阅读说明：本技术 基于深度学习的瘤样病变识别工作站 (Tumor-like lesion recognition workstation based on deep learning ) 是由 王玉峰 于 2019-04-25 设计创作，主要内容包括：本发明涉及智能医疗技术领域,尤其涉及一种基于深度学习的瘤样病变识别工作站。包括：肠镜设备、电脑主机、显示设备、WEB端界面、病变识别模型和数据传输模块；所述肠镜设备为医生肠镜检查所需要的所有相关仪器和设备；所述电脑主机用于配置好模型所需的相关环境框架；所述web端界面架设在电脑主机中；所述病变识别模型用于识别肠镜检查退镜视野中的相关病变；所述显示设备和电脑主机相连。本发明能够通过深度学习自动检查肠道中的瘤样病变并提示给医生供参考,不仅能更好更全面的检测出肠道瘤样病变,而且可以大大减轻医生的压力,让医生更专注于其它更富创造性的任务中,有着巨大的经济及社会效益。(The invention relates to the technical field of intelligent medical treatment, in particular to a tumor-like lesion identification workstation based on deep learning. The method comprises the following steps: the enteroscope equipment, the computer host, the display equipment, the WEB terminal interface, the lesion identification model and the data transmission module; the enteroscopy equipment is all related instruments and equipment required by enteroscopy of doctors; the computer host is used for configuring a relevant environment frame required by the model; the web end interface is erected in the computer host; the lesion identification model is used for identifying related lesions in a retroscopic view of enteroscopy; the display equipment is connected with the computer host. The invention can automatically check the tumor-like lesion in the intestinal tract through deep learning and prompt a doctor for reference, thereby not only detecting the tumor-like lesion in the intestinal tract better and more comprehensively, but also greatly relieving the pressure of the doctor, leading the doctor to concentrate on other more creative tasks and having great economic and social benefits.)

1. The utility model provides a tumour appearance pathological change discernment workstation based on degree of depth study which characterized in that: the method comprises the following steps: the enteroscope equipment, the computer host, the display equipment, the WEB terminal interface, the lesion identification model and the data transmission module;

the enteroscopy equipment is all related instruments and equipment required by enteroscopy of doctors; the computer host adopts a linux-Ubuntu16.04 operating system, and configures a relevant environment frame required by the model; the web end interface is erected in the computer host and is used for displaying a model identification result; the lesion identification model is used for identifying related lesions in the endoscope withdrawal field of enteroscopy and transmitting the lesions to the computer host through the data transmission module; the display equipment is connected with the computer host and is used for displaying the enteroscope video image, the lesion mark and the web end interface transmitted by the enteroscope camera.

2. The deep learning based neoplasia identification workstation of claim 1, wherein: the display device comprises a high-definition display and is used for displaying an enteroscope video image transmitted by an enteroscope camera, a lesion mark and a web-end interface.

3. The deep learning based neoplasia identification workstation of claim 1, wherein: the web-side interface can display relevant information of hospital departments, doctors and patients, a prompt box for examination operation, a real-time image of scene examination and output of an identification model.

4. The deep learning based neoplasia identification workstation of claim 1, wherein: the lesion recognition model algorithms include, but are not limited to YOLOV 3; the lesion recognition model can receive video images output by the enteroscope, each frame of image is sent to the model for target detection, if a target is detected, the type of the target is output, the position is framed, and the target is sent to the web end for display through the data transmission module.

Technical Field

The invention relates to the technical field of intelligent medical treatment, in particular to a tumor-like lesion identification workstation based on deep learning.

Background

Colorectal cancer is one of the most common malignancies worldwide, including colon and rectal cancers. The incidence of large intestine cancer is, from high to low, rectum, sigmoid colon, cecum, ascending colon, descending colon and transverse colon. With the aggravation of the aging of the population in China and the change of the life style of people, the incidence and mortality of colorectal cancer are rising year by year, and early discovery, early diagnosis and early treatment are one of the main strategies for reducing the mortality and improving the survival rate.

The enteroscopy is a method of inserting the enteroscopy circulation cavity into the ileocecal part through the anus and observing the colon lesion from the side of the mucous membrane. Enteroscopy can meet the needs of examination in almost all colon areas. The enteroscopy can be used for diagnosis, can be used for removing polyps or early-stage tiny cancer focuses, and can be used for biopsy of tissues by directional microscopy on the focuses. In the general investigation of colorectal cancer, it is often used as a "gold standard" for evaluating various preliminary screening effects. Has important significance for preventing colon cancer from being discovered early, so the method is the most effective means for diagnosing the colon cancer at present.

However, enteroscopy requires the human eye to identify the intestinal condition in a dynamic situation, and in addition, each doctor has many surgical examinations every day, and long-term visual fatigue is likely to cause some small and unobvious polyps to be missed.

Disclosure of Invention

The invention aims to overcome the defects of the technology and provide a tumor-like lesion identification workstation based on deep learning.

In order to achieve the purpose, the invention adopts the following technical scheme: the utility model provides a tumour appearance pathological change discernment workstation based on degree of depth study which characterized in that: the method comprises the following steps: the enteroscope equipment, the computer host, the display equipment, the WEB terminal interface, the lesion identification model and the data transmission module;

the enteroscopy equipment is all related instruments and equipment required by enteroscopy of doctors; the computer host adopts a linux-Ubuntu16.04 operating system, and configures a relevant environment frame required by the model; the web end interface is erected in the computer host and is used for displaying a model identification result; the lesion identification model is used for identifying related lesions in the endoscope withdrawal field of enteroscopy and transmitting the lesions to the computer host through the data transmission module; the display equipment is connected with the computer host and is used for displaying the enteroscope video image, the lesion mark and the web end interface transmitted by the enteroscope camera.

Preferably, the display device includes, but is not limited to, a high-definition display for displaying the enteroscope video image transmitted by the enteroscope camera, the lesion mark and the web-side interface.

Preferably, the web-end interface can display prompt boxes for hospital departments, doctors, relevant information of patients, prompt of examination operation, real-time images of scene examination and output of recognition models.

Preferably, the lesion recognition model algorithm includes, but is not limited to, YOLO V3; the lesion recognition model can receive video images output by the enteroscope, each frame of image is sent to the model for target detection, if a target is detected, the type of the target is output, the position is framed, and the target is sent to the web end for display through the data transmission module.

The invention has the advantages that the invention can automatically check the tumor-like lesion in the intestinal tract through deep learning and prompt the physician for reference, thereby not only detecting the tumor-like lesion in the intestinal tract better and more comprehensively, but also greatly reducing the pressure of the physician, enabling the physician to concentrate on other more creative tasks, and having great economic and social benefits.

Drawings

FIG. 1 is a schematic structural view of the present invention;

fig. 2 is a flowchart of the training of the lesion recognition model in the present invention.

Detailed Description

The following detailed description of the preferred embodiments will be made with reference to the accompanying drawings. As shown in fig. 1 and 2, a tumor-like lesion recognition workstation based on deep learning includes an enteroscope device, a computer host, a display device, a WEB interface, a lesion recognition model, and a data transmission module.

Including, but not limited to, light source hosts, colonoscopes, various examination and treatment accessories, suction and monitoring devices, and the like.

The computer host should be capable of running a web-side interface and recognition model.

The display device comprises a high-definition display and is used for displaying an enteroscope video image transmitted by an enteroscope camera, a lesion mark and a web-end interface.

The web-side interface can display relevant information of departments, doctors and patients of the hospital, a prompt of examination operation, a real-time image of scene examination and a prompt box output by a recognition model.

The lesion recognition model algorithms include, but are not limited to, YOLO V3, and the like; the lesion recognition model can receive video images output by the enteroscope, each frame of image is sent to the model for target detection, if a target is detected, the type of the target is output, the position is framed, and the target is sent to the web end for display through the data transmission module.

The process of building the lesion identification model is shown in fig. 2:

step 201, collecting an image. And acquiring enough screenshots of the intestinal lesions in the endoscope withdrawing process, wherein the screenshots of the intestinal conditions are derived from standard images intercepted in the operation process of a professional doctor.

Step 202, image preprocessing. Selecting qualified pictures, namely, pictures with determined lesion, clear images and proper angles, cutting off useless information, wherein the resolution of the processed pictures is not limited to 640 x 480, and finally naming the pictures in sequence by 6-bit integers through a script.

And step 203, labeling the data. Labeled software including but not limited to labelimg, marked the lesion location with a rectangular box by the software, generating a corresponding x.

And step 204, building an environment framework. Downloading and installing an Ubuntu16.04 system, python2.7, cuda8.0, cudnn6.0.21, opencv3.4.0 and Darknet on a computer host; the frameworks used include, but are not limited to, Darknet, Kreas, Tensorflow, and the like; the languages used include, but are not limited to, Python, C + +, C, and the like.

Step 205, a data set is established. The data set comprises pictures required by training and corresponding markup files, and the original x.xml file is converted into a x.txt file through a script and is placed into a frame.

Step 206, modify the training parameters. Parameters to be modified include, but are not limited to, training test switches, learning rate, step size, iteration number, etc., and the specific values are determined according to actual conditions.

Step 207, training the model. And calling a command to transfer a data set related file for training.

The data transmission module acquires a video image transmitted by the enteroscope equipment through a video acquisition card and is connected with the web-end interface and the lesion recognition model through WebSocket communication.

The foregoing is only a preferred embodiment of the present invention, and it should be noted that, for those skilled in the art, various modifications and decorations can be made without departing from the principle of the present invention, and these modifications and decorations should also be regarded as the protection scope of the present invention.