CN112633342A - Human body ultrasonic detection real-time guiding strategy based on deep learning - Google Patents
Human body ultrasonic detection real-time guiding strategy based on deep learning Download PDFInfo
- Publication number
- CN112633342A CN112633342A CN202011486938.0A CN202011486938A CN112633342A CN 112633342 A CN112633342 A CN 112633342A CN 202011486938 A CN202011486938 A CN 202011486938A CN 112633342 A CN112633342 A CN 112633342A
- Authority
- CN
- China
- Prior art keywords
- human body
- real
- neural network
- ultrasonic
- strategy
- 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.)
- Granted
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/25—Fusion techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Computational Linguistics (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Ultra Sonic Daignosis Equipment (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a human body ultrasonic detection real-time guiding strategy based on deep learning, which comprises a demonstrator of human body ultrasonic detection, a skill evaluation strategy based on deep learning and a real-time adjustment strategy based on a sampling principle. The demonstrator is used for collecting pressure signals and attitude signals in the detection process; the skill evaluation strategy divides the human body ultrasonic image according to the effective degree of the contained information, and a training set is made and used for training of a classification neural network, so that the neural network which can judge whether the ultrasonic image meets the diagnosis requirement is obtained; the real-time adjustment strategy is based on a skill evaluation strategy, multi-mode information collected by a demonstrator is added into a deep learning process, a multi-mode information fusion neural network is trained, a sampling set is manufactured, and a sampling principle is combined to realize the function of guiding the real-time adjustment of the ultrasonic probe. The invention can help to complete medical ultrasonic detection, and greatly promotes the application of artificial intelligence in the field of ultrasonic detection.
Description
Technical Field
The invention belongs to the field of data acquisition and deep learning, relates to a data acquisition, skill evaluation and real-time adjustment strategy for human body ultrasonic detection, and particularly relates to a strategy for acquiring human body ultrasonic detection data, applying the data to deep learning and combining a sampling principle to realize skill evaluation and real-time adjustment.
Background
In recent years, deep learning is a rapidly-developed machine learning method, and emphasizes that learning is performed from mass data, so that the problem that a traditional machine learning algorithm is difficult to process, such as high dimension, redundancy and high noise, existing in the mass data is solved, and application scenes in various fields are gradually expanded, such as biological recognition, intelligent driving, financial e-commerce, industrial manufacturing and the like. Medical ultrasound imaging has been a research hotspot and difficulty of machine learning: medical ultrasound imaging is highly specialized, its imaging quality depends on the experience and manipulation of the operator, and requires the operator to grasp a lot of medical knowledge. Common image recognition methods include: the method comprises a statistical decision method, a structural pattern recognition method, a fuzzy pattern recognition method, a support vector machine and an artificial neural network, and can realize ultrasonic image recognition work to a certain degree, and has obvious defects: only the ultrasonic image is considered, and a plurality of information in the process of operating the ultrasonic probe by a doctor, including signals such as the pressure and the posture of the probe, and the like, are ignored. How to acquire data of a doctor operating an ultrasonic probe, how to distinguish complicated ultrasonic images according to imaging effects, how to combine the operating data with the ultrasonic images for deep learning, and how to further guide adjustment of the probe, and the problems are all lack of a referential and systematic solution.
Disclosure of Invention
In view of many problems which cannot be solved by the prior art, the invention aims to provide a human body ultrasonic detection real-time guiding strategy based on deep learning.
The invention is realized by the following technologies:
the utility model provides a human ultrasonic testing real-time guide strategy based on degree of deep learning which characterized in that: the system comprises a demonstrator for human body ultrasonic detection, a skill evaluation strategy based on deep learning and a real-time adjustment strategy based on a sampling principle.
The demonstrator is used for acquiring pressure signals and posture signals generated by a doctor operating a probe in the human body ultrasonic detection process.
The skill evaluation strategy based on deep learning is to divide a human body ultrasonic image according to the effective degree of the contained information, make a training set and train a classification neural network, so that the obtained neural network can judge whether the ultrasonic image in the human body ultrasonic detection process meets the diagnosis requirement.
The real-time adjustment strategy based on the sampling principle is characterized in that on the basis of a skill evaluation strategy, multi-mode information (namely the pressure signal and the attitude signal) collected by a demonstrator is added into a deep learning process, a multi-mode information fusion neural network is trained, a sampling set is manufactured, and the function of guiding the real-time adjustment of an ultrasonic probe is realized by combining the sampling principle.
Further, the teach pendant includes: the device comprises a holding part, an IMU attitude sensor, a force sensor and a clamping part. The holding part is held by an operator, the IMU attitude sensor can output attitude information of the whole device, the force sensor is used for connecting the holding part and the clamping part and outputting force and moment signals of the clamping part at the lower end of the force sensor and the ultrasonic probe, and the clamping part is used for clamping the ultrasonic probe.
Furthermore, the upper part of the holding part is of a cylindrical handle structure, so that an operator can hold the holding part conveniently, and the lower part of the holding part is a circular end face. Eight through holes are formed in the circumferential direction of the outer side of the circular end face, four through holes are formed in the circumferential direction close to the center and are used for fixing the force sensor and the IMU attitude sensor respectively. And a circular groove is formed in the center of the circular end face and used for placing an IMU attitude sensor. Meanwhile, a wire guide groove is formed in the wiring opening direction of the IMU attitude sensor and used for wiring of the sensor.
Preferably, the IMU attitude sensor at least comprises three accelerometers, three gyroscopes and three magnetometers, and can acquire and output an attitude signal of the whole demonstrator.
Furthermore, the upper end face of the force sensor is provided with eight circumferential threaded holes which can be fixed with the holding part. The lower end surface is provided with four circumferential threaded holes which can be fixed with the clamping part.
Preferably, the force sensor is a six-dimensional force/moment sensor, and can acquire force and moment applied to the lower end face of the force sensor.
Furthermore, the clamping part consists of a movable clamping block, a fixed clamping block and a connecting bolt. Four through holes are formed in the circumferential direction of the upper end face of the fixing clamping block and can be fixed on the lower end face of the force sensor. The lower part of the movable clamping block and the fixed clamping block have the same structure: two pairs of same through holes are formed, and the opposite surfaces are curved surfaces with grooves. The connecting bolt passes through the through hole and is screwed down to realize the clamping function. The curved surface with the groove ensures that the clamping block and the probe have larger contact area on one hand, so that the clamping is more stable; on the other hand, the sticky soft material is additionally arranged on the groove, so that the probe is in flexible contact with the clamping block, and the effects of increasing friction force and protecting the probe are achieved.
Preferably, a layer of soft silica gel material is additionally arranged on the curved surface with the groove of the clamping block, so that a better clamping effect and a better protection effect can be achieved. The soft silica gel is formed by mixing two liquids and then gradually solidifying, when the two liquids are not completely solidified, the silica gel liquid is poured on the curved surface of the groove, and a better fixing effect is achieved after the silica gel liquid is solidified.
Further, the deep learning based skill assessment strategy comprises: the human body ultrasonic image training set and the classification neural network.
Furthermore, the human body ultrasonic image training set consists of an ultrasonic image and an evaluation label, the ultrasonic image is a real-time image on the screen of the ultrasonic instrument, and the evaluation label is manually marked, so that the imaging effect of the ultrasonic image can be qualitatively described.
Further, specific judgment indexes of the evaluation label include: whether the ultrasonic image is clear, whether the ultrasonic image contains the organ to be checked, whether the position of the organ in the image is centered, whether a shelter or an interference object exists in the image, and the like.
Further, the classification neural network is a convolutional neural network model and comprises a convolutional layer, an activation layer, a pooling layer and a full-link layer. The human body ultrasonic image training set is used for training the classification neural network, and the obtained network can judge whether the ultrasonic image in the human body ultrasonic detection process meets the diagnosis requirement or not.
Further, the real-time adjustment strategy based on the sampling principle comprises: and a multi-mode information fusion neural network and an ultrasonic probe real-time adjustment mode.
Further, the multi-modal information fusion neural network can simultaneously calculate various types of data and perform fusion processing on various types of information, and is not limited to calculation, identification and prediction of a single image and sensor signal. The neural network finally outputs the prediction result of the ultrasonic image evaluation label, and the result is influenced by the numerical weight of various signals.
Furthermore, the multi-modal information fusion neural network is a convolutional neural network model and comprises a convolutional layer, an activation layer, a pooling layer and a full connection layer.
Further, the multi-modal information fusion neural network is improved on the basis of the classification neural network. After the characteristics of the ultrasonic image are extracted from the characteristic layer, the characteristics are spliced with the sensor signals, and the prediction result of the ultrasonic image evaluation label is finally output through the multi-layer pooling layer, the convolution layer, the activation layer and the full connection layer.
Furthermore, the real-time adjustment mode of the ultrasonic probe is that firstly, the acquired data is screened and divided, and a sampling set is established. After the training of the multi-mode information fusion neural network is completed, the real-time ultrasonic image is read each time, the sampling set is combined for multiple times of prediction, the sampling result with the most ideal effect is obtained, and whether the operation requirement of a doctor is met is judged. If the prediction result meets the requirement, the prediction result is used as an adjustment strategy of the ultrasonic probe, otherwise, the sampling process is repeated.
Preferably, the sampling set contains not only all the signals collected, but also can be combined with a random function to sample within the signal fluctuation range.
The invention has the beneficial effects that:
the invention designs a demonstrator for acquiring data, distinguishes numerous and complicated ultrasonic images according to imaging effects, combines the acquired data with deep learning and sampling principles, and realizes the functions of ultrasonic detection skill evaluation and real-time adjustment. The demonstrator comprises a holding part, an IMU posture sensor, a force sensor and a clamping part, a skill evaluation strategy based on deep learning comprises a human body ultrasonic image training set and a classification neural network, and a real-time adjustment strategy based on a sampling principle comprises a multi-mode information fusion neural network and an ultrasonic probe real-time adjustment mode. And acquiring attitude and contact force signals through a demonstrator with an IMU attitude sensor and a force sensor, and establishing a sampling set. Meanwhile, a human body ultrasonic image training set is established through the ultrasonic images and the evaluation indexes of the artificial marks, a classification neural network is trained, and ultrasonic detection skill assessment is achieved. On the basis, the posture and contact force signals are fused into the classification neural network to obtain a multi-mode information fusion neural network, and the function of guiding the real-time adjustment of the ultrasonic probe is realized by combining a sampling set and the neural network. The invention is mainly oriented to the field of medical ultrasonic imaging, covers the aspects of structural design, data acquisition, deep learning, sampling prediction and the like, and has a set of complete modes. Meanwhile, the mode has better reference, feasibility and portability, can be applied to other fields, and has wide application prospect.
Drawings
FIG. 1 is a flow chart of the system of the present invention.
Fig. 2 is a conceptual diagram of the data acquisition process of the present invention.
FIG. 3 is a schematic diagram of the overall structure of the teaching device of the present invention.
FIG. 4 is a schematic view of a structure of a holding part of the teaching device of the present invention.
FIG. 5 is a bottom view of the grip portion of the teach pendant of the present invention.
FIG. 6 is a schematic diagram of a force sensor configuration for the teach pendant of the present invention.
FIG. 7 is a schematic view of the structure of the clamping part of the teaching device of the present invention.
FIG. 8 is a schematic view of a clamping block of the clamping part of the teaching device of the present invention.
FIG. 9 is a schematic diagram of a neural network according to the present invention.
FIG. 10 is a diagram illustrating evaluation label judgment indexes of the human ultrasound image training set according to the present invention.
FIG. 11 is a schematic structural diagram of the multi-modal information fusion neural network of the present invention.
Fig. 12 is a flow chart of the real-time adjustment mode of the ultrasonic probe according to the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention clearer, the following describes in detail a human body ultrasonic detection real-time guiding strategy based on deep learning, which is provided by the present invention, with reference to the accompanying drawings and embodiments. It is to be understood that the embodiments described are only a few embodiments of the invention, and not all embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
Fig. 1 is a system flowchart of a human body ultrasonic detection real-time guiding strategy based on deep learning according to the present invention. In the medical ultrasound examination process, information we mainly consider includes: ultrasonic images, probe attitude signals and probe contact force signals. And acquiring attitude and contact force signals through a demonstrator with an IMU attitude sensor and a force sensor, and establishing a sampling set. Meanwhile, a human body ultrasonic image training set is established through the ultrasonic images and the evaluation indexes of the artificial marks, a classification neural network is trained, and ultrasonic detection skill assessment is achieved. On the basis, the posture and contact force signals are fused into the classification neural network to obtain a multi-mode information fusion neural network, and the function of real-time adjustment of the ultrasonic probe is realized by combining a sampling set and the neural network. The skill assessment means to judge whether the ultrasound image meets the diagnosis requirements and thus whether the doctor's operation meets the requirements. The real-time adjustment of the ultrasonic probe refers to the adjustment of the posture of the ultrasonic probe and the contact force between the probe and the skin of a human body, so that an ultrasonic image with a better effect is obtained.
Fig. 2 is a conceptual diagram of the data acquisition process of the present invention. In the medical ultrasonic detection process, a doctor holds the demonstrator to perform fixed-point detection, and obtains the posture and contact force signals of the probe. Fig. 3 is a schematic diagram of the overall structure of the teaching device of the present invention. The demonstrator comprises a holding part 1, an IMU posture sensor 2, a force sensor 3 and a clamping part 4. The holding part 1 is held by an operator, the IMU posture sensor 2 can output the posture information of the whole device, the force sensor 3 is used for connecting the holding part 1 and the clamping part 4 and outputting the force and moment signals of the clamping part 4 at the lower end and the ultrasonic probe 5, and the clamping part 4 is used for clamping the ultrasonic probe 5.
Fig. 4 is a schematic view of the structure of the holding part of the teaching device of the present invention. The upper part of the holding part 1 is a cylindrical handle structure 11, which is convenient for an operator to hold. As shown in fig. 5, the bottom view of the grip of the demonstrator of the present invention is a bottom view of the grip, and eight through holes 13 are circumferentially opened on the outer side of the circular end surface of the grip, and four through holes 14 are circumferentially opened near the center for fixing the force sensor 3 and the IMU attitude sensor 2, respectively. And a circular groove is formed in the center of the circular end face and used for placing the IMU attitude sensor 2. Meanwhile, a wire guide groove 12 is formed in the wiring opening direction of the IMU attitude sensor 2 and used for sensor wiring.
Fig. 6 is a schematic diagram of a force sensor structure of the teaching machine of the present invention. The upper end face of the force sensor 3 has eight circumferential threaded holes 32, which can be fixed to the grip portion 1. The lower end surface has four circumferential threaded holes 31 which can be fixed with the holding portion 4. The force sensor 3 can select a six-dimensional force/moment sensor, and can acquire force and moment signals borne by the lower end face of the force sensor.
Fig. 7 is a schematic view of the structure of the clamping part of the teaching device of the present invention. The clamping part 4 is composed of a movable clamping block 43, a fixed clamping block 41 and a connecting bolt 42. Fig. 8 is a schematic structural view of a clamping part fixing clamp block of the demonstrator of the present invention. Four through holes 411 are circumferentially formed in the upper end face of the fixing clamping block 41 and can be fixed to the lower end face of the force sensor 3. The lower parts of the movable clamping block 43 and the fixed clamping block 41 have the same structure: two pairs of identical through holes 413 are formed, and the opposite surfaces are curved surfaces 412 with grooves. The connecting bolt 42 is passed through the through hole and tightened to perform a clamping function. The curved surface 412 with the groove ensures that the clamping block and the probe have larger contact area on one hand, so that the clamping is more stable; on the other hand, the groove is additionally provided with a viscous soft material, so that the ultrasonic probe 5 is in flexible contact with the clamping block, and the effects of increasing friction force and protecting the probe are achieved. A layer of soft silica gel material can be additionally arranged on the curved surface 412 with the groove of the clamping block, so that a better clamping effect and a better protection effect can be achieved. The soft silica gel is formed by mixing two liquids and then gradually solidifying, when the two liquids are not completely solidified, the silica gel liquid is poured on the curved surface of the groove, and a better fixing effect is achieved after the silica gel liquid is solidified.
Fig. 9 is a schematic structural diagram of the neural network of the present invention. The classification neural network is a convolution neural network model and comprises a convolution layer, an activation layer, a pooling layer and a full-connection layer. The human body ultrasonic image training set is used for training the classification neural network, and the obtained network can judge whether the ultrasonic image in the human body ultrasonic detection process meets the diagnosis requirement or not. The human body ultrasonic image training set consists of an ultrasonic image and an evaluation label, the ultrasonic image is a real-time image on a screen of an ultrasonic instrument, and the evaluation label is manually marked, so that the imaging effect of the ultrasonic image can be qualitatively described.
Fig. 10 shows evaluation label judgment indexes of the human ultrasound image training set according to the present invention. The specific judgment indexes of the evaluation label include: whether the ultrasonic image is clear, whether the ultrasonic image contains the organ to be checked, whether the position of the organ in the image is centered, whether a shelter or an interference object exists in the image, and the like.
Fig. 11 is a schematic structural diagram of the multi-modal information fusion neural network according to the present invention. The multi-mode information fusion neural network can simultaneously calculate various types of data and perform fusion processing on various types of information, and is not limited to calculation, identification and prediction of single images and sensor signals. The multi-modal information fusion neural network finally outputs the prediction result of the ultrasonic image label, and the result is influenced by the weight of various signal values. The neural network is a convolutional neural network model and comprises a convolutional layer, an activation layer, a pooling layer and a full-link layer. The multi-mode information fusion neural network is improved on the basis of a classification neural network, after the characteristics of an ultrasonic image are extracted in a characteristic layer, the ultrasonic image is spliced with a sensor signal, and a prediction result of an ultrasonic image evaluation label is finally output through a multi-layer pooling layer, a convolution layer, an activation layer and a full connection layer.
Fig. 12 is a flow chart of a real-time adjustment method of the ultrasonic probe according to the present invention. The real-time adjustment mode of the ultrasonic probe comprises the steps of firstly screening and dividing acquired data and establishing a sampling set. After the training of the multi-mode information fusion neural network is completed, the real-time ultrasonic image is read each time, the sampling set is combined for multiple times of prediction, the sampling result with the most ideal effect is obtained, and whether the operation requirement of a doctor is met is judged. If the prediction result meets the requirement, the prediction result is used as an adjustment strategy of the ultrasonic probe, otherwise, the sampling process is repeated. The sampling set not only contains all the collected signals, but also can be combined with a random function to carry out sampling in a signal fluctuation range.
Claims (10)
1. The utility model provides a human ultrasonic testing real-time guide strategy based on degree of deep learning which characterized in that: the system comprises a demonstrator for human body ultrasonic detection, a skill evaluation strategy based on deep learning and a real-time adjustment strategy based on a sampling principle;
the demonstrator is used for acquiring a pressure signal and an attitude signal generated by a doctor operating a probe in the human body ultrasonic detection process;
the skill evaluation strategy based on deep learning is to divide a human body ultrasonic image according to the effective degree of the contained information, make a training set and train a classification neural network, so as to obtain the neural network which can judge whether the ultrasonic image meets the diagnosis requirement in the human body ultrasonic detection process;
the real-time adjustment strategy based on the sampling principle is characterized in that on the basis of a skill evaluation strategy, multi-mode information collected by a demonstrator is added into a deep learning process, a multi-mode information fusion neural network is trained, a sampling set is manufactured, and the function of guiding the ultrasonic probe to be adjusted in real time is realized by combining the sampling principle.
2. The human body ultrasonic detection real-time guiding strategy as claimed in claim 1, wherein: the demonstrator comprises a holding part, an IMU attitude sensor, a force sensor and a clamping part, wherein the holding part is held by an operator, the IMU attitude sensor can output attitude information of the whole device, the force sensor is used for connecting the holding part and the clamping part and outputting force and moment signals of the clamping part at the lower end and the ultrasonic probe, and the clamping part is used for clamping the ultrasonic probe.
3. The human body ultrasonic detection real-time guiding strategy as claimed in claim 2, wherein: the utility model discloses a IMU gesture sensor, including the portion of gripping, the circular recess that is used for placing IMU gesture sensor is equipped with on the circular terminal surface, the circular recess that grips the portion of upper portion be the cylinder handle structure that makes things convenient for the operator to grip, and the lower part is circular terminal surface, be equipped with the circular recess that is used for placing IMU gesture sensor on the circular terminal surface, be equipped with the through-hole that is used for fixed IMU gesture sensor in the circular recess, be equipped with the through-hole that is used for fixed force.
4. The human body ultrasonic detection real-time guiding strategy as claimed in claim 3, wherein: the upper end face of the force sensor is provided with a circumferential threaded hole fixed with the holding part, and the lower end face of the force sensor is provided with a circumferential threaded hole fixed with the clamping part.
5. The human body ultrasonic detection real-time guiding strategy as claimed in claim 4, wherein: the clamping part comprises activity clamp splice, fixed clamp splice and connecting bolt, and fixed clamp splice up end circumference is opened has the through-hole that can fix terminal surface under force sensor, and the lower partial structure of activity clamp splice and fixed clamp splice is the same: two pairs of identical through holes are formed, the opposite surfaces of the through holes are curved surfaces with grooves, and the connecting bolts penetrate through the paired through holes and are screwed down to achieve the clamping function.
6. The human body ultrasonic detection real-time guiding strategy as claimed in claim 1, wherein: the skill evaluation strategy based on deep learning comprises a human body ultrasonic image training set and a classification neural network, wherein the human body ultrasonic image training set is composed of an ultrasonic image and an evaluation label, the ultrasonic image is a real-time image on a screen of an ultrasonic instrument, and the evaluation label is manually marked, so that the imaging effect of the ultrasonic image can be described qualitatively.
7. The human body ultrasonic detection real-time guiding strategy as claimed in claim 6, wherein: the specific judgment indexes of the evaluation label include: whether the ultrasonic image is clear or not, whether the ultrasonic image contains the examined organ or not, whether the position of the organ in the image is centered or not, and whether a shelter or an interference object exists in the image or not.
8. The human body ultrasonic detection real-time guiding strategy as claimed in claim 6, wherein: the classification neural network is a convolution neural network model and comprises a convolution layer, an activation layer, a pooling layer and a full-connection layer, and the human body ultrasonic image training set is used for training the classification neural network to obtain the neural network capable of judging whether the ultrasonic image meets the diagnosis requirement in the human body ultrasonic detection process.
9. The human body ultrasonic detection real-time guiding strategy as claimed in claim 1, wherein: the real-time adjustment strategy based on the sampling principle comprises the following steps: a multi-mode information fusion neural network and an ultrasonic probe real-time adjustment mode;
the multi-mode information fusion neural network can calculate various types of data at the same time, and performs fusion processing on various types of information without being limited to calculation, identification and prediction of single images and sensor signals; the multi-mode information fusion neural network is a convolutional neural network model and comprises a convolutional layer, an activation layer, a pooling layer and a full-connection layer; the multi-mode information fusion neural network is improved on the basis of a classification neural network, after the characteristics of an ultrasonic image are extracted in a characteristic layer, the ultrasonic image is spliced with a sensor signal, and a prediction result of an ultrasonic image evaluation label is finally output through a multi-layer pooling layer, a convolution layer, an activation layer and a full connection layer.
10. The human body ultrasonic detection real-time guidance strategy of claim 9, characterized in that: the real-time adjustment mode of the ultrasonic probe comprises the steps of firstly screening and dividing acquired data and establishing a sampling set; after the training of the multi-mode information fusion neural network is completed, reading the real-time ultrasonic image each time, performing multiple predictions by combining a sampling set, taking a sampling result with the most ideal effect, and judging whether the operation requirements of doctors are met; if the prediction result meets the requirement, the prediction result is used as an adjustment strategy of the ultrasonic probe, otherwise, the sampling process is repeated.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011486938.0A CN112633342B (en) | 2020-12-16 | 2020-12-16 | Human body ultrasonic detection real-time guiding strategy based on deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011486938.0A CN112633342B (en) | 2020-12-16 | 2020-12-16 | Human body ultrasonic detection real-time guiding strategy based on deep learning |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN112633342A true CN112633342A (en) | 2021-04-09 |
| CN112633342B CN112633342B (en) | 2022-08-16 |
Family
ID=75313846
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202011486938.0A Active CN112633342B (en) | 2020-12-16 | 2020-12-16 | Human body ultrasonic detection real-time guiding strategy based on deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112633342B (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113288204A (en) * | 2021-04-21 | 2021-08-24 | 佛山纽欣肯智能科技有限公司 | Semi-autonomous B-ultrasonic detection system of robot |
| EP4250311A1 (en) * | 2022-03-24 | 2023-09-27 | Koninklijke Philips N.V. | Ultrasound system with customization unit |
| WO2023180167A1 (en) * | 2022-03-24 | 2023-09-28 | Koninklijke Philips N.V. | Ultrasound system with customization unit |
Citations (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20130296707A1 (en) * | 2010-12-18 | 2013-11-07 | Massachusetts Institute Of Technology | User interface for ultrasound scanning system |
| US20160113630A1 (en) * | 2014-10-23 | 2016-04-28 | Samsung Electronics Co., Ltd. | Ultrasound imaging apparatus and method of controlling the same |
| US20160206292A1 (en) * | 2008-08-05 | 2016-07-21 | Guardsman Scientific, Inc. | Systems and methods for managing a patient |
| CN109199445A (en) * | 2018-11-14 | 2019-01-15 | 中聚科技股份有限公司 | A kind of Intelligence Ultrasound fetal rhythm monitoring system |
| CN109512461A (en) * | 2018-11-14 | 2019-03-26 | 中聚科技股份有限公司 | A kind of medical supersonic fetal rhythm monitoring probe location regulation method and holding meanss of popping one's head in |
| CN109602451A (en) * | 2019-02-16 | 2019-04-12 | 河南科技大学第附属医院（河南省显微外科研究所） | A kind of medical ultrasonic auxiliary auto-check system |
| CN109674494A (en) * | 2019-01-29 | 2019-04-26 | 深圳瀚维智能医疗科技有限公司 | Ultrasonic scan real-time control method, device, storage medium and computer equipment |
| CN110675398A (en) * | 2019-10-22 | 2020-01-10 | 深圳瀚维智能医疗科技有限公司 | Mammary gland ultrasonic screening method and device and computer equipment |
| US20200069285A1 (en) * | 2018-08-31 | 2020-03-05 | General Electric Company | System and method for ultrasound navigation |
| CN110974294A (en) * | 2019-12-19 | 2020-04-10 | 上海尽星生物科技有限责任公司 | Ultrasonic scanning method and device |
| CN111260786A (en) * | 2020-01-06 | 2020-06-09 | 南京航空航天大学 | Intelligent ultrasonic multi-mode navigation system and method |
-
2020
- 2020-12-16 CN CN202011486938.0A patent/CN112633342B/en active Active
Patent Citations (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20160206292A1 (en) * | 2008-08-05 | 2016-07-21 | Guardsman Scientific, Inc. | Systems and methods for managing a patient |
| US20130296707A1 (en) * | 2010-12-18 | 2013-11-07 | Massachusetts Institute Of Technology | User interface for ultrasound scanning system |
| US20160113630A1 (en) * | 2014-10-23 | 2016-04-28 | Samsung Electronics Co., Ltd. | Ultrasound imaging apparatus and method of controlling the same |
| US20200069285A1 (en) * | 2018-08-31 | 2020-03-05 | General Electric Company | System and method for ultrasound navigation |
| CN109199445A (en) * | 2018-11-14 | 2019-01-15 | 中聚科技股份有限公司 | A kind of Intelligence Ultrasound fetal rhythm monitoring system |
| CN109512461A (en) * | 2018-11-14 | 2019-03-26 | 中聚科技股份有限公司 | A kind of medical supersonic fetal rhythm monitoring probe location regulation method and holding meanss of popping one's head in |
| CN109674494A (en) * | 2019-01-29 | 2019-04-26 | 深圳瀚维智能医疗科技有限公司 | Ultrasonic scan real-time control method, device, storage medium and computer equipment |
| CN109602451A (en) * | 2019-02-16 | 2019-04-12 | 河南科技大学第附属医院（河南省显微外科研究所） | A kind of medical ultrasonic auxiliary auto-check system |
| CN110675398A (en) * | 2019-10-22 | 2020-01-10 | 深圳瀚维智能医疗科技有限公司 | Mammary gland ultrasonic screening method and device and computer equipment |
| CN110974294A (en) * | 2019-12-19 | 2020-04-10 | 上海尽星生物科技有限责任公司 | Ultrasonic scanning method and device |
| CN111260786A (en) * | 2020-01-06 | 2020-06-09 | 南京航空航天大学 | Intelligent ultrasonic multi-mode navigation system and method |
Non-Patent Citations (3)
| Title |
|---|
| CHANGFAN CHEN, ET.AL: "Ultrasound thermometry using an ultrasound element and deep learning for HIFU", 《IEEE INTERNATIONAL ULTRASONICS SYMPOSIUM (IUS)》 * |
| PIERRE CHATELAIN, ET.AL: "Optimization of ultrasound image quality via visual servoing", 《IEEE INTERNATIONAL CONFERENCE ON ROBOTICS AND AUTOMATION (ICRA)》 * |
| 沈运等: "远程控制超声扫描的远端操纵***", 《机械与电子》 * |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113288204A (en) * | 2021-04-21 | 2021-08-24 | 佛山纽欣肯智能科技有限公司 | Semi-autonomous B-ultrasonic detection system of robot |
| CN113288204B (en) * | 2021-04-21 | 2023-09-29 | 佛山纽欣肯智能科技有限公司 | Semi-autonomous B ultrasonic detection system of robot |
| EP4250311A1 (en) * | 2022-03-24 | 2023-09-27 | Koninklijke Philips N.V. | Ultrasound system with customization unit |
| WO2023180167A1 (en) * | 2022-03-24 | 2023-09-28 | Koninklijke Philips N.V. | Ultrasound system with customization unit |
Also Published As
| Publication number | Publication date |
|---|---|
| CN112633342B (en) | 2022-08-16 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112633342B (en) | Human body ultrasonic detection real-time guiding strategy based on deep learning | |
| CN111046823A (en) | Student classroom participation degree analysis system based on classroom video | |
| CN112155729A (en) | Intelligent automatic planning method and system for surgical puncture path and medical system | |
| DE10032363A1 (en) | Six-axis force sensor applied to fingertip detects deformation of elastic part using e.g. strain gauge | |
| Cula et al. | Bidirectional imaging and modeling of skin texture | |
| CN112206006A (en) | Intelligent auxiliary identification equipment and method for autonomously evaluating benign and malignant thyroid nodules | |
| CN110505383B (en) | Image acquisition method, image acquisition device and endoscope system | |
| CN110456050B (en) | Portable intelligent digital parasite in vitro diagnostic instrument | |
| CN107307870A (en) | A kind of driving condition brain monitoring system and method based near infrared spectrum | |
| CN106767750A (en) | A kind of air navigation aid and system for intelligent robot | |
| CN116531094A (en) | Visual and tactile fusion navigation method and system for cornea implantation operation robot | |
| Huve et al. | Online recognition of the mental states of drivers with an fnirs-based brain-computer interface using deep neural network | |
| Almanza et al. | Robotic hex-nut sorting system with deep learning | |
| CN113171118B (en) | Ultrasonic inspection operation guiding method based on generation type countermeasure network | |
| CN114878087A (en) | Artificial intelligence-based pressure vessel air tightness detection method and device | |
| CN113520473A (en) | Intelligent unmanned nucleic acid sampling system based on flexible sensing | |
| Astudillo et al. | Cost-efficient brainwave controller for automated vehicles route decisions | |
| CN114093488A (en) | Doctor skill level judging method and device based on bone recognition | |
| Choi et al. | Prediction of four degrees of freedom arm movement using EMG signal | |
| Jain et al. | EEG Cortical Source Feature based Hand Kinematics Decoding using Residual CNN-LSTM Neural Network | |
| Gupta et al. | Optimization of Data Quality Related EMG Feature Extraction Parameters to Increase Hand Movement Classification Accuracy | |
| Takahashi et al. | Remarks on fuzzy reasoning-based brain activity recognition with a compact near infrared spectroscopy device and its application to robot control interface | |
| CN107343016A (en) | Biological electricity analyzes and processes platform | |
| Zammouri et al. | Prototype of a BCI-Based Autonomous Communicating Robot for Disability Assessment | |
| CN110134242A (en) | A kind of arm rigidity discrimination method and system based on surface electromyogram signal |
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 | ||
| GR01 | Patent grant |