Melanin cancer image segmentation method and network

Abstract

The invention discloses a segmentation method and network for a melanin cancer image. The method comprises the following steps that (1) image preprocessing is conducted on a dermatoscopy image; (2) a lightweight skin cancer segmentation network for melanin cancer image segmentation is constructed, and the network adopts the structure of an encoder decoder, and the method comprises the following steps that (2.1) the encoder ensures information exchange between channels and extracts more useful feature information while reducing network parameters by using a deep separable convolution and channel random fusion means; (2.2) a decoder adopts an attention mechanism and combines a new feature fusion mode to ensure that more relevant feature information in a low-level stage and feature information in a high-level stage are fused so as to improve the segmentation accuracy; and (2.3) a score is calculated through a global average pooling layer to obtain a segmentation result.

Description

technical field [0001] The present invention relates to the technical field of medical image processing, in particular to a method for segmenting melanoma images, and a segmentation network for melanoma images, which provide semantic segmentation results for input skin cancer images and provide automatic monitoring systems for melanoma Reliable pending data. Background technique [0002] Melanoma is a tumor derived from malignant transformation of melanocytes. It is highly malignant and mostly occurs in the skin. Although the morbidity and mortality of melanoma are not so high, they are increasing year by year at a growth rate of 6% to 7%. According to the report of the American Oncology Annual Meeting (ASCO), it has become one of the fastest growing tumors. Especially in my country, the incidence of melanin is rapidly increasing, with about 20,000 new cases each year, and the mortality rate is also increasing rapidly year by year. Clinical findings show that the earlier t...

AI Technical Summary

Problems solved by technology

However, for this problem, the post-processing operation of CRF is relatively cumbersome, which leads to relatively slow training and difficult training.

Moreover, with the development of technology, the tools for obtaining test images are constantly being updated, and the current image preprocessing scheme may have application limitations, which may affect the accuracy of segmentation results

Images