An application method of a multi-class segmentation loss function in realizing multi-class segmentation of spinal tissue images

Abstract

A multi-class segmentation loss function and its construction method and application relate to the fields of deep learning and medical image processing, and solve the problems of imbalance between classes and poor segmentation accuracy in the prior art. The invention includes: setting a basic loss function; setting a limit weight according to the basic loss function, which is used to adjust the loss function weight corresponding to the segmentation target; combining the basic loss function and the limit weight to obtain multiple types of segmentation loss functions. The present invention has the advantages of significantly suppressing the influence of the imbalance between classes on the multi-class segmentation of the spinal tissue image, improving the segmentation performance of the full convolutional neural network, improving the spinal tissue image segmentation effect, and enhancing the segmentation performance of the loss function. Deep learning and medical image processing technology have extremely high application and promotion value.

Description

technical field [0001] The invention relates to the technical field of deep learning and medical image processing, in particular to a multi-class segmentation loss function and a construction method and application thereof. Background technique [0002] Diseases such as spinal lesions often need to be diagnosed with the help of medical images, and the development of artificial intelligence technology has made computer-aided diagnosis of spinal diseases and other diseases based on algorithms such as deep learning more and more researchers pay attention. Among them, accurate tissue region segmentation is an important prerequisite for realizing its intelligent diagnosis. The current common deep learning segmentation technology performs pixel-level convolution and backpropagation by inputting training images and label files, and finally obtains the classification probability value of each pixel to achieve the task of semantic segmentation. As an important part of backpropagatio...

AI Technical Summary

Problems solved by technology

[0005] The purpose of the present invention is to provide a multi-class segmentation loss function and its construction method and application to solve the problems of imbalance between classes and poor segmentation accuracy in the multi-class segmentation of existing spinal tissue images

Images