Organ detection and identification positioning method based on deep learning

Abstract

The invention belongs to the field of medicine, and relates to an organ detection and identification positioning method based on deep learning. Through deep learning, feature extraction and feature region prediction of organs are realized, a basic network part of a training network is utilized to perform feature extraction on an input image, an auxiliary network part is utilized to predict a convolution feature layer by using a convolution kernel, a fixed prediction set is generated, and multi-scale prediction aiming at different convolution feature layers is realized. For the generated prediction set, an optimal bounding box is found through screening, a redundant bounding box is removed, a window with the highest score in bounding box neighborhoods is selected, and the window with the low score is inhibited. A screening result is output, organ detection and identification positioning in the acquired image is completed through text information and rectangular frame representation, soas to realize identification and positioning of a complex key area, and obtain position information and category information of a key organ part.

Description

technical field [0001] The present invention relates to an organ detection, identification and positioning method based on deep learning, more specifically, the present invention relates to a deep learning-based organ detection, identification and positioning method in the medical field. Background technique [0002] In the medical operation process, although after long-term training, experienced doctors are already familiar with the structure of organs in various parts of the human body, but in the highly tense operation process, doctors often need to perform different operations with their left and right hands. When concentrating on a certain operation, the instrument in the other hand is very likely to accidentally slip and touch the "red line" part. The "red line" part is an organ part that must not be touched in the human body, such as heart surgery. Touching the Koch triangle can cause accidental bleeding at least, and threaten life at worst. Moreover, in the traditio...

AI Technical Summary

Problems solved by technology

Moreover, in the traditional operation process, because the doctor cannot accurately locate the location of the organ, and cannot accurately divide the size of the operable space between adjacent organs, the doctor needs to try many times to determine the surgical tool with the right size, which not only increases the operation time but also increased risk of surgery

For traditional image processing and recognition methods, due to the high similarity in color and texture of images of key parts such as organs collected by current medical image acquisition devices, it is difficult to automatically obtain position information and category information of different organ parts using traditional methods

Images