Research On Multi-feature Fusion Classification Algorithms For Medical Image Based On Deep Learning

Hepatocellular carcinoma is the third largest cancer in the world and has a serious impact on people’s health.With the rapid development of medical imaging technology,computer-assisted technology is of great significance to the classification and diagnosis of medical images.In this paper,the magnetic resonance images of HCC patients from a hospital in Guangzhou are used as research data.Based on the overall design of medical image multi-feature fusion classification algorithm,magnetic resonance image preprocessing and feature extraction,completed the design and implementation of high-low level feature fusion classification algorithm and multi-feature fusion classification algorithm based on deep learning.The main innovation of this paper is to present a multi-feature fusion classification algorithm based on deep learning.By designing a feature cascaded fusion layer network,the deep features of multi-period and multi-view HCC tumors are combined,combined with anti-overfitting Dropout technology and global averaging pooling technology,has achieved higher classification accuracy and better stability in the classification of benign and malignant HCC tumors;in addition,this paper also presents multi-kernel learning based high-low level feature fusion classification algorithm,which is superior to the traditional low-level Classification results of texture features and high-level deep features.The main work of this article is as follows:(1)Completed the overall design of a multi-feature fusion classification algorithm for medical images.Based on the analysis of the image fusion algorithm and the characteristics of HCC magnetic resonance imaging,the process design of high-low level feature fusion classification algorithm and the multi-feature fusion classification algorithm based on deep learning are completed.(2)Completed three-dimensional resampling and feature extraction of magnetic resonance images.Three-dimensional resampling of MR images was achieved,including acquisition of clinical magnetic resonance images,extraction of ROIs from HCC tumors,and implementation of three-dimensional resampling.On the basis of three-dimensional resampling,feature extraction of sample images was completed,including texture features and deep feature extraction.(3)Completed the design and implementation of high-low level feature fusion classification algorithm.On the basis of analyzing multi-kernel learning feature fusion method and SVM classifier method,the feature fusion classification algorithm of highlevel deep features and low-level texture features was presented.Based on the preparation of experimental data,the high and low-level feature fusion was designed.The classification algorithm experiment completed the classification performance analysis of texture features and deep features and the comparison analysis of high and low-level feature fusion classification algorithm experiments.(4)Completed the design and implementation of multi-feature fusion classification algorithm based on deep learning.A multi-feature fusion network structure and a cascaded fusion layer network based on deep learning are presented.Based on the Dropout and GAP technologies implemented in the multi-feature fusion deep network,a multi-feature fusion classification algorithm experiment is designed to complete Dropout and GAP.The analysis of the experimental results of the technology and the comparison analysis of the multi-feature fusion classification algorithm experiments based on deep learning.