Brain MRI image classification method and device based on three-dimensional convolutional neural network

Abstract

The invention provides a classification method based on a three-dimensional convolution neural network, which is applied to brain MRI images. On the basis of a main network, an auxiliary supervision branch network is designed to supervise and learn a middle layer, and finally the final classification result is acquired by combining the main network with the branch network. The method can makes full use of the three-dimensional convolution neural network to extract the important three-dimensional information of an image, and uses the auxiliary supervision branch network to extract more robust local information of the image to make up for the deficiency of a two-dimensional convolutional neural network in the aspect of three-dimensional feature extraction. The middle layer is supervised andlearnt, so that the network extracts features with a distinguishing ability as early as possible in the learning process. The learning speed is very fast, which has an important influence on the finalclassification result. The auxiliary supervision convolution neural network is added, which can improve the accuracy and robustness of brain MRI image classification, and accelerates the convergenceof the learning process.

Description

technical field [0001] The invention relates to a brain MRI image classification method and device based on a three-dimensional convolutional neural network. Background technique [0002] Brain tumors are one of the most common and deadly diseases in the world. Doctors divide tumors into two categories, malignant and benign, according to their pathological form, growth pattern, and degree of harm to patients. Analysis of images of tumors can help doctors assess how the disease is progressing to suggest and alter treatment options. Nuclear magnetic resonance technology is a non-invasive medical imaging technology. By analyzing MRI image sequences, we can obtain high-resolution 3D images with anatomical and functional information, which is conducive to improving the level of diagnosis and treatment of diseases. . In recent years, machine learning methods based on supervised learning have been increasingly used in the classification of MRI images, and have achieved good recog...

AI Technical Summary

Problems solved by technology

Convolutional neural networks identify specific disease types based on complex hierarchical feature representations that are difficult for humans or traditional classification

[0005] However, the above medical images are all two-dimensional images, and they all use two-dimensional convolutional neural networks to achieve classification results. Since MRI images are 3D images, there are deficiencies in directly using traditional two-dimensional convolutional neural networks to classify images: Two The three-dimensional network structure cannot use the three-dimensional spatial information of the image, and will lose a lot of useful information, thus limiting the classification performance of the model

In addition, the convolutional neural network is a multi-layer learning network. The traditional method only supervises the last layer of the network, ignoring the influence of the middle layer supervision on the classification effect of the model.

Images