Pathological brain image classification method based on deep stacked sparse autoencoder

Abstract

The invention discloses a pathological brain image classification method based on a deep stacked sparse autoencoder, comprising the following steps of: 1. downloading different types of pathological brain magnetic resonance images, including normal brain images and pathological brain images, from the website of Harvard medical college; 2, using a data enhancement method to increase the number of images in order to balance the data distribution; 3, using the method of deep stacked sparse autoencoder to automatically extract the characteristics of the input image, accurately classifying the image by a Softmax classifier to otbain different disease types 4, training the entire network by a batch conjugate gradient method, and in a fine-adjusting stage, using a scaled conjugate gradient methodto fine adjust the network. The method avoids the loss of some characteristic information. Compared with the traditional method just capable of achieving two classifications of the diseased brain images, the method can accurately obtain the effective classifications of different types of diseases, significantly reduces the doctors' workload, and has a practical application value.

Description

technical field [0001] The present invention relates to a classification method for different pathological brain diseases, in particular to a method for classifying pathological brain images based on a deep stacked sparse autoencoder. Background technique [0002] Due to the rapid development of computer technology, magnetic resonance imaging technology has been widely used in the medical field, especially in the detection of pathological brain diseases, but how to obtain the classification results of brain diseases with high precision is still our key research direction. However, due to the high cost of acquiring MRI brain image data, the number of images used in a large number of scientific studies is relatively small, resulting in poor generalization performance. [0003] In the research on traditional disease brain image classification methods, we found the following problems: First, in traditional image classification problems, image features cannot be extracted automat...

AI Technical Summary

Problems solved by technology

However, due to the high cost of acquiring MRI brain image data, the number of images used in a large number of scientific studies is relatively small, resulting in poor generalization performance

[0003] In the research on traditional disease brain image classification methods, we found the following problems: First, in traditional image classification problems, image features cannot be extracted automatically, but need to be manually extracted, which will lead to partial loss of image information; secondly, In traditional disease brain image classification research, because it is quite difficult to obtain various disease brain images, the number of images used by researchers is small; finally, when using traditional disease brain image classification methods for image classification, only binary classification can be achieved, that is, Can only be divided into normal brain and abnormal brain

Images