A Convolutional Neural Network Based Lesion Image Classification and Segmentation Method

Abstract

The invention relates to a classification and segmentation method of lesion images based on convolutional neural networks, which specifically includes the following steps: (1) collecting standard white light images of patients, and labeling the collected white light images with categories based on strict histological evidence. Segment and annotate it and use it as an image database; (2) Construct the lesion classification network Dual-stream ELNet to obtain the lesion classification model; (3) Construct the lesion U-Net segmentation network to obtain the lesion segmentation network model; (4) White light the lesion to be tested The image is input into the lesion classification network Dual‑stream ELNet to obtain the category of the lesion; the white light image of the lesion of the said category is input into the specified U‑Net segmentation network model to obtain the lesion segmentation result. The lesion classification network Dual‑stream ELNet extracts global and local features based on the Global Stream and Local Stream models, effectively improving the final classification results.

Description

technical field [0001] The invention relates to a lesion classification and segmentation method, in particular to a convolutional neural network-based lesion classification and segmentation method. Background technique [0002] In recent years, with the development of science and technology, endoscopic technology has been widely used clinically, which can achieve the purpose of observing the internal organs of the human body with the least damage. However, each endoscopy will generate a large number of data images. In order to detect lesion images, doctors need to spend a lot of time viewing images, and at the same time, missed and false detections may occur due to visual fatigue. Therefore, developing a set of methods for automatic detection of endoscopic lesion images is a key problem that needs to be solved urgently. At present, in the field of automatic detection of endoscopic lesion images, many researchers have adopted traditional machine learning methods, and tradit...

AI Technical Summary

Problems solved by technology

[0003] The technical problem to be solved by the present invention is to provide a classification and segmentation method of lesion images based on convolutional neural network to solve the problems in the prior art that manual extraction of image features is required, resulting in cumbersome steps and unsatisfactory results.

Images