阅读说明：本技术 一种基于图像识别的预防医源性肠壁穿孔的方法 (Method for preventing iatrogenic intestinal wall perforation based on image recognition ) 是由 王玉峰 于 2019-04-25 设计创作，主要内容包括：本发明公开一种基于图像识别的预防医源性肠壁穿孔的方法,包括步骤：肠镜视频截图处理,获取训练用图片数据；从图片数据中获取目标图片,形成有效图片数据；利用使用OpenCV大量有效图片数据处理,将图片无效信息区域截取,保留包含肠道背景的图片部分；制作VOC数据集；搭建深度学习框架；使用VOC数据集在深度学习框架中训练,目标检测算法采用YOLOv3算法,最终得到目标检测模型,用来实时识别肠镜手术检查过程中的结肠镜镜头与肠壁距离小于阈值的图像。本发明用于在结肠镜手术检查过程中实时分析肠镜视频,发现视频中距离肠壁过近的图像时,给予提示,避免术者操作过程中由于用力过度导致医源性肠壁穿孔。(The invention discloses a method for preventing iatrogenic intestinal wall perforation based on image identification, which comprises the following steps: performing enteroscope video screenshot processing to obtain picture data for training; acquiring a target picture from the picture data to form effective picture data; intercepting an invalid picture information area by utilizing a large amount of effective picture data processing by using OpenCV, and reserving a picture part containing an intestinal background; making a VOC data set; building a deep learning framework; the VOC data set is used for training in a deep learning framework, the YOLOv3 algorithm is adopted in a target detection algorithm, and finally a target detection model is obtained and used for identifying images with the distance between a colonoscope lens and the intestinal wall smaller than a threshold value in the enteroscopy process in real time. The invention is used for analyzing the enteroscope video in real time in the process of colonoscope operation examination, and gives a prompt when finding the image too close to the intestinal wall in the video, thereby avoiding iatrogenic intestinal wall perforation caused by over exertion in the operation process of an operator.)

1. A method for preventing iatrogenic intestinal wall perforation based on image recognition is characterized by comprising the following steps:

(1) performing enteroscope video screenshot processing to obtain picture data for training;

(2) acquiring a target picture from the picture data to form effective picture data;

(3) Processing a large amount of effective picture data obtained in the step (2) by using OpenCV, intercepting an invalid information area of the picture, and reserving a picture part containing an intestinal background;

(4) making a VOC data set by using the data obtained in the step (3);

(5) building a deep learning framework for training data in the VOC data set;

(6) the VOC data set is used for training in a deep learning framework, the YOLOv3 algorithm is adopted in a target detection algorithm, and finally a target detection model is obtained and used for identifying images with the distance between a colonoscope lens and the intestinal wall smaller than a threshold value in the enteroscopy process in real time.

2. The method according to claim 1, wherein the step (1) is specifically that a Python script is used to process a large number of enteroscopy videos, and every 5 frames of the videos are saved as a JPEG-formatted picture, so as to obtain a large number of pictures at each stage during the enteroscopy operation.

3. The method for preventing iatrogenic intestinal wall perforation based on image recognition according to claim 1, wherein the specific step in the step (2) is to select a picture with a distance between a colonoscope lens and the intestinal wall smaller than a threshold value from a plurality of obtained enteroscope video screenshots, especially a key frame before iatrogenic perforation occurs in the surgical procedure.

4. The method for preventing iatrogenic intestinal wall perforation based on image recognition according to claim 1, wherein the invalid information area of the picture comprises a black border of the picture.

5. The method for preventing iatrogenic intestinal wall perforation based on image recognition according to claim 1, wherein the specific steps in the step (4) are as follows:

and (3) carrying out annotation processing on the picture processed by the OpenCV picture by using an image annotation tool, marking out the intestinal environment when the distance between the colonoscope lens and the intestinal wall in the picture is less than a threshold value by using a rectangular frame, generating an XML format document, and then converting the XML document into a TXT format document by using a Python compiling script.

Technical Field

The invention relates to the technical field of image processing, in particular to a method for preventing iatrogenic intestinal wall perforation based on image identification.

Background

Iatrogenic colon injury of colon perforation caused by enteroscopy rarely occurs clinically, but once the iatrogenic colon injury occurs, bacteria in colon enter abdominal cavity, abdominal cavity pollution is caused, peritonitis and septicemia can be rapidly developed, the consequences are serious, and the risks of complication and death are obviously increased. For diagnostic colonoscopy, the occurrence probability of iatrogenic colon perforation is 0.01% -0.35%; for therapeutic colonoscopy, iatrogenic colon perforation occurs with a probability of 0.029% to 2.14%, with some reports even up to 3%. Moreover, with the attention of people on their health, the increasing clinical application of techniques such as colonoscopy, the excision of polyps under colonoscopy, and the excision of early colon tumors under mucosa, the occurrence of iatrogenic colon injuries is on the rise. And there is ample evidence that the incidence of therapeutic colonic perforations is significantly higher than that of diagnostic colonic perforations.

The most common site of iatrogenic colon perforation is the junction between the rectum and the sigmoid colon, and the most typical clinical feature is the visualization of extra-colonic structures during colonoscopy, but in many cases, perforation is not necessarily realized during colonoscopy, and patients have the manifestations of abdominal pain, elevated body temperature, peritoneal irritation, elevated blood leukocytes, etc. hours after colonoscopy. Therefore, it is highly desirable to prevent perforation of the colon during colonoscopic surgery to ensure substantial safety of the patient.

In recent years, precise medicine and artificial intelligence have become popular topics in the medical and health industries, and medical images have become important research hotspots in the medical field for artificial intelligence, wherein the development of computer vision-based medical image processing and automatic analysis technology is rapid.

Disclosure of Invention

The invention aims to provide a method for preventing iatrogenic intestinal wall perforation based on image identification, which aims at overcoming the technical defects in the prior art.

The technical scheme adopted for realizing the purpose of the invention is as follows:

a method for preventing iatrogenic bowel wall perforation based on image recognition, comprising the steps of:

(1) performing enteroscope video screenshot processing to obtain picture data for training;

(2) acquiring a target picture from the picture data to form effective picture data;

(3) processing a large amount of effective picture data obtained in the step (2) by using OpenCV, intercepting an invalid information area of the picture, and reserving a picture part containing an intestinal background;

(4) Making a VOC data set by using the data obtained in the step (3);

(5) building a deep learning framework for training data in the VOC data set;

(6) the VOC data set is used for training in a deep learning framework, the YOLOv3 algorithm is adopted in a target detection algorithm, and finally a target detection model is obtained and used for identifying images with the distance between a colonoscope lens and the intestinal wall smaller than a threshold value in the enteroscopy process in real time.

Specifically, the step (1) is to use Python to write a script to process a large number of enteroscopy videos, and store every 5 frames of the videos as a picture with a JPEG format, so as to obtain a large number of pictures at each stage in the enteroscopy operation process.

The specific step in the step (2) is to select a picture with the distance between the colonoscope lens and the intestinal wall being smaller than a threshold value from a large number of obtained enteroscope video screenshots, especially a key frame before a iatrogenic perforation phenomenon occurs in the operation process.

The invalid information area of the picture comprises a black border of the picture.

The specific steps in the step (4) are as follows:

and (3) carrying out annotation processing on the picture processed by the OpenCV picture by using an image annotation tool, marking out the intestinal environment when the distance between the colonoscope lens and the intestinal wall in the picture is less than a threshold value by using a rectangular frame, generating an XML format document, and then converting the XML document into a TXT format document by using a Python compiling script.

Compared with the prior art, the invention has the beneficial effects that:

according to the invention, a YOLOv3 target detection algorithm is used for constructing the identification model for the short distance between the colonoscope lens and the intestinal wall, the model detection speed is high, the model detection method can be used for processing an enteroscope video, and the operation condition that iatrogenic intestinal wall perforation is likely to happen is found in real time. The image recognition method is used for assisting detection, can assist a doctor to finish examination in clinical colonoscopy, and avoids iatrogenic intestinal wall perforation caused by excessive force application of an operator in the enteroscopy operation process as much as possible.

The method for preventing iatrogenic intestinal wall perforation based on image identification is used for analyzing an enteroscope video in real time in the process of colonoscopy operation examination, and when an image which is too close to the intestinal wall in the video is found, a prompt can be given, so that iatrogenic intestinal wall perforation caused by over-exertion of force in the operation process of an operator is avoided.

Drawings

Fig. 1 is a flow chart of a method for preventing iatrogenic bowel wall perforation based on image recognition.

Detailed Description

The invention is described in further detail below with reference to the figures and specific examples. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

As shown in fig. 1, the method for preventing iatrogenic intestinal wall perforation based on image recognition of the present invention comprises:

step 101, enteroscopy video screenshot processing. And processing a large amount of enteroscopy videos by using a Python writing script, and storing every 5 frames of the videos into a picture with a JPEG format, thereby obtaining a large amount of pictures at each stage in the enteroscopy operation process.

And 102, sorting the valid data. From the large number of enteroscopy video screenshots obtained in step 101, a picture of the colonoscope when the colonoscope lens is too close to the intestinal wall, especially a key frame before the iatrogenic perforation occurs during the operation, is selected.

And step 103, processing the OpenCV picture. And processing a large number of effective pictures obtained in the step 102 by using OpenCV, intercepting invalid information areas such as black borders in the pictures, and reserving picture parts containing intestinal backgrounds.

And 104, preparing a VOC data set.

The original pictures were labeled using Labelimg. Marking out the intestinal environment when the distance between the colonoscope lens in the picture and the intestinal wall is smaller than a threshold value by using a rectangular frame, generating an XML format document, and then using Python to write a script to convert the XML document into a TXT format document.

In this embodiment, the image annotation tool may be software that can be used for creating a target detection task data set, such as Labelimg, Labelme, vantic, and Sloth, which is not limited in this embodiment.

And 105, building a deep learning framework.

In this embodiment, the deep learning frame may be Darknet, tensflow, Caffe, or the like, which is not limited in this embodiment, and the processed image data may be trained by using the deep learning frame.

Step 106, training the data acquisition model.

The VOC data set acquired in step 104 is trained in the deep learning framework set up in step 105, and the target detection algorithm adopts YOLOv3 algorithm. And finally, obtaining a target detection model too close to the intestinal wall, and identifying the image when the lens is too close to the intestinal wall in the enteroscopy process in real time.

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.