive medical images obtained by tomography, wherein at least one of the consecutive medical images comprises a suspected nodule unit.
Considering that tomography is a continuous shooting process, if a certain medical image includes a suspected nodule unit, the adjacent medical image of the medical image also includes the suspected nodule unit to a large extent. Therefore, the accuracy of the nodule identification model can be improved by selecting a plurality of medical image samples of the target structure consisting of N groups of medical images.
In an alternative embodiment, the tumor identification model comprises: a feature extraction sub-model and a classification sub-model; the method comprises the following steps of identifying the medical image of the suspected nodule unit by utilizing a pre-trained tumor identification model to obtain the identification result score of the suspected nodule unit, wherein the identification result score comprises the following steps: inputting the medical image of the suspected nodule unit into a feature extraction sub-model, and extracting the features of the suspected nodule unit; and inputting the characteristics of the suspected nodule units into a classification submodel for classification calculation, and determining the identification result scores of the suspected nodule units according to the classification calculation scores.
In the identification process of the medical image of the suspected nodule unit, the features of the medical image of the suspected nodule unit are extracted firstly, and then the identification score is calculated by using the extracted features. And the identification score is calculated according to the characteristics of the medical image of the suspected nodule unit, so that the calculation accuracy can be ensured while the calculation amount is reduced, and the identification efficiency is improved.
In an alternative embodiment, the method for segmenting a virtual image of a target structure from an original medical image further comprises: preprocessing an original medical image of a target structure; and inputting the preprocessed original medical image into a nodule identification model to segment a virtual image of the target structure.
Before extracting the substantial image, part of irrelevant pixel points can be screened out by preprocessing the original medical image. Therefore, the amount of calculation can be reduced, and the influence of other factors on the recognition result can be eliminated.
In an alternative embodiment, the pre-processing of the original medical image of the target structure comprises: screening each data point based on the Hounsfield unit HU value of each data point in the original medical image; and resampling the screened data points to obtain the preprocessed original medical image.
Through screening out the data point that HU value does not accord with the selection scope, can screen out irrelevant data points such as skin, air in advance, not only reduced image recognition's calculated amount, improved identification efficiency, also can avoid the probability of misidentifying air, skin etc. as suspected knot unit to improve the identification accuracy.
In an alternative embodiment, obtaining the position information of the suspected nodule unit in the original medical image of the target structure by using the pre-trained nodule recognition model includes: obtaining a corresponding probability map by using a pre-trained nodule identification model; and clustering the pixels according to the probability value of each pixel in the probability map to obtain the position information of the suspected nodule unit in the original medical image. The probability value of the pixel point represents the probability value of the pixel point of the image of the suspected nodule unit of the pixel point.
By acquiring the probability image, the probability value of each point belonging to the image pixel point of the suspected nodule unit can be acquired. The higher the probability value is, the higher the probability that the pixel end is an image pixel point of a suspected nodule unit is. By utilizing the probability value and the clustering algorithm, the image similarity of the suspected target nodule unit can be accurately clustered in the image, and the suspected nodule unit is further identified.
In an alternative embodiment, the raw medical image of the target structure comprises raw medical images of the target structure at a plurality of different angles.
Since a single medical image can only represent part of the image information of the target structure. Therefore, the original medical images at different angles can comprehensively reflect the characteristics of the target structure, and the target structure can be conveniently identified in an all-around manner.
According to another aspect of an embodiment of the present invention, there is provided a medical image recognition apparatus including:
the position information acquisition module is used for acquiring the position information of the suspected nodule unit in the original medical image of the target structure by utilizing a pre-trained nodule identification model;
the image segmentation module is used for segmenting the medical image of the suspected nodule unit from the medical image according to the position information of the suspected nodule unit in the original medical image;
and the image identification module is used for identifying the medical image of the suspected nodule unit by utilizing a pre-trained tumor identification model to obtain an identification result of the suspected nodule unit, wherein the identification result comprises a benign nodule or a malignant nodule.
According to the medical image recognition apparatus in another aspect of the embodiment of the present invention, the medical image of the suspected nodule unit can be segmented from the original medical image of the target structure by using the nodule recognition model. And inputting the medical image of the suspected nodule unit into a pre-trained tumor recognition model to obtain a recognition result of the suspected nodule unit. Wherein, the identification result of the suspected nodule unit comprises a benign nodule or a malignant nodule. Compared with the subjective identification medical image of human eyes, the method can be used for objectively and quantitatively identifying the nodules, the tumors and the malignant tumors in the medical image through the two models. Therefore, the recognition accuracy of the medical image is ensured.
According to still another aspect of an embodiment of the present invention, there is provided a medical image recognition apparatus including:
a memory for storing a program;
and a processor for executing the program stored in the memory to execute the medical image recognition method provided by one aspect of the embodiment of the invention.
According to a further aspect of the embodiments of the present invention, there is provided a computer storage medium storing computer program instructions, which when executed by a processor, implement a medical image recognition method provided by an aspect of the embodiments of the present invention.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required to be used in the embodiments of the present invention will be briefly described below, and for those skilled in the art, other drawings can be obtained according to the drawings without creative efforts.
Fig. 1 is a schematic flow chart diagram illustrating a medical image recognition method according to an embodiment of the present invention;
FIG. 2 is a logic diagram illustrating an exemplary captured probability map in accordance with an embodiment of the present invention;
FIG. 3 shows a schematic flow chart of a medical image recognition method provided by another embodiment of the present invention;
FIG. 4 shows a schematic flow chart of a medical image recognition method provided by a further embodiment of the invention;
FIG. 5 is a logic diagram illustrating exemplary obtaining a recognition score for a target structure in accordance with embodiments of the present invention;
FIG. 6 shows a schematic flow chart of a medical image recognition method according to a further embodiment of the invention;
FIG. 7 is a logical diagram illustrating an exemplary extraction of a substantive image, according to an embodiment of the present invention;
fig. 8 shows a schematic structural diagram of a medical image recognition apparatus provided according to an embodiment of the present invention;
fig. 9 is a block diagram of an exemplary hardware architecture of a medical image recognition apparatus in an embodiment of the present invention.
Detailed Description
Features and exemplary embodiments of various aspects of the present invention will be described in detail below, and in order to make objects, technical solutions and advantages of the present invention more apparent, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not to be construed as limiting the invention. It will be apparent to one skilled in the art that the present invention may be practiced without some of these specific details. The following description of the embodiments is merely intended to provide a better understanding of the present invention by illustrating examples of the present invention.
It is noted that, herein, relational terms such as first and second, and the like may be used solely to distinguish one entity or action from another entity or action without necessarily requiring or implying any actual such relationship or order between such entities or actions. Also, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising … …" does not exclude the presence of other identical elements in a process, method, article, or apparatus that comprises the element.
The embodiment of the invention provides a medical image identification method, a medical image identification device, medical image identification equipment and a medical image identification medium, which can be applied to an identification scene of a medical image and used for identifying suspected nodule units from the medical image of a target structure and further calculating identification scores of the suspected nodule units. The score may be used as a basis for identifying the type to which the suspected nodule unit belongs or for determining an identification score for the target structure. The target structure may be a biological organ such as a lung or a thyroid gland, which may have a possibility of being diseased from a nodule to a cancer. Specific types of suspected nodule units may include benign nodules, malignant nodules, and the like.
Since cancer of the human body's partial structure is derived from the lesion of the nodule unit, there is a structural similarity between the benign and malignant nodules, but there is a difference. According to the medical image identification scheme provided by the embodiment of the invention, the medical image of the suspected nodule unit can be identified and segmented from the medical image by utilizing the nodule identification model trained by the commonality between the nodule unit and the tumor unit. And then according to a tumor recognition model trained by the difference of the benign nodules and the malignant nodules, recognizing and scoring the medical image of the suspected nodule units. Compared with the human eye identification medical image, the identification accuracy of the medical image is guaranteed.
For a better understanding of the present invention, a medical image recognition method, apparatus, device and medium according to embodiments of the present invention will be described in detail below with reference to the accompanying drawings, and it should be noted that these embodiments are not intended to limit the scope of the present disclosure.
Fig. 1 is a schematic flow chart illustrating a medical image recognition method according to an embodiment of the present invention. As shown in fig. 1, the medical image recognition method 100 in the present embodiment may include the following steps:
and S110, obtaining the position information of the suspected nodule unit in the original medical image of the target structure by using a pre-trained nodule identification model.
First, the suspected nodule unit refers to a small protrusion with a circular shape existing on the surface or in the internal tissue of the living body, and may specifically include a benign nodule or a malignant nodule. The suspected nodule units have structural commonality. Therefore, the nodule recognition module can be trained according to the commonalities. In particular, the training process of the nodule recognition model may comprise two steps.
In the first step, medical image samples of a plurality of target structures are marked. The medical image sample is obtained by tomography of a target structure containing suspected nodule units. In the sample marking process, suspected nodule units in the medical images of the multiple target structures need to be marked as positive samples, and other units except the suspected nodule units need to be marked as negative samples. By marking the suspected nodule unit as a positive sample and marking other units except the suspected nodule unit as negative samples, the nodule identification model can be trained according to the difference between the suspected nodule unit and other units, and the accuracy of the nodule identification model is improved.
The medical image samples of the target structures in the first step may include N groups of medical images, where each group of medical images corresponds to one suspected nodule unit. Considering that tomography is a continuous shooting process, if a certain medical image includes a suspected nodule unit, the adjacent medical image of the medical image also includes the suspected nodule unit to a large extent. In order to improve the identification accuracy of the nodule identification model, each group of medical images comprises a plurality of continuous medical images obtained by tomography. At least one of the plurality of consecutive medical images contains a suspected nodule unit.
In one embodiment, the set of medical images may include 3 medical images: a certain medical image containing the suspected nodule unit, and a medical image before and after the certain medical image. Illustratively, 100 pictures taken continuously by tomography are numbered starting from 1 in the order of taking. And if the 67 th medical image contains the suspected nodule image, selecting the 66 th medical image, the 67 th medical image and the 68 th medical image as a group of medical image samples.
And secondly, training a nodule identification model by using the marked medical image sample. Wherein the nodule identification model may be a convolutional neural network model. Illustratively, a U-Net neural network model with 20 convolutional layers, which comprises 10 layers of convolution and 10 layers of deconvolution. Wherein, the size of the convolution kernel is 3 × 3, and the loss function can adopt MSE mean square error.
Next, the manner of using the nodule recognition model in S110 may be: the method I is to directly input the original medical image of the target structure into the nodule identification model. Or, in the second mode, the original medical image of the target structure is processed, and the processed medical image is input into the nodule identification model. In order to facilitate the omnidirectional identification of the target structure, the original medical images of the target structure may include original medical images of the target structure at a plurality of different angles. The original medical image of the target structure may be a plurality of different slices or slices of the same slice but at different angles obtained by the tomography of the target structure by the medical scanning device. For example, the information file may be a DICOM medical image format or an MHD format medical image. In an alternative embodiment, the plurality of raw medical images input to the nodule recognition model include information in three different directions XYZ. For example, the three directions XYZ are perpendicular to each other. Finally, in the process of acquiring the position information of the suspected nodule unit in the original medical image of the target structure by using the nodule identification model, the original medical image or the substantial image of the target structure needs to be input into the nodule identification model to obtain a probability map, and then the position information of the suspected nodule unit in the medical image is acquired by using the probability map. Wherein, each pixel point in the probability map corresponds to a probability value. The probability value represents the probability value of the image pixel point of the pixel point belonging to the suspected nodule unit. For accurate acquisition of the position information of the suspected nodule unit in the medical image, S110 may be refined into S111 and S112.
And S111, acquiring a probability map corresponding to the original medical image of the target structure by using the pre-trained nodule identification model.
Wherein, 3 original medical images (which may be scanned images of different layers or scanned images of different angles of the same layer) obtained by tomography scanning the target structure from three different positions can be used to determine the probability map corresponding to the original medical image of the target structure.
Illustratively, FIG. 2 is a logical diagram illustrating an exemplary captured probability map according to an embodiment of the present invention. As shown in fig. 2, first, 3 parenchymal images P31 to P33 of the original medical image are input to the nodule recognition model M1, and a first reference probability image P41, a second reference probability image P42, and a third reference probability image P43 are obtained, respectively. Next, the probability values of the pixels at the same actual position in the 3 reference probability images P41 to P43 are averaged, and the average value is used as the probability value of the actual position. Specifically, if the actual position a on the lung is found, the corresponding pixel a is found in the first reference probability image P411Pixel a1Corresponding probability value b1In the second reference probability image P42, corresponding to the pixel a2Pixel a2Corresponding probability value b2In the third reference probability image P43, the corresponding pixel a3Pixel a3Corresponding probability value b3Then the probability value B corresponding to the actual position AAIs equal to (b)1+b2+b3)/3. Finally, a nodule similarity probability image P5 corresponding to the original medical image of the target structure is generated by using the probability values of the actual positions. In the nodule similarity probability image P5, the higher the similarity probability, the higher the gray level, and the black indicates the lowest probability value.
And S112, clustering the pixel points in the nodule similarity probability image according to the probability value of each actual position point to obtain the position information of the suspected nodule unit in the original medical image of the target structure. The DBSCAN algorithm can be used for clustering all the pixel points, and one pixel cluster obtained by clustering represents a suspected nodule unit.
In addition, because the probability map corresponds to the actual position, structural feature information such as the center coordinate, the radius and the like of the pixel cluster can be obtained after clustering, and actual structural information such as the actual center coordinate, the actual radius and the like of the suspected nodule unit can also be obtained. In some embodiments, information such as the actual center coordinates, the actual radius, and the like of the suspected nodule unit may be output as the recognition result of the structural position feature of the suspected nodule unit.
By acquiring the probability image, the probability value of each point belonging to the image pixel point of the suspected nodule unit can be acquired. The higher the probability value is, the higher the probability that the pixel end is an image pixel point of a suspected nodule unit is. By utilizing the probability value and the clustering algorithm, the image similarity of the suspected target nodule unit can be accurately clustered in the image, and the suspected nodule unit is further identified. And S120, segmenting the medical image of the suspected nodule unit in the medical image according to the position information of the suspected nodule unit in the original medical image.
In order to ensure the identification accuracy, the medical images of the suspected nodule units may be segmented from the medical images of a plurality of angles, respectively. In one embodiment, the medical images at multiple angles can include information about the suspected nodule unit in three XYZ directions.
In addition, during the process of segmenting the medical image of the suspected nodule unit, the size of the image can be 0.7 x 0.7mm3/px3The medical image of the suspected nodule unit is cut out from the original medical image.
And S130, recognizing the medical image of the suspected nodule unit by using a pre-trained tumor recognition model to obtain a recognition result of the suspected nodule unit, wherein the recognition result comprises benign nodules or malignant nodules. Wherein the tumor recognition module may be trained using the difference between benign nodules and malignant nodules.
According to the medical image identification method provided by the embodiment of the invention, the medical image of a suspected nodule unit can be segmented from the original medical image of the target structure by utilizing the nodule identification model. And inputting the medical image of the suspected nodule unit into a pre-trained tumor recognition model to obtain a recognition result of the suspected nodule unit. Wherein, the identification result of the suspected nodule unit comprises a benign nodule and a malignant tumor. Compared with the subjective identification medical image of human eyes, the method can be used for objectively and quantitatively identifying benign nodules and malignant nodules in the medical image through the two models. Therefore, the recognition accuracy of the medical image is ensured.
Fig. 3 shows a schematic flow chart of a medical image recognition method according to another embodiment of the present invention. As shown in fig. 3, the difference from the medical image recognition method of the previous embodiment is that the recognition score of the suspected nodule unit may be calculated first, and then the recognition result of the suspected nodule unit may be determined according to the recognition score of the suspected nodule unit. S130 may be refined into S131 and S132.
And S131, recognizing the medical image of the suspected nodule unit by using a pre-trained tumor recognition model, and acquiring a recognition score of the suspected nodule unit. Wherein the identification score of the suspected nodule unit can be used to distinguish benign nodules from malignant nodules.
In the process of calculating the identification score of the suspected nodule unit, the features of the medical image of the suspected nodule unit may be extracted first, and then the identification score may be calculated by using the extracted features. The tumor identification model may specifically comprise: and the characteristic extraction submodel is used for extracting the characteristics of the suspected nodule units and the classification submodel for calculating the identification scores of the suspected nodule units.
Accordingly, S131 specifically includes S1311 and S1312.
And S1311, inputting the medical image of the suspected nodule unit into a feature extraction sub-model, and extracting features of the suspected nodule unit. For example, the feature extraction submodel may use a neural network VGG 19.
And S1312, inputting the characteristics of the suspected nodule units into the classification submodels for classification calculation, and determining the identification scores of the suspected nodule units according to the classification calculation scores. For example, the classification sub-model may use an extreme gradient Boosting (XGBoost) classifier.
And S132, determining the identification result of the suspected nodule unit according to the identification score of the suspected nodule unit.
In one possible approach, a first score threshold may be preset to determine the identification result of the suspected nodule unit. Specifically, if the identification score of the suspected nodule unit is less than or equal to the first score threshold, the suspected nodule unit is identified as a benign nodule. And if the identification score of the suspected nodule unit is greater than the first score threshold value, identifying the suspected nodule unit as a malignant nodule. For example, the first score threshold may be set to 0.4.
By the medical image identification method provided by the embodiment, the suspected nodule unit can be objectively and quantitatively scored, and the identification result of the suspected nodule unit is determined according to the objective and quantitative scoring, so that the identification precision is improved.
Fig. 4 shows a schematic flow chart of a medical image recognition method according to a further embodiment of the present invention. As shown in fig. 4, the difference from the medical image recognition method of the previous embodiment is that after S131, the method 600 may further include:
and S133, inputting the identification scores of one or more suspected nodule units of the target structure into a decision model to obtain the identification scores of the target structure. The recognition score of the target structure represents the probability of the target structure having malignant nodules, and the higher the recognition score of the target structure is, the higher the probability of the target structure having malignant nodules is. The lower the score of the target structure, the higher the probability that a benign nodule exists in the target structure.
Since there may be multiple suspected nodule units in a human body. Therefore, in one possible method, if M suspected nodule units are identified from the target structure, K suspected nodule units may be selected in order of the identification scores of the suspected nodule units from high to low to obtain the identification score of the target structure. Wherein M is an integer greater than K. For example, K is 5. The identification score of the target structure and the identification score of the suspected nodule unit may be probability scores.
In one embodiment, if the recognition score of the target structure can be a probability score, the decision model can be a Noisy-or decision model. Wherein the decision model adopts a Noisy-or decision method shown in formula (1).
P＝1＝Π(1-pi) (1)
Wherein P is the recognition score of the target structure, PiIs the identification score of the ith suspected nodule unit.
As an example, FIG. 5 is a flow chart illustrating exemplary acquisition of target structure identification, according to an embodiment of the present inventionLogic diagram of the scores. As shown in fig. 5, 5 medical images P61-P65 of suspected nodule units are input into the tumor identification unit M2 to obtain identification scores P of the corresponding suspected nodule units1-P6. The identification scores P of 5 suspected nodule units1-P6After the decision model M3 is input, the recognition score P of the target structure is obtained.
With the medical image recognition method provided by the embodiment, since the suspected nodule units between the target structures are related to each other, for example, if there is a malignant nodule in the lung, the malignant nodule may spread into a plurality of malignant nodules. The probability of a malignant nodule in the lung is high if the lung identifies multiple units of the malignant nodule. Therefore, the identification score of the target structure is calculated by using the identification results of one or more suspected nodule units of one or more target structures, so that the target structure can be accurately identified.
Fig. 6 shows a schematic flow chart of a medical image recognition method according to another embodiment of the present invention. As shown in fig. 6, the difference from the medical image recognition method of the previous embodiment is that the nodule recognition model is used in S110, which may be processing an original medical image of a target structure and inputting the processed medical image into the nodule recognition model.
Specifically, prior to S110, the method 100 may further include S140, S150, and S160 described below.
S140, acquiring an original medical image of the target structure.
S150, segmenting a substantial image of the target structure from the original medical image of the target structure. The virtual image representation of the target structure contains only images of the target structure.
Before the substantial image is segmented, in order to reduce the calculation amount and eliminate the influence of other factors on the identification result. The original medical image of the target structure can be preprocessed, and then the substantial image of the target structure is segmented from the preprocessed original medical image. The preprocessing process comprises the steps of data filtering, resampling and the like. Specifically, the pre-treatment process may include the following two steps of screening and resampling.
In a first step, each data point in the original medical image is screened based on its Hounsfield Unit (HU) value. In one optional example, the selection range of the HU value may be set to [ -1000,400 ]. That is, data points having HU values less than-1000 or greater than 400 are filtered out. Through screening out the data point that HU value does not accord with the selection scope, can screen out irrelevant data points such as skin, air in advance, not only reduced image recognition's calculated amount, improved identification efficiency, also can avoid the probability of misidentifying air, skin etc. as suspected knot unit to improve the identification accuracy.
And secondly, resampling the screened data points to obtain a preprocessed original medical image. Illustratively, may be in terms of 1 × 1mm3/px3The data interval of (2) is resampled, that is, after resampling, the distance between adjacent data in rows and columns is 1 mm. By resampling, the data is provided with the same data interval.
In segmenting the real-time image, a pre-selected trained mask segmentation model may be utilized to segment a substantial image of the target structure. In particular segmenting the real-time image can be refined into the following two steps.
Firstly, carrying out image segmentation processing on an original medical image by using a pre-trained mask segmentation model to obtain a mask image of a target structure. The mask image is used to extract a region of interest, i.e., a region where the target structure is located, from the original medical image. Illustratively, the mask image may be a binary mask consisting of 0 and 1. That is, the value of the pixel point in the region of the target structure is 1, and the value of the pixel point outside the region is 0.
Considering that a plurality of different original medical images may be input, the mask segmentation module may be used to obtain mask images corresponding to the plurality of original medical images one to one. For example, when the first medical image, the second medical image and the third medical image are input into the mask segmentation model, a first mask image corresponding to the first medical image, a second mask image corresponding to the second medical image and a third mask image corresponding to the third medical image are obtained.
The mask segmentation model may be trained using a plurality of medical image samples of the target structure. Optionally, the number of medical image samples is 500. In an alternative embodiment, the mask segmentation model may be a convolutional neural network model. Illustratively, a U-Net neural network model with 10 convolutional layers, which comprises 5 layers of convolution and 5 layers of deconvolution. Wherein, the size of the convolution kernel is 3 × 3, and the loss function is Jaccard similarity coefficient. The Jaccard similarity coefficient S satisfies formula (2):
S＝|A∩B|/|A∪B|＝|A∩B|/(|A|+|B|-|A∩B|) (2)
wherein A is a predicted value, and B is a tag value of the tag. The larger the S, the higher the similarity between the predicted value and the label.
It should be noted that, with the network model in the present embodiment, the size of the input medical image is the same as that of the output mask image. For example, the size of the input medical image may be 512 x 512, and the size of the output mask image may also be 512 x 512.
And secondly, extracting a substantial image from the original medical image by using the mask image. The method for extracting the substantial image comprises the following steps:
the method is to directly use the mask image to extract the essential image from the original medical image.
And secondly, determining a minimum rectangular frame containing the mask region according to the mask image, determining a mapping rectangular frame corresponding to the rectangular frame in the original medical image according to the position of the minimum rectangular frame in the mask image, and cutting out an image in the mapping rectangular frame from the original medical image to be used as a substantial image. The position of the minimum rectangular frame in the mask image may be determined by the positions of the corners of the rectangular frame, for example, the position coordinates of the upper left corner and the lower right corner of the rectangular frame.
With respect to mode two, fig. 7 is a logic diagram illustrating an exemplary extraction of a substantial image according to an embodiment of the present invention. As shown in fig. 7, the original medical image P1 corresponds to the mask image P2. The white rectangular box in the mask image is the smallest rectangular box that contains the mask area. The substantial image P3 of the target structure can be cut out of the original medical image with a white rectangular box.
By using the mask image, a substantial image can be accurately extracted from the original medical image. In addition, in the embodiment of the invention, the mask image is obtained by utilizing the mask training model, so that the matching degree of the mask image and the original medical image is extremely high. Therefore, the recognition accuracy can be improved.
S160, the parenchymal image of the target structure is input into the nodule identification model. Optionally, a plurality of parenchymal images of the target structure at different angles are input to the nodule identification model.
Compared to the original medical image, the virtual image does not contain other tissues than the target structure, such as a steel plate, clothes, bones, and the like. Therefore, the parenchymal image is input into the nodule identification model to be processed, the data processing amount is reduced, and the interference of other tissues on the identification result is also reduced.
An apparatus according to an embodiment of the present invention will be described in detail below with reference to the accompanying drawings.
Based on the same inventive concept, the embodiment of the invention provides a medical image recognition device. Fig. 8 shows a schematic structural diagram of a medical image recognition apparatus provided according to an embodiment of the present invention. As shown in fig. 8, the medical image recognition apparatus 800 includes:
a position information obtaining module 810, configured to obtain position information of a suspected nodule unit in an original medical image of a target structure by using a pre-trained nodule recognition model;
an image segmentation module 820, configured to segment the medical image of the suspected nodule unit from the medical image according to the position information of the suspected nodule unit in the original medical image;
the image recognition module 830 is configured to recognize the medical image of the suspected nodule unit by using a pre-trained tumor recognition model, so as to obtain a recognition result of the suspected nodule unit. Wherein, the identification result of the suspected nodule comprises: benign nodules and malignant nodules.
In some embodiments of the present invention, the image recognition module 830 includes:
the first score determining unit is used for identifying the medical image of the suspected nodule unit by utilizing a pre-trained tumor identification model to obtain an identification score of the suspected nodule unit;
and the identification result determining unit is used for determining the identification result of the suspected nodule unit according to the identification score of the suspected nodule unit.
In some embodiments of the present invention, the image recognition module 830 further comprises:
and the second score calculating unit is used for determining and inputting the identification scores of one or more suspected nodule units of the target structure into the decision model to obtain the identification score of the target structure.
In some embodiments of the present invention, the medical image recognition apparatus 800 further comprises:
the image acquisition module is used for acquiring an original medical image of a target structure;
the real image segmentation module is used for segmenting a real image of a target structure from the original medical image;
and the image input module is used for inputting the parenchymal image into the nodule identification model.
In some embodiments of the present invention, the substantive image segmentation module is specifically configured to:
carrying out image segmentation processing on the original medical image by using a pre-trained mask segmentation model to obtain a mask image of a target structure;
using the mask image, a substantial image is extracted from the original medical image.
In some embodiments of the present invention, the medical image recognition apparatus 800 further comprises:
the system comprises a sample marking module, a data processing module and a data processing module, wherein the sample marking module is used for marking medical image samples of a plurality of target structures, suspected nodule units in the medical images of the plurality of target structures are marked as positive samples, and other units except the suspected nodule units are marked as negative samples;
and the model training module is used for training the nodule identification model by using the marked medical image samples.
In some embodiments of the present invention, the medical image samples of the target structure include N sets of medical images, each set of medical images including a plurality of consecutive medical images obtained by tomography, wherein at least one of the consecutive medical images contains a suspected nodule unit.
In some embodiments of the invention, the tumor identification model comprises: a feature extraction sub-model and a classification sub-model;
the first score determining unit is specifically configured to: inputting the medical image of the suspected nodule unit into a feature extraction sub-model, and extracting the features of the suspected nodule unit; and inputting the characteristics of the suspected nodule units into a classification submodel for classification calculation, and determining the identification scores of the suspected nodule units according to the classification calculation scores.
In some embodiments of the invention, the substantial image segmentation module is specifically configured to:
preprocessing an original medical image of a target structure;
and segmenting a substantial image of the target structure from the preprocessed original medical image.
In some embodiments of the invention, the substantial image segmentation module is specifically configured to:
screening each data point based on the Hounsfield unit HU value of each data point in the original medical image;
and resampling the screened data points to obtain the preprocessed original medical image.
In some embodiments of the present invention, the location information obtaining module 810 is specifically configured to:
obtaining a corresponding probability map by using a pre-trained nodule identification model;
and clustering the pixels according to the probability value of each pixel in the probability map to obtain the position information of the suspected nodule unit in the original medical image. The probability value of the pixel point represents the probability value of the pixel point of the image of the suspected nodule unit of the pixel point.
In some embodiments of the present invention, the original medical image of the target structure comprises an original medical image of the target structure at a plurality of different angles.
Other details of the medical image recognition apparatus according to the embodiment of the present invention are similar to those of the method according to the embodiment of the present invention described above with reference to fig. 1 to 7, and are not repeated herein.
Fig. 9 is a block diagram of an exemplary hardware architecture of a medical image recognition apparatus in an embodiment of the present invention.
As shown in fig. 9, the medical image recognition apparatus 900 includes an input apparatus 901, an input interface 902, a central processor 903, a memory 904, an output interface 905, and an output apparatus 906. The input interface 902, the central processing unit 903, the memory 904, and the output interface 905 are connected to each other through a bus 910, and the input device 901 and the output device 906 are connected to the bus 910 through the input interface 902 and the output interface 905, respectively, and further connected to other components of the medical image recognition device 900.
Specifically, the input device 901 receives input information from the outside, and transmits the input information to the central processor 903 through the input interface 902; central processor 903 processes input information based on computer-executable instructions stored in memory 904 to generate output information, stores the output information temporarily or permanently in memory 904, and then transmits the output information to output device 906 via output interface 905; the output device 906 outputs the output information to the outside of the medical image recognition device 900 for use by the user.
That is, the medical image recognition apparatus shown in fig. 9 may also be implemented to include: a memory storing computer-executable instructions, and a processor. The processor, when executing the computer executable instructions, may implement the method and apparatus of the medical image recognition device described in connection with fig. 1-2.
In one embodiment, the medical image recognition device 900 shown in fig. 9 may be implemented as a device that may include: a memory for storing a program; a processor for executing the program stored in the memory to execute the medical image recognition method of the embodiment of the present invention.
Embodiments of the present invention further provide a computer storage medium, where computer program instructions are stored on the computer storage medium, and when the computer program instructions are executed by a processor, the method for recognizing medical images according to the embodiments of the present invention is implemented in combination with the medical image recognition methods shown in fig. 1 to 7.
It is to be understood that the invention is not limited to the specific arrangements and instrumentality described above and shown in the drawings. A detailed description of known methods is omitted herein for the sake of brevity. In the above embodiments, several specific steps are described and shown as examples. However, the method processes of the present invention are not limited to the specific steps described and illustrated, and those skilled in the art can make various changes, modifications and additions or change the order between the steps after comprehending the spirit of the present invention.
The functional blocks shown in the above structural block diagrams may be implemented as hardware, software, firmware, or a combination thereof. When implemented in hardware, it may be, for example, an electronic circuit, an Application Specific Integrated Circuit (ASIC), suitable firmware, plug-in, function card, or the like. When implemented in software, the elements of the invention are the programs or code segments used to perform the required tasks. The program or code segments may be stored in a machine-readable medium or transmitted by a data signal carried in a carrier wave over a transmission medium or a communication link. A "machine-readable medium" may include any medium that can store or transfer information. Examples of a machine-readable medium include electronic circuits, semiconductor memory devices, ROM, flash memory, Erasable ROM (EROM), floppy disks, CD-ROMs, optical disks, hard disks, fiber optic media, Radio Frequency (RF) links, and so forth. The code segments may be downloaded via computer networks such as the internet, intranet, etc.
As will be apparent to those skilled in the art, for convenience and brevity of description, the specific working processes of the systems, modules and units described above may refer to the corresponding processes in the foregoing method embodiments, and are not described herein again.
Claims (15)
1. A medical image recognition method, characterized in that the method comprises:
obtaining the position information of a suspected nodule unit in an original medical image of a target structure by utilizing a pre-trained nodule identification model;
according to the position information of the suspected nodule unit in the original medical image, segmenting the medical image of the suspected nodule unit in the medical image;
recognizing the medical image of the suspected nodule unit by utilizing a pre-trained tumor recognition model to obtain a recognition result of the suspected nodule unit,
wherein the recognition result comprises a benign nodule or a malignant nodule.
2. The method according to claim 1, wherein the identifying the medical image of the suspected nodule unit by using a pre-trained tumor identification model to obtain the identification result of the suspected nodule unit comprises:
recognizing the medical image of the suspected nodule unit by utilizing a pre-trained tumor recognition model to obtain a recognition score of the suspected nodule unit;
and determining the identification result of the suspected nodule unit according to the identification score of the suspected nodule unit.
3. The method of claim 2, wherein after obtaining the identification score of the suspected nodule unit, the method further comprises:
and determining the identification scores of one or more suspected nodule units of the target structure to be input into a decision-making model, so as to obtain the identification score of the target structure.
4. The method of claim 1, wherein before obtaining the position information of the suspected nodule unit in the original medical image of the target structure by using the pre-trained nodule recognition model, the method further comprises:
acquiring an original medical image of the target structure;
segmenting a substantial image of the target structure from the original medical image;
the parenchymal image is input to a nodule recognition model.
5. The method of claim 4, wherein segmenting the substantial image of the target structure from the medical image comprises:
carrying out image segmentation processing on the original medical image by using a pre-trained mask segmentation model to obtain a mask image of the target structure;
extracting the substantial image from the original medical image using the mask image.
6. The method of claim 1, further comprising:
marking a plurality of medical image samples of the target structure, wherein suspected nodule units in the medical images of the target structure are marked as positive samples, and other units except the suspected nodule units are marked as negative samples;
and training the nodule identification model by using the marked medical image sample.
7. The method of claim 6, wherein the medical image samples of the target structures comprise N sets of medical images, each set of medical images comprising a plurality of consecutive medical images obtained by tomography, wherein at least one of the consecutive medical images contains the suspected nodule unit.
8. The method of claim 2, wherein the tumor identification model comprises: a feature extraction sub-model and a classification sub-model;
the identifying the medical image of the suspected nodule unit by using the pre-trained tumor identification model to obtain the identification score of the suspected nodule unit includes:
inputting the medical image of the suspected nodule unit into the feature extraction sub-model, and extracting the features of the suspected nodule unit;
inputting the characteristics of the suspected nodule units into the classification submodels for classification calculation, and determining the identification scores of the suspected nodule units according to the classification calculation scores.
9. The method of claim 3, wherein segmenting the substantial image of the target structure from the original medical image comprises:
preprocessing an original medical image of the target structure;
and segmenting a substantial image of the target structure from the preprocessed original medical image.
10. The method of claim 9, wherein the pre-processing of the raw medical image of the target structure comprises:
screening each data point in the original medical image based on the Hounsfield Unit (HU) value of each data point;
and resampling the screened data points to obtain the preprocessed original medical image.
11. The method of claim 1, wherein obtaining the position information of the suspected nodule unit in the original medical image of the target structure by using the pre-trained nodule recognition model comprises:
obtaining a corresponding probability map by using a pre-trained nodule identification model;
clustering the pixels according to the probability value of each pixel in the probability map to obtain the position information of the suspected nodule unit in the original medical image,
and the probability value of the pixel point represents the probability value of the image pixel point of the pixel point belonging to the suspected nodule unit.
12. The method of any one of claims 1 to 11, wherein the original medical image of the target structure comprises an original medical image of the target structure at a plurality of different angles.
13. A medical image recognition apparatus, characterized in that the apparatus comprises:
the position information acquisition module is used for acquiring the position information of the suspected nodule unit in the original medical image of the target structure by utilizing a pre-trained nodule identification model;
the image segmentation module is used for segmenting the medical image of the suspected nodule unit from the medical image according to the position information of the suspected nodule unit in the original medical image;
and the image identification module is used for identifying the medical image of the suspected nodule unit by utilizing a pre-trained tumor identification model to obtain an identification result of the suspected nodule unit, wherein the identification result comprises a benign nodule or a malignant nodule.
14. A medical image recognition apparatus, characterized in that the apparatus comprises:
a memory for storing a program;
a processor for executing the program stored in the memory to perform the medical image recognition method of any one of claims 1-12.
15. A computer storage medium having computer program instructions stored thereon which, when executed by a processor, implement the medical image recognition method of any one of claims 1-12.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910905767.1A CN110766659A (en) | 2019-09-24 | 2019-09-24 | Medical image recognition method, apparatus, device and medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910905767.1A CN110766659A (en) | 2019-09-24 | 2019-09-24 | Medical image recognition method, apparatus, device and medium |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110766659A true CN110766659A (en) | 2020-02-07 |
Family
ID=69329893
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910905767.1A Pending CN110766659A (en) | 2019-09-24 | 2019-09-24 | Medical image recognition method, apparatus, device and medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110766659A (en) |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111369524A (en) * | 2020-03-02 | 2020-07-03 | 北京推想科技有限公司 | Nodule recognition model training method, nodule recognition method and device |
| CN111626301A (en) * | 2020-05-07 | 2020-09-04 | 北京海益同展信息科技有限公司 | Image screening method and device, electronic equipment and storage medium |
| CN112541907A (en) * | 2020-12-17 | 2021-03-23 | 推想医疗科技股份有限公司 | Image identification method, device, server and medium |
| CN112950577A (en) * | 2021-02-26 | 2021-06-11 | 南方科技大学 | Image processing method, image processing device, electronic equipment and storage medium |
| WO2021169161A1 (en) * | 2020-02-26 | 2021-09-02 | 上海商汤智能科技有限公司 | Image recognition method, recognition model training method and apparatuses related thereto, and device |
| CN113903452A (en) * | 2021-09-18 | 2022-01-07 | 宁波大学医学院附属医院 | Auxiliary diagnosis system for pulmonary nodules of doctors |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106780460A (en) * | 2016-12-13 | 2017-05-31 | 杭州健培科技有限公司 | A kind of Lung neoplasm automatic checkout system for chest CT image |
| CN108010021A (en) * | 2017-11-30 | 2018-05-08 | 上海联影医疗科技有限公司 | A kind of magic magiscan and method |
| CN108010013A (en) * | 2017-11-03 | 2018-05-08 | 天津大学 | A kind of lung CT image pulmonary nodule detection methods |
| CN109003260A (en) * | 2018-06-28 | 2018-12-14 | 深圳视见医疗科技有限公司 | CT image pulmonary nodule detection method, device, equipment and readable storage medium storing program for executing |
| CN109559303A (en) * | 2018-11-22 | 2019-04-02 | 广州达美智能科技有限公司 | Recognition methods, device and the computer readable storage medium of calcification point |
| CN109886933A (en) * | 2019-01-25 | 2019-06-14 | 腾讯科技（深圳）有限公司 | A kind of medical image recognition method, apparatus and storage medium |
| CN110148192A (en) * | 2019-04-18 | 2019-08-20 | 上海联影智能医疗科技有限公司 | Medical image imaging method, device, computer equipment and storage medium |
-
2019
- 2019-09-24 CN CN201910905767.1A patent/CN110766659A/en active Pending
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106780460A (en) * | 2016-12-13 | 2017-05-31 | 杭州健培科技有限公司 | A kind of Lung neoplasm automatic checkout system for chest CT image |
| CN108010013A (en) * | 2017-11-03 | 2018-05-08 | 天津大学 | A kind of lung CT image pulmonary nodule detection methods |
| CN108010021A (en) * | 2017-11-30 | 2018-05-08 | 上海联影医疗科技有限公司 | A kind of magic magiscan and method |
| CN109003260A (en) * | 2018-06-28 | 2018-12-14 | 深圳视见医疗科技有限公司 | CT image pulmonary nodule detection method, device, equipment and readable storage medium storing program for executing |
| CN109559303A (en) * | 2018-11-22 | 2019-04-02 | 广州达美智能科技有限公司 | Recognition methods, device and the computer readable storage medium of calcification point |
| CN109886933A (en) * | 2019-01-25 | 2019-06-14 | 腾讯科技（深圳）有限公司 | A kind of medical image recognition method, apparatus and storage medium |
| CN110148192A (en) * | 2019-04-18 | 2019-08-20 | 上海联影智能医疗科技有限公司 | Medical image imaging method, device, computer equipment and storage medium |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2021169161A1 (en) * | 2020-02-26 | 2021-09-02 | 上海商汤智能科技有限公司 | Image recognition method, recognition model training method and apparatuses related thereto, and device |
| CN111369524A (en) * | 2020-03-02 | 2020-07-03 | 北京推想科技有限公司 | Nodule recognition model training method, nodule recognition method and device |
| CN111626301A (en) * | 2020-05-07 | 2020-09-04 | 北京海益同展信息科技有限公司 | Image screening method and device, electronic equipment and storage medium |
| CN111626301B (en) * | 2020-05-07 | 2023-09-26 | 京东科技信息技术有限公司 | Image screening method and device, electronic equipment and storage medium |
| CN112541907A (en) * | 2020-12-17 | 2021-03-23 | 推想医疗科技股份有限公司 | Image identification method, device, server and medium |
| CN112950577A (en) * | 2021-02-26 | 2021-06-11 | 南方科技大学 | Image processing method, image processing device, electronic equipment and storage medium |
| CN112950577B (en) * | 2021-02-26 | 2024-01-16 | 南方科技大学 | Image processing method, device, electronic equipment and storage medium |
| CN113903452A (en) * | 2021-09-18 | 2022-01-07 | 宁波大学医学院附属医院 | Auxiliary diagnosis system for pulmonary nodules of doctors |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110766659A (en) | Medical image recognition method, apparatus, device and medium | |
| CN111985536B (en) | Based on weak supervised learning gastroscopic pathology image Classification method | |
| CN109635846B (en) | Multi-type medical image judging method and system | |
| Kowal et al. | Computer-aided diagnosis of breast cancer based on fine needle biopsy microscopic images | |
| CN110599508A (en) | Spine image processing method based on artificial intelligence and related equipment | |
| CN114092450A (en) | Real-time image segmentation method, system and device based on gastroscopy video | |
| CN112819747A (en) | Method for automatically diagnosing benign and malignant nodules based on lung tomography image | |
| CN112820399A (en) | Method and device for automatically diagnosing benign and malignant thyroid nodules | |
| CN114972272A (en) | Grad-CAM-based segmentation method for new coronary pneumonia lesions | |
| CN112712540B (en) | Lung bronchus extraction method based on CT image | |
| JP6132485B2 (en) | Image processing apparatus, operation method of image processing apparatus, and image processing program | |
| CN113780421B (en) | Brain PET image identification method based on artificial intelligence | |
| Dickson et al. | A Dual Channel Multiscale Convolution U-Net Methodfor Liver Tumor Segmentation from Abdomen CT Images | |
| CN105956587A (en) | Method for automatically extracting meniscus from knee-joint magnetic resonance image sequence based on shape constraint | |
| Khan et al. | Segmentation of single and overlapping leaves by extracting appropriate contours | |
| CN109948706B (en) | Micro-calcification cluster detection method combining deep learning and feature multi-scale fusion | |
| CN114913185A (en) | Texture segmentation method and system for lung CT image | |
| CN114155234A (en) | Method and device for identifying position of lung segment of focus, storage medium and electronic equipment | |
| CN113222009A (en) | Method for estimating nodule aspect ratio based on thyroid ultrasound video | |
| Paul et al. | Computer-Aided Diagnosis Using Hybrid Technique for Fastened and Accurate Analysis of Tuberculosis Detection with Adaboost and Learning Vector Quantization | |
| CN117058292B (en) | Tone scale map rendering system based on digital pathological image | |
| CN117315378B (en) | Grading judgment method for pneumoconiosis and related equipment | |
| CN113627231B (en) | Automatic segmentation method for liquid region in retina OCT image based on machine vision | |
| CN111369580B (en) | Intelligent segmentation method and terminal for X-ray image | |
| CN111986137B (en) | Biological organ lesion detection method, apparatus, device, and readable storage medium |
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 | ||
| TA01 | Transfer of patent application right | ||
| TA01 | Transfer of patent application right |
Effective date of registration: 20210719 Address after: Room 103-2, building a, phase II, science and technology building, 1057 Nanhai Avenue, Yanshan community, zhaoshang street, Nanshan District, Shenzhen, Guangdong 518052 Applicant after: Xi Ren Ma Da Zhou (Shenzhen) Medical Technology Co.,Ltd. Address before: 202, east side, 2 / F, building F, Zhizao street, Zhongguancun, 45 Chengfu Road, Haidian District, Beijing Applicant before: Xiren Ma Diyan (Beijing) Technology Co.,Ltd. |
|
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20200207 |