CN115153647A - Intelligent pancreatic cancer detection method and platform based on ultrasonic endoscope - Google Patents
Intelligent pancreatic cancer detection method and platform based on ultrasonic endoscope Download PDFInfo
- Publication number
- CN115153647A CN115153647A CN202210783361.2A CN202210783361A CN115153647A CN 115153647 A CN115153647 A CN 115153647A CN 202210783361 A CN202210783361 A CN 202210783361A CN 115153647 A CN115153647 A CN 115153647A
- Authority
- CN
- China
- Prior art keywords
- network
- image
- training
- ultrasonic endoscope
- pancreas
- 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
Images
Classifications
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B8/00—Diagnosis using ultrasonic, sonic or infrasonic waves
- A61B8/46—Ultrasonic, sonic or infrasonic diagnostic devices with special arrangements for interfacing with the operator or the patient
- A61B8/461—Displaying means of special interest
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B8/00—Diagnosis using ultrasonic, sonic or infrasonic waves
- A61B8/08—Detecting organic movements or changes, e.g. tumours, cysts, swellings
- A61B8/0833—Detecting organic movements or changes, e.g. tumours, cysts, swellings involving detecting or locating foreign bodies or organic structures
- A61B8/085—Detecting organic movements or changes, e.g. tumours, cysts, swellings involving detecting or locating foreign bodies or organic structures for locating body or organic structures, e.g. tumours, calculi, blood vessels, nodules
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B8/00—Diagnosis using ultrasonic, sonic or infrasonic waves
- A61B8/52—Devices using data or image processing specially adapted for diagnosis using ultrasonic, sonic or infrasonic waves
- A61B8/5207—Devices using data or image processing specially adapted for diagnosis using ultrasonic, sonic or infrasonic waves involving processing of raw data to produce diagnostic data, e.g. for generating an image
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B8/00—Diagnosis using ultrasonic, sonic or infrasonic waves
- A61B8/52—Devices using data or image processing specially adapted for diagnosis using ultrasonic, sonic or infrasonic waves
- A61B8/5215—Devices using data or image processing specially adapted for diagnosis using ultrasonic, sonic or infrasonic waves involving processing of medical diagnostic data
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B8/00—Diagnosis using ultrasonic, sonic or infrasonic waves
- A61B8/52—Devices using data or image processing specially adapted for diagnosis using ultrasonic, sonic or infrasonic waves
- A61B8/5215—Devices using data or image processing specially adapted for diagnosis using ultrasonic, sonic or infrasonic waves involving processing of medical diagnostic data
- A61B8/523—Devices using data or image processing specially adapted for diagnosis using ultrasonic, sonic or infrasonic waves involving processing of medical diagnostic data for generating planar views from image data in a user selectable plane not corresponding to the acquisition plane
Landscapes
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Engineering & Computer Science (AREA)
- Heart & Thoracic Surgery (AREA)
- Molecular Biology (AREA)
- Biophysics (AREA)
- Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
- Pathology (AREA)
- Radiology & Medical Imaging (AREA)
- Biomedical Technology (AREA)
- Veterinary Medicine (AREA)
- Medical Informatics (AREA)
- Physics & Mathematics (AREA)
- Surgery (AREA)
- Animal Behavior & Ethology (AREA)
- General Health & Medical Sciences (AREA)
- Public Health (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Vascular Medicine (AREA)
- Image Processing (AREA)
Abstract
The invention relates to the technical field of medicine and artificial intelligence, in particular to an intelligent pancreatic cancer change detection method and a platform based on an ultrasonic endoscope, which comprises the following steps: intercepting a training image from an ultrasonic endoscope video stream; constructing a positioning classification network, and training the classification network by using a training image to realize the classification of the pancreas position of the ultrasonic endoscope to obtain the pancreas position of the image; constructing a target detection network, respectively extracting and training features according to the image features of each position of pancreas to obtain the target detection network, sending the classified images into the target detection network to obtain an actual canceration region and a suspected canceration region as SE-AlexNet network training samples; cascading the networks after training to construct a pancreatic canceration detection platform; and transmitting the pancreatic visual information to a pancreatic canceration detection platform for lesion judgment. The method can accurately judge the canceration part, ensure the accuracy and the reliability of the judgment result and realize effective auxiliary diagnosis.
Description
Technical Field
The invention relates to the technical field of medicine and artificial intelligence, in particular to an intelligent pancreatic canceration detection method and platform based on an ultrasonic endoscope.
Background
Current methods for early diagnosis of pancreatic cancer generally include Computed Tomography (CT), magnetic Resonance Imaging (MRI), and Endoscopic Ultrasound (EUS), among others. Although CT and MRI have a certain effect on early pancreatic cancer diagnosis, the sensitivity is too low, and ultrasound endoscopy combines ultrasound with ordinary endoscopy, is more accurate than CT and MRI in the aspects of pancreatic cancer detection and staging, and is widely considered as the most reliable and accurate detection method for diagnosing pancreatic masses including pancreatic cancer. Meanwhile, CT and MRI are expensive, influenced by factors such as radioactivity and anaphylaxis of contrast agents, and are limited to be used by patients with metal foreign matters in vivo and renal insufficiency patients. The ultrasonic examination has the advantages of low cost, no wound, no radioactivity, repeatability and the like, and becomes a common examination means for pancreatic local lesions.
The investigation on the existing related intelligent pancreatic cancer diagnosis method finds that the existing intelligent pancreatic cancer diagnosis method still has the following difficulties which need to be mainly solved:
(1) Due to the influence of multiple factors such as technology and equipment, most hospitals are artificially judged by doctors according to the examination result in the pancreatic cancer diagnosis in reality, and the high difficulty in the operation of the EUS and the difficulty in returning the ultrasound image highly require the operation skill and the image recognition capability of the operator. Meanwhile, the accuracy and reliability of the judgment result cannot be guaranteed due to the influence of subjectivity and experience of an operator on the diagnosis result;
(2) The returned images may have the characteristics of unclear target contour, large noise, unstable imaging and the like, which greatly hinders the research on the characteristics of pancreatic lesions in the images. If the algorithm detection effect is good, the display difficulty existing in the returned image must be solved firstly;
(3) When the imaging examination is carried out, the heterogeneity of the pancreatic tumor and the overlapping characteristic of the image symptoms often cause the detection result to have false negative and poor diagnosis effect
Disclosure of Invention
The invention aims to provide an intelligent pancreatic canceration detection method and platform based on an ultrasonic endoscope, which can be used for effectively detecting a canceration part of pancreatic visual information returned by the ultrasonic endoscope, assisting a doctor in accurately diagnosing the pancreatic canceration part and realizing effective auxiliary diagnosis.
The embodiment of the invention is realized by the following technical scheme: an intelligent pancreatic cancer detection method based on an ultrasonic endoscope comprises the following steps:
firstly, intercepting a video stream of an ultrasonic endoscope, and taking an intercepted image as a training image;
secondly, constructing a positioning classification network, and training the classification network by using the training images, so that the classification network can be combined with the position of the ultrasonic endoscope entering the pancreas to carry out primary position classification during feature extraction, and the position of the pancreas where the images are located is obtained;
step three, constructing a target detection network, respectively extracting and training features according to the image features of all the positions of pancreas to obtain the target detection network, sending the classified images into the target detection network, and correspondingly extracting actual cancerous area images and suspected cancerous area images;
step four, constructing an AlexNet network, adding an SE attention mechanism for the AlexNet network, and training the SE-AlexNet network by using the actual canceration area image and the suspected canceration area image;
fifthly, constructing a pancreatic canceration detection platform by the positioning classification network, the target detection network and the SE-AlexNet network which are trained in a cascade manner;
and step six, transmitting the pancreatic visual information of the ultrasonic endoscope to the pancreatic canceration detection platform, and judging whether the pancreatic part is diseased or not.
According to a preferred embodiment, the first step further comprises: and preprocessing the image obtained by interception.
According to a preferred embodiment, the pre-processing includes image thresholding, image denoising, and image enhancement processing.
According to a preferred embodiment, the pancreas positions comprise a pancreas head, a pancreas neck, a pancreas body and a pancreas tail, and the target detection networks are obtained by training according to image characteristics of four pancreas positions.
The invention also provides an intelligent pancreatic cancer detection platform based on the ultrasonic endoscope, which is applied to the method for training the network model through cloud computing and comprises the following steps:
the image preprocessing module is used for intercepting the video stream of the ultrasonic endoscope and preprocessing the intercepted image to be used as a training image;
the positioning and classifying network module is used for positioning the position of the pancreas where the image is located according to the direction of the ultrasonic endoscope entering the pancreas and by combining the characteristic values of the images of the head of the pancreas, the neck of the pancreas, the body of the pancreas and the tail of the pancreas;
the target detection network module is used for preliminarily extracting an actual pancreatic cancer area and a suspected pancreatic cancer area based on the preprocessed image and according to network training characteristics;
the classification network module is used for providing an AlexNet classification network based on an SE attention mechanism, and is used for paying attention to the relationship between Feature map channels containing image features, acquiring the importance of each channel and adjusting calculation resources according to the importance;
according to a preferred embodiment, the platform further comprises a data transmission module, which is used for downloading the network model trained by the cloud platform to the PC terminal.
According to a preferred embodiment, the platform further comprises a threshold judgment module, which is used for judging the accuracy of the result according to a preset threshold, performing transmission control on the data transmission module, and returning the corresponding ultrasonic endoscope image for secondary training if the result is not accurate.
According to a preferred embodiment, the platform further includes a thread processing module, configured to perform thread control of model training according to the memory operation condition.
The technical scheme of the embodiment of the invention at least has the following advantages and beneficial effects: the invention combines the running mode of the ultrasonic endoscope and extracts the characteristic vector by combining the intelligent detection platform, realizes the real-time resolution of the pancreatic part where the image is located, can carry out effective canceration part detection on pancreatic visual information returned by the ultrasonic endoscope, assists doctors in accurately diagnosing the pancreatic canceration part, and realizes effective auxiliary diagnosis.
Drawings
Fig. 1 is a schematic flow chart of an intelligent pancreatic cancer detection method based on an ultrasonic endoscope according to embodiment 1 of the present invention;
fig. 2 is a schematic diagram of a data transmission call provided in embodiment 1 of the present invention;
fig. 3 is a schematic view of thread processing according to embodiment 1 of the present invention;
fig. 4 is a schematic diagram of data interaction between a PC and a cloud terminal provided in embodiment 1 of the present invention;
fig. 5 is a schematic diagram of cloud network training provided in embodiment 1 of the present invention;
fig. 6 is a schematic diagram of a threshold determination process provided in embodiment 1 of the present invention;
fig. 7 is a schematic diagram of a quadratic combining training network according to embodiment 1 of the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are some, but not all embodiments of the present invention. The components of embodiments of the present invention generally described and illustrated in the figures herein may be arranged and designed in a wide variety of different configurations.
Example 1
Referring to fig. 1, an intelligent pancreatic cancer detection method based on an ultrasonic endoscope includes the following steps:
step one, intercepting a video stream of the ultrasonic endoscope, and taking an intercepted image as a training image. For an original ultrasonic endoscope image, image preprocessing is carried out according to the characteristics of an acquired image, and the image preprocessing specifically comprises the following steps: performing threshold segmentation on the image, removing redundant black background in the original ultrasonic endoscope image, and cutting to obtain a region of interest (ROI) with a fixed size so as to reduce irrelevant regions and improve the image quality; carrying out noise reduction processing on the image, and removing interference signals in the original ultrasonic endoscope image to make the regional characteristics more obvious; the image is enhanced, the contrast of the background of the characteristic region is enhanced, and the identification efficiency is improved; after the original ultrasonic endoscope image is subjected to the preprocessing flow, the textures of the pancreas and other organ regions are enhanced, the texture difference among different tissues is more obvious, and the recognition accuracy and speed of the model obtained through final training can be remarkably improved.
Further, in order to ensure the sufficient number of training samples, the image preprocessing step of the implementation further includes image amplification, and in one implementation of the embodiment of the present invention, the preprocessed image is expanded in a horizontal flipping manner, so as to increase the number of training samples; for image expansion, it is worth mentioning that besides the horizontal flipping method given above, the expansion process can be performed in many other ways, such as the common geometric transformation: such as rotation, scaling, clipping, translation, affine transformation, etc., color space transformation: such as contrast change, brightness change, saturation change, channel separation, gray circle, histogram enhancement, color disturbance and the like, the pixel point is operated as follows: such as blurring, sharpening, adding noise, dropout, posing, random advancing, etc., the above expansion mode may be selected according to the specific expansion requirement and the image quality requirement, which is not limited in this embodiment.
The research of the applicant finds that the pancreas can be sequentially divided into four parts, namely a pancreas head, a pancreas neck, a pancreas body and a pancreas tail, and because the local EUS ultrasonic images of the four parts have the same characteristics and no obvious characteristics, the four positions are difficult to distinguish only by images; however, manually resolving the EUS ultrasound image is mostly affected by subjectivity and experience of an operator, so that it is difficult to accurately judge a cancerous part. Based on this, the second method step provided by the embodiment of the present invention specifically includes: and constructing a positioning classification network, training the classification network by using the training images, enabling the classification network to be capable of combining with an ultrasonic endoscope to enter the pancreas direction during feature extraction, and performing primary position classification on feature vector values extracted from images of the pancreas head, the pancreas neck, the pancreas body and the pancreas tail, accurately positioning to obtain the position of the pancreas where the image is located, and assisting medical staff to detect the pancreas. Specifically, if the ultrasonic endoscope enters the pancreas from the pancreas head, image feature vectors are extracted according to a positioning classification network, a target area is detected, and then the position of the ultrasonic endoscope and the image feature vectors are combined, so that the fact that an upcoming image is a pancreas neck image instead of a pancreas body or a pancreas tail can be judged. Similarly, if the ultrasonic endoscope enters the pancreas from other parts of the pancreas, the position of the endoscope at the moment can be positioned in the above mode, so that the accuracy of judging the canceration position is improved.
Preferably, the method provided by the embodiment of the invention specifically comprises the third step of constructing a target detection network, respectively performing feature extraction and training according to the image features of the four pancreatic positions to obtain four target detection networks, sending the classified images into the target detection network, and correspondingly extracting the actual cancerous region images and the suspected cancerous region images.
Preferably, the method provided by the embodiment of the present invention specifically includes the following fourth step: constructing an AlexNet network, adding an SE attention mechanism for the AlexNet network, namely an SE-AlexNet network, and training the SE-AlexNet network by using the actual cancerous region image and the suspected cancerous region image; it should be noted that the SE-AlexNet network is a classification network that introduces an SE attention mechanism, and the SE attention mechanism enables the model to automatically adjust the computational resources according to the importance and improves the network identification performance according to feature map channel feature vectors each including an image feature. In the embodiment of the invention, the SE module is arranged between the AlexNet trunk feature extraction part and the full connection layer and is used for highlighting the feature vector of the ROI area and improving the network identification rate; and a Support Vector Machine (SVM) is used instead of the BP network. The Support Vector Machine (SVM) has very good popularization capability in the applications of nonlinear classification, function approximation, pattern recognition and the like, and gets rid of the constraint of constructing a learning machine from the perspective of bionics formed for a long time. Compared with a neural network, the Support Vector Machine (SVM) method has a firmer mathematical theory basis, can effectively solve the problem of constructing a high-dimensional data model under the condition of limited samples, and has the advantages of strong generalization capability, convergence to global optimum, insensitivity in dimension and the like.
Preferably, the fifth step of the method provided by the embodiment of the present invention specifically includes constructing a pancreatic cancer detection platform by using the localization classification network, the target detection network and the SE-AlexNet network, which are completed by cascade training. Further, the pancreatic cancer detection platform of the embodiment can transmit image data to the cloud through a gigabit network, referring to fig. 4 and 5, the cloud processes the image data and trains the positioning classification network, the target detection network and the SE-AlexNet network, and returns 9 network models after training through cloud computing, so that the trained network models are led into a local PC terminal, and the PC terminal can acquire a result quickly by only leading in an image to be detected. And the cloud computing is used for replacing a local training process, so that the hardware and time cost consumption of a local PC (personal computer) end can be effectively reduced.
Preferably, the sixth step of the method provided by the embodiment of the present invention specifically includes transmitting the visual information of the pancreas of the ultrasonic endoscope to the platform for detecting pancreatic cancer change, and determining whether a lesion occurs in the pancreas. The embodiment of the invention applies an artificial intelligence technology to assist medical staff in diagnosing visual information of the pancreas of the ultrasonic endoscope, utilizes the contrast of the enhanced image to position and classify the pancreas part, detects the target of the suspected canceration area of the pancreas and finally classifies the suspected canceration area and the actual canceration area, and applies a deep learning mode to assist the medical staff in detecting whether the pancreas part is diseased or not, thereby reducing the requirements of the operation skill and the image recognition capability of an operator and ensuring the accuracy and the reliability of a judgment result.
The embodiment of the invention also provides an intelligent pancreatic cancer detection platform based on an ultrasonic endoscope, which is applied to the method and used for training a network model through cloud computing, and comprises the following steps:
the image preprocessing module is used for intercepting the video stream of the ultrasonic endoscope and preprocessing the intercepted image to be used as a training image;
the positioning and classifying network module is used for positioning the position of the pancreas in combination with the characteristic values of images of the head, the neck, the body and the tail of the pancreas according to the direction of the ultrasonic endoscope entering the pancreas;
the target detection network module is used for preliminarily extracting an actual pancreatic cancer area and a suspected pancreatic cancer area based on the preprocessed image and according to network training characteristics;
and the classification network module provides an AlexNet classification network based on an SE attention mechanism, and is used for paying attention to the relationship among Feature map channels containing image features through the SE attention module, acquiring the importance of each channel and adjusting computing resources according to the importance. Specifically, the AlexNet network extracts image features by using laminated convolutional layers, adopts ReLU as an activation function, and applies a Dropout mechanism to prevent the network from over-training due to small sample size of a data set. And a support vector machine technology is used for replacing a BP network, so that the problem of high-dimensional data model construction under the condition of limited samples can be effectively solved.
And the data transmission module is used for downloading the network model trained by the cloud platform to the PC terminal, so that the hardware cost of the PC terminal is reduced, and cloud computing is realized. It should be noted that the data transmission process of the platform includes the following steps: the method comprises the steps of collecting images of the ultrasonic endoscope, uploading the images to a cloud end, carrying out cloud computing by the cloud end, and carrying out computing including a positioning classification network training process, a target detection network training process, an SE-AlexNet network training process, an image preprocessing process and an image amplification process. And after training, the SE-AlexNet network, the positioning classification network and the target detection network are completed and are transmitted back to the PC end from the cloud end, and the PC end receives data, namely 9 trained network models. If the detection result is found to be different from the manual judgment result by the PC terminal, the PC terminal autonomously returns the corresponding ultrasonic endoscope image to perform secondary perfect training, repeats the process, and returns a total of 9 network models of the new SE-AlexNet network, the positioning classification network and the target detection network which are trained to the PC terminal, and the process is circulated in sequence.
And the threshold judgment module is used for judging the accuracy of the result according to a preset threshold, controlling the transmission of the data transmission module, and if the result is not accurate, transmitting back the corresponding ultrasonic endoscope image for secondary training so as to realize cloud autonomous network training. The method comprises the following specific steps: if the result of the artificial detection is different from that of the intelligent network detection, returning a difference image, and if the number of the images does not reach a threshold value, keeping the original network model unchanged; if the number of the images reaches the threshold value, the positioning classification network, the target detection network and the SE-AlexNet network are secondarily combined and trained according to the returned data set, namely, the difference images exist, and finally the trained network is transmitted to the PC terminal, specifically referring to fig. 6 and 7.
The secondary combination training positioning classification network, the target detection network and the SE-AlexNet network are adopted, namely, a plurality of classification network models and target detection models are stored in the cloud. In an implementation manner of the embodiment of the present invention, the classification network model includes: VGG-16, VGG-19, resNet, inclusion-v 1, inclusion-v 2, inclusion-v 3. The object detection network comprises: fast R-CNN, YOLOv1, YOLOv2, YOLOv3, SSD. And combining the network model secondarily according to the returned data set, and finally retraining the weight of the SE-AlexNet network. Referring to fig. 5, specifically: the ultrasonic endoscope image can be divided into the following parts by the positioning classification network: the method comprises the steps of selecting a combination of a classification network model with the best training effect and a target detection network model from 6 classification network models and 5 target detection network models according to the final SE-AlexNet training result, extracting target area training SE-AlexNet classification network weight, finally selecting a network model combination with the best classification effect and storing the network model combination according to different characteristics of each part.
And the thread processing module is used for performing thread control of model training according to the memory running condition, for example, performing multi-thread simultaneous processing training and accelerating the network training speed. The embodiment of the invention reasonably utilizes the storage space of a computer, and simultaneously processes the images to be detected in a multi-thread mode, as shown in fig. 1, 2 and 3, when a plurality of images are detected simultaneously, an empty thread is automatically judged according to the thread occupation condition, the images to be detected are placed in the corresponding empty thread, and the training is called to complete the processing algorithm of each network model, so that the multi-thread parallel operation processing is realized, and the integral operation speed of the platform is improved. The parallel operation processing can fully utilize a CPU to call an operation space, a main program calls a data module as a main control module, data transmission and calling among other functional modules are carried out, and all functions are processed in a modularized mode, so that an image algorithm can be completed only by data calling, the operation time is shortened, and the time cost is reduced. The platform can simultaneously realize the network model training and canceration detection without influencing each other.
The present invention has been described in terms of the preferred embodiment, and it is not intended to be limited to the embodiment. Any modification, equivalent replacement, or improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.
Claims (8)
1. An intelligent pancreatic cancer detection method based on an ultrasonic endoscope is characterized by comprising the following steps:
firstly, intercepting a video stream of an ultrasonic endoscope, and taking an intercepted image as a training image;
secondly, constructing a positioning classification network, and training the classification network by using the training images, so that the classification network can be combined with the position of the ultrasonic endoscope entering the pancreas to carry out primary position classification when the result is output, and the position of the pancreas where the image is located is obtained;
step three, constructing a target detection network, respectively extracting and training features according to the image features of each position of the pancreas to obtain the target detection network, sending the classified images into the target detection network, and correspondingly extracting actual cancerous area images and suspected cancerous area images;
step four, constructing an AlexNet network, adding an SE attention mechanism for the AlexNet network, and training the SE-AlexNet network by using the actual cancerous area image and the suspected cancerous area image;
fifthly, constructing a pancreatic canceration detection platform by the positioning classification network, the target detection network and the SE-AlexNet network which are trained in a cascade manner;
and step six, transmitting the pancreatic visual information of the ultrasonic endoscope to the pancreatic canceration detection platform, and judging whether the pancreatic part is diseased or not.
2. The ultrasound endoscope-based intelligent pancreatic cancer detection method of claim 1, wherein said first step further comprises: and preprocessing the image obtained by interception.
3. The intelligent ultrasonic-endoscope-based pancreatic carcinogenesis detection method of claim 2, wherein said preprocessing includes image thresholding, image denoising, and image enhancement processing.
4. The intelligent ultrasonic endoscope-based pancreatic carcinogenesis detecting method according to claim 1, wherein said pancreatic positions include a head, a neck, a body and a tail, and said target detecting network is trained based on image features of four positions of a pancreas to obtain four target detecting networks.
5. An intelligent pancreas canceration detection platform based on ultrasonic endoscope, applied to the method of any one of claims 1-4, and used for training a network model through cloud computing, wherein the intelligent pancreas canceration detection platform comprises:
the image preprocessing module is used for intercepting the video stream of the ultrasonic endoscope and preprocessing the intercepted image to be used as a training image;
the positioning and classifying network module is used for positioning the position of the pancreas in combination with the characteristic values of images of the head, the neck, the body and the tail of the pancreas according to the direction of the ultrasonic endoscope entering the pancreas;
the target detection network module is used for preliminarily extracting an actual pancreatic canceration region and a suspected canceration region according to network training characteristics based on the preprocessed image;
and the classification network module is used for providing an AlexNet classification network based on an SE attention mechanism, and is used for paying attention to the relationship between Feature map channels containing image features, acquiring the importance of each channel and adjusting computing resources according to the importance.
6. The ultrasonic endoscope-based intelligent pancreatic cancer detection platform of claim 5, wherein said platform further comprises a data transmission module for downloading the network model trained by the cloud platform to the PC.
7. The intelligent ultrasonic endoscope-based pancreatic cancer detection platform of claim 5, further comprising a threshold judgment module for judging the accuracy of the result according to a preset threshold and performing transmission control on the data transmission module, and if the result is not accurate, returning the corresponding ultrasonic endoscope image for secondary training.
8. The ultrasonic endoscope-based intelligent pancreatic cancer detection platform of claim 5, wherein said platform further comprises a thread processing module for performing thread control of model training according to memory operating conditions.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210783361.2A CN115153647A (en) | 2022-07-05 | 2022-07-05 | Intelligent pancreatic cancer detection method and platform based on ultrasonic endoscope |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210783361.2A CN115153647A (en) | 2022-07-05 | 2022-07-05 | Intelligent pancreatic cancer detection method and platform based on ultrasonic endoscope |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115153647A true CN115153647A (en) | 2022-10-11 |
Family
ID=83490866
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210783361.2A Pending CN115153647A (en) | 2022-07-05 | 2022-07-05 | Intelligent pancreatic cancer detection method and platform based on ultrasonic endoscope |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115153647A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116649963A (en) * | 2023-07-27 | 2023-08-29 | 四川省医学科学院·四川省人民医院 | Early warning analysis method and system based on patient limb pose detection |
| CN117152485A (en) * | 2023-07-21 | 2023-12-01 | 华能(广东)能源开发有限公司汕头电厂 | Coal leakage detection method, device and equipment for coal-fired unit |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107767365A (en) * | 2017-09-21 | 2018-03-06 | 华中科技大学鄂州工业技术研究院 | A kind of endoscopic images processing method and system |
| CN109598708A (en) * | 2018-11-28 | 2019-04-09 | 武汉大学人民医院（湖北省人民医院） | The system and method for cancer of pancreas is identified and positioned under a kind of endoscopic ultrasonography |
| CN109614995A (en) * | 2018-11-28 | 2019-04-12 | 武汉大学人民医院（湖北省人民医院） | The system and method for pancreatic duct and pancreas structure is identified under a kind of endoscopic ultrasonography |
| CN111784721A (en) * | 2020-07-01 | 2020-10-16 | 华南师范大学 | Ultrasonic endoscopic image intelligent segmentation and quantification method and system based on deep learning |
| CN111784686A (en) * | 2020-07-20 | 2020-10-16 | 山东省肿瘤防治研究院（山东省肿瘤医院） | Dynamic intelligent detection method, system and readable storage medium for endoscope bleeding area |
| CN113129293A (en) * | 2021-04-26 | 2021-07-16 | 长沙理工大学 | Medical image classification method, medical image classification device, computer equipment and storage medium |
| CN113808057A (en) * | 2021-08-03 | 2021-12-17 | 上海大学 | Endoscope image enhancement method based on unsupervised learning |
-
2022
- 2022-07-05 CN CN202210783361.2A patent/CN115153647A/en active Pending
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107767365A (en) * | 2017-09-21 | 2018-03-06 | 华中科技大学鄂州工业技术研究院 | A kind of endoscopic images processing method and system |
| CN109598708A (en) * | 2018-11-28 | 2019-04-09 | 武汉大学人民医院（湖北省人民医院） | The system and method for cancer of pancreas is identified and positioned under a kind of endoscopic ultrasonography |
| CN109614995A (en) * | 2018-11-28 | 2019-04-12 | 武汉大学人民医院（湖北省人民医院） | The system and method for pancreatic duct and pancreas structure is identified under a kind of endoscopic ultrasonography |
| CN111784721A (en) * | 2020-07-01 | 2020-10-16 | 华南师范大学 | Ultrasonic endoscopic image intelligent segmentation and quantification method and system based on deep learning |
| CN111784686A (en) * | 2020-07-20 | 2020-10-16 | 山东省肿瘤防治研究院（山东省肿瘤医院） | Dynamic intelligent detection method, system and readable storage medium for endoscope bleeding area |
| CN113129293A (en) * | 2021-04-26 | 2021-07-16 | 长沙理工大学 | Medical image classification method, medical image classification device, computer equipment and storage medium |
| CN113808057A (en) * | 2021-08-03 | 2021-12-17 | 上海大学 | Endoscope image enhancement method based on unsupervised learning |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117152485A (en) * | 2023-07-21 | 2023-12-01 | 华能(广东)能源开发有限公司汕头电厂 | Coal leakage detection method, device and equipment for coal-fired unit |
| CN116649963A (en) * | 2023-07-27 | 2023-08-29 | 四川省医学科学院·四川省人民医院 | Early warning analysis method and system based on patient limb pose detection |
| CN116649963B (en) * | 2023-07-27 | 2023-10-20 | 四川省医学科学院·四川省人民医院 | Early warning analysis method and system based on patient limb pose detection |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110599448B (en) | Migratory learning lung lesion tissue detection system based on MaskScoring R-CNN network | |
| CN108010021B (en) | Medical image processing system and method | |
| CN106682435B (en) | System and method for automatically detecting lesion in medical image through multi-model fusion | |
| US10510144B2 (en) | System and method for detection of suspicious tissue regions in an endoscopic procedure | |
| CN110930416B (en) | MRI image prostate segmentation method based on U-shaped network | |
| WO2021093448A1 (en) | Image processing method and apparatus, server, medical image processing device and storage medium | |
| CN115153647A (en) | Intelligent pancreatic cancer detection method and platform based on ultrasonic endoscope | |
| CN109492547B (en) | Nodule identification method and device and storage medium | |
| WO2022037642A1 (en) | Method for detecting and classifying lesion area in clinical image | |
| CN111899244B (en) | Image segmentation method, network model training method, device and electronic equipment | |
| CN112700461B (en) | System for pulmonary nodule detection and characterization class identification | |
| Souaidi et al. | A new automated polyp detection network MP-FSSD in WCE and colonoscopy images based fusion single shot multibox detector and transfer learning | |
| Sun et al. | Automatic diagnosis of thyroid ultrasound image based on FCN-AlexNet and transfer learning | |
| Zhao et al. | Attention residual convolution neural network based on U-net (AttentionResU-Net) for retina vessel segmentation | |
| Mansour | Using genetic algorithm for identification of diabetic retinal exudates in digital color images | |
| CN113409273B (en) | Image analysis method, device, equipment and medium | |
| CN116228690A (en) | Automatic auxiliary diagnosis method for pancreatic cancer and autoimmune pancreatitis based on PET-CT | |
| CN113889238B (en) | Image identification method and device, electronic equipment and storage medium | |
| Huang et al. | Semantic segmentation of breast ultrasound image with pyramid fuzzy uncertainty reduction and direction connectedness feature | |
| CN111275719B (en) | Calcification false positive recognition method, device, terminal and medium and model training method and device | |
| Ashame et al. | Abnormality Detection in Eye Fundus Retina | |
| CN114140830A (en) | Repeated identification inhibition method based on circulating tumor cell image | |
| CN115187621A (en) | Automatic U-Net medical image contour extraction network integrating attention mechanism | |
| CN113222989A (en) | Image grading method and device, storage medium and electronic equipment | |
| Acharya et al. | Long short-term memory model based microaneurysm sequence classification in fundus images |
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 |