CN111916206A - CT image auxiliary diagnosis system based on cascade connection - Google Patents

CT image auxiliary diagnosis system based on cascade connection Download PDF

Info

Publication number
CN111916206A
CN111916206A CN202010772672.XA CN202010772672A CN111916206A CN 111916206 A CN111916206 A CN 111916206A CN 202010772672 A CN202010772672 A CN 202010772672A CN 111916206 A CN111916206 A CN 111916206A
Authority
CN
China
Prior art keywords
cascade
image
server
network
convolution kernel
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
Application number
CN202010772672.XA
Other languages
Chinese (zh)
Other versions
CN111916206B (en
Inventor
刘然
赵洋
易琳
刘亚琼
陈希
崔珊珊
王斐斐
陈丹
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Chongqing University
Chongqing University Cancer Hospital
Original Assignee
Chongqing University
Chongqing University Cancer Hospital
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Chongqing University, Chongqing University Cancer Hospital filed Critical Chongqing University
Priority to CN202010772672.XA priority Critical patent/CN111916206B/en
Publication of CN111916206A publication Critical patent/CN111916206A/en
Application granted granted Critical
Publication of CN111916206B publication Critical patent/CN111916206B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
    • G16H50/00—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics
    • G16H50/20—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for computer-aided diagnosis, e.g. based on medical expert systems
    • 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
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
    • G16H30/00—ICT specially adapted for the handling or processing of medical images
    • G16H30/20—ICT specially adapted for the handling or processing of medical images for handling medical images, e.g. DICOM, HL7 or PACS
    • 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
    • G06T2207/20081—Training; Learning
    • 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
    • G06T2207/20084—Artificial neural networks [ANN]
    • 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)
  • Health & Medical Sciences (AREA)
  • Medical Informatics (AREA)
  • General Health & Medical Sciences (AREA)
  • Public Health (AREA)
  • Epidemiology (AREA)
  • Radiology & Medical Imaging (AREA)
  • Primary Health Care (AREA)
  • Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
  • Biomedical Technology (AREA)
  • Pathology (AREA)
  • Databases & Information Systems (AREA)
  • Data Mining & Analysis (AREA)
  • Quality & Reliability (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Apparatus For Radiation Diagnosis (AREA)

Abstract

The invention relates to a cascade-based CT image auxiliary diagnosis system, which comprises a server end, wherein the server end comprises a server, and an input device and an output device which are connected with the server, and an image preprocessing module and a cascade detection model are stored in the server; the cascade detection model comprises a convolutional neural network, a region suggestion network, an ROI Pooling layer and a plurality of cascade detectors. The invention combines medical imaging with computer image processing to complete automatic detection of pathological changes. Aiming at the problem that the CT focus features are complex and difficult to extract, the cascade detection model introduces a deformable convolution kernel to replace a standard convolution kernel, namely 2D offset is added in the standard convolution kernel, so that the focus features can be better extracted. The experimental result shows that the invention can effectively improve the detection quality of pathological changes, help doctors to complete case classification and focus identification more quickly and accurately, and reduce the workload of doctors.

Description

CT image auxiliary diagnosis system based on cascade connection
Technical Field
The present invention relates to the field of image processing technologies, and in particular, to a CT image processing apparatus.
Background
The incidence of cancer is on the rise with aging population, industrialization, progress of urbanization, change of lifestyle, and the like. Cancer has now become one of the major obstacles affecting our quality of life and fitness level, especially lung cancer, lymphoma. This is because most lung cancer, lymphoma patients do not have distinct specific symptoms in the early stage, and therefore, they are not found until the cancer is examined in a hospital at an advanced stage, at which time the best period of therapeutic recovery has been missed. With the development of medical imaging, doctors can diagnose cancer in time at an early stage through CT images, thereby solving the problem of high cancer fatality rate. With the appearance of a large number of CT images, heavy working pressure and severe fatigue feeling are brought to doctors by a large number of repeated image screening, so that the false detection and the missed detection rate are improved.
In recent years, with the development of deep learning, computer-aided diagnosis has become a very successful research field. How to combine medical imaging with computer image processing technology and automatically extract image features by applying a deep learning algorithm to detect and diagnose lesions becomes a key problem for realizing computer-aided diagnosis (CAD) of high-performance CT images.
Disclosure of Invention
In view of the above, in order to solve the above key problems, the present invention aims to provide a CT image aided diagnosis system based on cascade, so as to help solve the problems of large workload of manually performing CT image screening, and easily causing fatigue, false detection and missing detection.
The invention relates to a cascade-based CT image auxiliary diagnosis system, which comprises a server end, an input device and an output device, wherein the input device and the output device are connected with the server;
the image preprocessing module is used for converting an original CT picture into a format which can be accepted by a neural network;
the cascade detection model comprises a convolutional neural network, a region suggestion network, an ROI Powing layer and a plurality of cascade detectors;
the convolutional neural network is used for extracting the characteristics of the image input by the image preprocessing module and inputting the extracted characteristic diagram into an ROI Pooling layer;
the area suggestion network is used for generating a group of target suggestion areas with different sizes and proportions at each anchor point position of the feature map generated by the convolutional neural network, and inputting the generated target suggestion areas into an ROI Pooling layer;
the ROI Pooling layer is used for adjusting a feature map input by the convolutional neural network and a target suggestion region input by the region suggestion network into feature maps with the same size and inputting the feature maps into each detector;
in the plurality of cascaded detectors, the optimized bounding box of the previous detector is used as the input of the next detector, and the detectors are used for carrying out bounding box classification and regression.
Further, the convolutional neural network adopts a residual error network as a backbone network, the residual error network adopts a deformable convolution kernel, and the deformable convolution kernel increases spatial sampling positions by utilizing extra offset in a standard convolution kernel and automatically learns the offset from a target detection task. The deformable convolution process is represented as:
Figure RE-GDA0002674891270000021
wherein the grid R defines the receptive field size and the expansion coefficient, pnRepresents the nth position in the grid R, p0Representing the first position in the grid R, x is the input feature map, w is the sample weight, and y represents the output feature map; Δ pnRepresenting the offset position relative to a standard convolution kernel; after the deformable convolution, the sampling point is at the irregular offset position pn+ΔpnThe above.
Further, the image preprocessing module, when invoked for execution by a server, implements the steps of:
1) converting the CT picture with the original storage format of DICOM into a 16-bit portable network graphic file to realize lossless conversion;
2) scaling the CT picture processed in the step 1) to a floating point number within the range of 0-255 by a sliding window technology, namely changing the CT picture into a gray scale image;
3) selecting front and back axial slices adjacent to the pathological section, splicing the front and back axial slices and the pathological section together as three channels to form an RGB image as the input of a cascade detection model; wherein, the pathological section is a central channel, and the adjacent sections are respectively used as a front channel and a rear channel;
4) and (4) resizing all the images processed in the step 3 into 512 by 512 pixels.
Further, the cascade-based CT image auxiliary diagnosis system further includes a mobile terminal, and the server is used for identifying and processing the CT image sent by the mobile terminal.
The invention has the beneficial effects that:
the invention discloses a cascade-based CT image auxiliary diagnosis system, which combines medical imaging technology and computer image processing technology to complete automatic detection of lesions. The Cascade detection model provided by the invention is based on a two-stage target detection network Cascade R-CNN, and aims at the problem that the CT focus features are complex and difficult to extract, a deformable convolution kernel is introduced into the Cascade detection model to replace a standard convolution kernel, and a 2D offset is added into the standard convolution kernel, so that the focus features can be better extracted. Experimental results show that the lesion detection network used by the cascade-connected CT image-based auxiliary diagnosis system can effectively improve the lesion detection quality, help doctors to complete case classification and lesion identification more quickly and accurately, and reduce the workload of doctors.
Drawings
Fig. 1 is a schematic diagram of a server side.
Fig. 2 is a network flow diagram of a cascade detection model.
FIG. 3 is a schematic diagram of a deformable convolution kernel.
Fig. 4 is a field map of the standard convolution (a) and the deformable convolution (b).
Fig. 5 is a graph of the results of different detection models for the prediction of small target lesions. The rectangular box in graph (c) is the true label for the small target lesion; panel (d) is the actual prediction of small target lesions in panel (c) using fast R-CNN. From the graph (d), it can be seen that the prediction box does not appear in the true tag region, and that fast R-CNN does not detect a small target lesion. And the part (e) is the actual prediction result of the small target lesion in the part (c) by adopting a cascade detection model. From graph (e), the prediction box appears in the true label region, which detects the small target lesion in graph (c).
Fig. 6 is a graph of the prediction results of different detection models for false positive lesion areas. The rectangular box in graph (f) is the true label of the target lesion; panel (g) is the actual prediction of the target lesion in panel (f) using fast R-CNN. From the graph (g), it can be seen that the prediction box does not appear at the true label, but appears elsewhere, indicating that Faster R-CNN does not detect the target lesion, but has a misprediction. And (h) the actual prediction result of the small target lesion in the image (f) is obtained by adopting the cascade detection model, and the fact that the prediction frame appears in the real label region can be known from the image, so that the target lesion is detected by the cascade detection model, and the false prediction of a false positive lesion region is successfully avoided.
Detailed Description
The invention is further described below with reference to the figures and examples.
The embodiment of the cascade-based CT image auxiliary diagnosis system comprises a server side, wherein the server side comprises a server, and an input device and an output device which are connected with the server, and an image preprocessing module and a cascade detection model are stored in the server.
The input device includes an I/O device, and the server may acquire the CT image from the input device and transmit the processing result of the image to the output device.
The image preprocessing module is used for converting the original CT picture into a format which can be accepted by a neural network. The image preprocessing module realizes the following steps when being called and executed by a server:
1) obtaining and reading an initial medical CT image from input equipment, and converting the CT image with an original storage format of DICOM into a 16-bit Portable Network Graphics (PNG) file to realize lossless conversion;
2) scaling the CT picture processed in the step 1) to a floating point number within the range of 0-255 by a sliding window technology, namely changing the CT picture into a gray scale image;
3) since the CT images have a three-dimensional structure, each set consists of tens of axial slices. The image preprocessing module selects a front axial slice and a rear axial slice which are adjacent to the pathological section, and the front axial slice and the rear axial slice are spliced together with the pathological section as three channels to form an RGB image which is used as the input of a cascade detection model, so that more detailed information is provided; wherein, the pathological section is a central channel, and the adjacent sections are respectively used as a front channel and a rear channel;
4) and (4) resizing all the images processed in the step 3 into 512 by 512 pixels.
And after finishing the CT image preprocessing, the server calls the trained cascade detection model to generate the predicted focus identification position.
In this embodiment, the cascade detection model includes a convolutional neural network, a region suggestion network, an ROI firing layer, and a plurality of cascade detectors.
In the current computer vision target detection network, a two-stage target detection network is mainly selected for pursuing accuracy. The two-stage detection Network generates a candidate Region by means of a Region suggestion Network (RPN), and the candidate Region selects positive and negative samples and then is further classified and regressed to adjust the position of a candidate frame. Currently, the mainstream two-phase target detection network is Faster R-CNN. The Cascade detection model provided in this embodiment is an improved two-stage target detection network Cascade R-CNN, and a flowchart of the Cascade detection model for detecting a CT lesion region is shown in fig. 2.
The convolutional neural network is used for extracting the characteristics of the image input by the image preprocessing module and inputting the extracted characteristic diagram into an ROI Pooling layer. In this embodiment, the convolutional neural Network uses a Residual Network (ResNet) as a backbone Network, and the ResNet can accelerate convergence of the Network through a jump connection.
The region suggestion network RPN is used for generating a group of target suggestion regions with different sizes and proportions at each anchor point position of the feature map generated by the convolutional neural network, and inputting the generated target suggestion regions into an ROI Pooling layer.
And the ROI Pooling layer is used for adjusting the feature map input by the convolutional neural network and the target suggestion region input by the region suggestion network into feature maps with the same size and inputting the feature maps into each detector.
In the plurality of cascaded detectors, the optimized bounding box of the previous detector is used as the input of the next detector, and the detectors are used for carrying out bounding box classification and regression. With the optimization of the cascade detector, the Intersection Over Unit (IOU) of the prediction bounding box and the group Truth is continuously improved, so that false positive lesion areas can be effectively avoided. Finally, the addition of the cascaded IOU does not result in a drastic reduction of the positive samples and at the same time improves the accuracy of the bounding box. In this embodiment, 3 cascaded detectors are used, and the threshold values are set to 0.3,0.4, and 0.5, respectively. Of course the number of detectors may be adjusted as desired in different embodiments.
Since the shape of the lesion is generally irregular and has abstract regular expression, it is difficult to extract complex lesion features by directly using standard convolution. To better extract features, we introduce a Deformable Convolution (DCN) in the ResNet as an improvement to the above implementation. The deformable convolution kernel increases the spatial sampling position by using an extra offset in the standard convolution kernel and automatically learns the offset from the object detection task. It adds a two-dimensional offset to the regular grid sample locations of the convolution kernel. Shown in fig. 3 is a 3 x 3 deformable convolution kernel where the upper pass learns the offset through additional convolution layers, and a different offset is learned at each location of the input signature. After the shift, the sliding window of the standard convolution kernel is no longer a regular window, as shown by the rectangular box of the input feature map. And finally, obtaining an output characteristic diagram through deformable convolution according to the input characteristic diagram and the calculated offset. During training, the convolution kernel and the offset of the output characteristic are simultaneously learned through back propagation. The deformable convolution process is represented as:
Figure RE-GDA0002674891270000061
where the grid R defines the receptive field size and the expansion coefficient, e.g., R { (-1, -1), (-1,0) … … (-1, -1), (-1, -1) }, p in a common 3 x 3 convolution kernelnRepresents the nth position in the grid R, p0Representing the first position in the grid R, x is the input feature map, w is the sample weight, and y represents the output feature map; Δ pnIndicating that the sample point is at an irregular offset position p after a deformable convolution with respect to the offset position of the standard convolution kerneln+ΔpnThe above. Due to the offset Δ pnThe position of the feature map after the offset is calculated by a bilinear interpolation method, and four pixel points around the corresponding coordinate are searched to replace the sampling pixel points after the offset.
Fig. 4 shows the receptive field in a standard convolution and a deformable convolution. It can be seen that the receptive field of the standard convolution is relatively fixed, while the receptive field of the deformable convolution is adaptable. After the deformable convolution is adopted, the receptive field is enlarged, and the sampling points are concentrated near a lesion area, so that the network focuses attention on the position of a lesion more when extracting features, and targets with different sizes and shapes are better covered.
As an improvement to the above embodiment, the cascade-based CT image aided diagnosis system further includes a mobile terminal, and the server is configured to identify and process a CT image sent by the mobile terminal. Mobile terminals include, but are not limited to, desktop computers, tablet computers, cell phones, and the like. The improvement enables the server side to directly carry out automatic detection on the locally stored CT images and also to identify and process the CT images sent by the mobile terminal. Firstly, a mobile terminal and a server end establish TCP socket connection; through TCP socket, the mobile terminal sends a request message to the server, and transmits the CT image to be processed as request data; and after receiving the terminal response, the server calls the cascade detection model stored on the local computer to generate the position of the focus and returns the prediction result to the terminal.
The following uses the DeepLesion dataset to evaluate the cascade-based CT image-assisted diagnosis system in this embodiment.
The DeepLesion data set is different from the traditional medical image data set with a single type, and the DeepLesion comprises a plurality of lesions, including lung nodules, liver lesions, lymph node enlargement, kidney lesions, bone tissue lesions and the like. The hyper-parameter settings in the cascade detection model in this embodiment are as follows in table 1.
TABLE 1 hyper-parameter settings in a cascaded detection model
Figure BDA0002617234310000071
This embodiment divides all CT images into a training set, a validation set, and a test set. In the training process of the cascade detection model in the present embodiment, ResNet is used as a feature extractor. The initial weights at stages C1 to C5 in the ResNet network are pre-training weights obtained on ImageNet, which we only need to fine-tune during the training phase. In the regional recommendation network, the top 2000 candidate boxes with higher scores are selected according to the ranking of the confidence scores.
In order to evaluate the cascade detection model in this embodiment, the embodiment performs experiments on different models and compares the results. In subjective evaluation, the model prediction results of the same slice are visualized and displayed, and the experimental results are shown in fig. 5 and 6. It can be seen that, compared with the fast R-CNN model, the cascade-based CT image aided diagnosis system of the present embodiment can reduce the occurrence of missing detection and false detection to some extent. As can be seen from the results in FIG. 5, the cascade detection model in this example successfully detected some small target lesions that were not detected by fast R-CNN. From the results of fig. 6, it can be seen that the cascade detection model in this embodiment also successfully avoids the false prediction of the false-positive lesion area. In addition, we evaluated by objective evaluation indexes-separately calculated the sensitivity of the model under different FPS (false positives). As shown in Table 2, compared with Faster R-CNN, the cascade detection model in this embodiment has a larger improvement under all FPS values. Therefore, the cascade-based CT image aided diagnosis system can effectively reduce the occurrence of missed detection and false detection, and can be used as aided diagnosis equipment for assisting doctors in detecting lesions.
TABLE 2 sensitivity index for the two models at different FPS's (0.5,1,2,4)
Figure BDA0002617234310000081
Finally, although the present invention has been described in detail with reference to the preferred embodiments, it will be understood by those skilled in the art that various changes and modifications may be made therein without departing from the spirit and scope of the present invention as defined by the appended claims.

Claims (4)

1. The CT image auxiliary diagnosis system based on the cascade connection is characterized by comprising a server, wherein the server comprises a server, the server further comprises an input device and an output device which are connected with the server, and an image preprocessing module and a cascade detection model are stored in the server.
The image preprocessing module is used for converting the original CT picture into a format which can be accepted by a neural network;
the cascade detection model comprises a convolutional neural network, a region suggestion network, an ROI Powing layer and a plurality of cascade detectors;
the convolutional neural network is used for extracting the characteristics of the image input by the image preprocessing module and inputting the extracted characteristic diagram into an ROI Pooling layer;
the area suggestion network is used for generating a group of target suggestion areas with different sizes and proportions at each anchor point position of the feature map generated by the convolutional neural network, and inputting the generated target suggestion areas into an ROI Pooling layer;
the ROI Pooling layer is used for adjusting the feature map input by the convolutional neural network and the target suggestion region input by the region suggestion network into feature maps with the same size and inputting the feature maps into each detector;
in the plurality of cascaded detectors, the optimized bounding box of the previous detector is used as the input of the next detector, and the detectors are used for carrying out bounding box classification and regression.
2. The cascade-based CT image-assisted diagnosis system according to claim 1, wherein: the convolutional neural network adopts a residual error network as a backbone network, the residual error network adopts a deformable convolution kernel, the deformable convolution kernel increases a spatial sampling position by utilizing extra offset in a standard convolution kernel, and automatically learns the offset from a target detection task, and the deformable convolution process is represented as follows:
Figure RE-FDA0002674891260000011
wherein the grid R defines the receptive field size and the expansion coefficient, pnRepresents the nth position in the grid R, p0Representing the first position in the grid R, x is the input feature map, w is the sample weight, and y represents the output feature map; Δ pnRepresenting the offset position relative to a standard convolution kernel; after the deformable convolution, the sampling point is at the irregular offset position pn+ΔpnThe above.
3. The cascade-based CT image-assisted diagnosis system according to claim 1, wherein: the image preprocessing module realizes the following steps when being called and executed by a server:
1) converting the CT picture with the original storage format of DICOM into a 16-bit portable network graphic file to realize lossless conversion;
2) scaling the CT picture processed in the step 1) to a floating point number within the range of 0-255 by a sliding window technology, namely changing the CT picture into a gray scale image;
3) selecting front and back axial slices adjacent to the pathological section, splicing the front and back axial slices and the pathological section together as three channels to form an RGB image as the input of a cascade detection model; wherein, the pathological section is a central channel, and the adjacent sections are respectively used as a front channel and a rear channel;
4) and (4) resizing all the images processed in the step 3 into 512 by 512 pixels.
4. The cascade-based CT image-assisted diagnosis system according to claim 1, wherein: the CT image processing system further comprises a mobile terminal, and the server side is used for identifying and processing the CT image sent by the mobile terminal.
CN202010772672.XA 2020-08-04 2020-08-04 CT image auxiliary diagnosis system based on cascade connection Active CN111916206B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202010772672.XA CN111916206B (en) 2020-08-04 2020-08-04 CT image auxiliary diagnosis system based on cascade connection

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202010772672.XA CN111916206B (en) 2020-08-04 2020-08-04 CT image auxiliary diagnosis system based on cascade connection

Publications (2)

Publication Number Publication Date
CN111916206A true CN111916206A (en) 2020-11-10
CN111916206B CN111916206B (en) 2023-12-08

Family

ID=73288141

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202010772672.XA Active CN111916206B (en) 2020-08-04 2020-08-04 CT image auxiliary diagnosis system based on cascade connection

Country Status (1)

Country Link
CN (1) CN111916206B (en)

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113537350A (en) * 2021-07-16 2021-10-22 商汤集团有限公司 Image processing method and device, electronic equipment and storage medium
CN114140668A (en) * 2021-12-09 2022-03-04 深圳市资福医疗技术有限公司 Stomach part recognition model training method, part recognition method and device
CN116504076A (en) * 2023-06-19 2023-07-28 贵州宏信达高新科技有限责任公司 Expressway traffic flow prediction method based on ETC portal data
CN117152509A (en) * 2023-08-28 2023-12-01 北京透彻未来科技有限公司 Stomach pathological diagnosis and typing system based on cascade deep learning

Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108230323A (en) * 2018-01-30 2018-06-29 浙江大学 A kind of Lung neoplasm false positive screening technique based on convolutional neural networks
CN108806793A (en) * 2018-04-17 2018-11-13 平安科技（深圳）有限公司 Lesion monitoring method, device, computer equipment and storage medium
CN109886286A (en) * 2019-01-03 2019-06-14 武汉精测电子集团股份有限公司 Object detection method, target detection model and system based on cascade detectors
US10325179B1 (en) * 2019-01-23 2019-06-18 StradVision, Inc. Learning method and learning device for pooling ROI by using masking parameters to be used for mobile devices or compact networks via hardware optimization, and testing method and testing device using the same
CN109949297A (en) * 2019-03-20 2019-06-28 天津工业大学 Pulmonary nodule detection method based on Reception and Faster R-CNN
CN110599448A (en) * 2019-07-31 2019-12-20 浙江工业大学 Migratory learning lung lesion tissue detection system based on MaskScoring R-CNN network
WO2020020472A1 (en) * 2018-07-24 2020-01-30 Fundación Centro Tecnoloxico De Telecomunicacións De Galicia A computer-implemented method and system for detecting small objects on an image using convolutional neural networks

Patent Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108230323A (en) * 2018-01-30 2018-06-29 浙江大学 A kind of Lung neoplasm false positive screening technique based on convolutional neural networks
CN108806793A (en) * 2018-04-17 2018-11-13 平安科技（深圳）有限公司 Lesion monitoring method, device, computer equipment and storage medium
WO2020020472A1 (en) * 2018-07-24 2020-01-30 Fundación Centro Tecnoloxico De Telecomunicacións De Galicia A computer-implemented method and system for detecting small objects on an image using convolutional neural networks
CN109886286A (en) * 2019-01-03 2019-06-14 武汉精测电子集团股份有限公司 Object detection method, target detection model and system based on cascade detectors
US10325179B1 (en) * 2019-01-23 2019-06-18 StradVision, Inc. Learning method and learning device for pooling ROI by using masking parameters to be used for mobile devices or compact networks via hardware optimization, and testing method and testing device using the same
CN109949297A (en) * 2019-03-20 2019-06-28 天津工业大学 Pulmonary nodule detection method based on Reception and Faster R-CNN
CN110599448A (en) * 2019-07-31 2019-12-20 浙江工业大学 Migratory learning lung lesion tissue detection system based on MaskScoring R-CNN network

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
JIFENG DAI等: "Deformable Convolutional Networks", 《PROCEEDINGS OF THE IEEE INTERNATIONAL CONFERENCE ON COMPUTER VISION》, pages 764 - 773 *

Cited By (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113537350A (en) * 2021-07-16 2021-10-22 商汤集团有限公司 Image processing method and device, electronic equipment and storage medium
CN113537350B (en) * 2021-07-16 2023-12-22 商汤集团有限公司 Image processing method and device, electronic equipment and storage medium
CN114140668A (en) * 2021-12-09 2022-03-04 深圳市资福医疗技术有限公司 Stomach part recognition model training method, part recognition method and device
CN116504076A (en) * 2023-06-19 2023-07-28 贵州宏信达高新科技有限责任公司 Expressway traffic flow prediction method based on ETC portal data
CN117152509A (en) * 2023-08-28 2023-12-01 北京透彻未来科技有限公司 Stomach pathological diagnosis and typing system based on cascade deep learning
CN117152509B (en) * 2023-08-28 2024-04-30 北京透彻未来科技有限公司 Stomach pathological diagnosis and typing system based on cascade deep learning

Also Published As

Publication number Publication date
CN111916206B (en) 2023-12-08

Similar Documents

Publication Publication Date Title
CN110111313B (en) Medical image detection method based on deep learning and related equipment
CN109493308B (en) Medical image synthesis and classification method for generating confrontation network based on condition multi-discrimination
CN111916206B (en) CT image auxiliary diagnosis system based on cascade connection
CN111445478B (en) Automatic intracranial aneurysm region detection system and detection method for CTA image
CN111524137B (en) Cell identification counting method and device based on image identification and computer equipment
CN112233117A (en) New coronary pneumonia CT detects discernment positioning system and computing equipment
CN109363699B (en) Method and device for identifying focus of breast image
CN109754007A (en) Peplos intelligent measurement and method for early warning and system in operation on prostate
US11967181B2 (en) Method and device for retinal image recognition, electronic equipment, and storage medium
CN109363697B (en) Method and device for identifying focus of breast image
US11587231B2 (en) Comprehensive detection device and method for cancerous region
CN112132166A (en) Intelligent analysis method, system and device for digital cytopathology image
CN111798424B (en) Medical image-based nodule detection method and device and electronic equipment
CN113313680B (en) Colorectal cancer pathological image prognosis auxiliary prediction method and system
Arjmand et al. Deep learning in liver biopsies using convolutional neural networks
CN112053325A (en) Breast mass image processing and classifying system
Jiang et al. Breast cancer detection and classification in mammogram using a three-stage deep learning framework based on PAA algorithm
JP2012252559A (en) Image processing device, image processing method, recording medium, and program
CN112750137A (en) Liver tumor segmentation method and system based on deep learning
CN116682109B (en) Pathological microscopic image analysis method, device, equipment and storage medium
de Araújo et al. Automated detection of segmental glomerulosclerosis in kidney histopathology
CN111062909A (en) Method and equipment for judging benign and malignant breast tumor
CN116934686A (en) OCT (optical coherence tomography) image detection method and device based on multi-direction image fusion
CN113379691B (en) Breast lesion deep learning segmentation method based on prior guidance
CN115375632A (en) Lung nodule intelligent detection system and method based on CenterNet model

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