Medical image data analysis method and system based on fused deep tensor nerve network

Abstract

The invention belongs to the technical field of computer application, specifically relates to the technical field of medical image analysis and specifically relates to a medical image data analysis method and system based on a fused deep tensor nerve network. The method is characterized by, to begin with, integrating effective information in medical images through a tensor convolutional neural network, and then, entering a tensor recursive neural network; and by carrying out analysis on historical medical image data and current medical image data of a patient, outputting an analysis result ofthe current medical image data and analysis result prediction of medical images in the future for providing analysis reference for doctors and evaluating a treatment scheme received by the patient respectively. Compared with a conventional recursive neural network, the network in the invention introduces tensor CP decomposition and tensor column decomposition, and the parameter scale of the tensorrecursive neural network is far smaller than the parameter scale of the conventional network in processing the same tensor data; and therefore, the medical image data analysis method and system can effectively improve reliability and efficiency of image analysis, and provide basis for adjustment and optimization of the treatment schemes.

Description

technical field [0001] The invention belongs to the technical field of computer applications, in particular to the technical field of medical image analysis, in particular to a medical image data analysis method and system fused with a deep tensor neural network. Background technique [0002] For example, the name of the invention is the method of deep learning to analyze medical data and its intelligent analyzer, and the application number is 201510294286.3. The Chinese invention patent applies the deep learning model to analyze the medical data and outputs the corresponding pathological analysis results. Its concrete implementation scheme is as follows: [0003] Collect the medical material data of the same type that has been filed and the medical analysis data that matches the medical material data as medical training data and store them in the computer through the input device; [0004] The not less than two-dimensional medical image data in the medical training data an...

AI Technical Summary

Problems solved by technology

The convolutional neural network cannot extract the information of the time dimension, and the information of the time dimension exists, which turns into noise

[0012] And through the existing technology, it can be seen that the recurrent neural network is a neural network that can display and model time series data, while medical images are generally stored and expressed in 2D and 3D matrix or tensor models, with a large scale, such as MRI images each The slice field of view is 512×512. In traditional recurrent neural network analysis, vectorization processing is used to perform feature extraction or other vectorization processing on medical images, which will destroy the unique spatial structure information of medical images and the relationship between various factors to varying degrees. Intrinsic dependencies cannot make full use of all effective information of medical images, and the reliability of model output results is low. At the same time, there are a large number of fully connected structures in traditional recurrent neural networks, such as fully connected from the input layer to the first hidden layer, and fully connected between hidden layers. , The hidden layer itself is fully connected, and the hidden layer is fully connected to the output layer. These fully connected structures generate huge weight parameters, resulting in high memory costs, high computational complexity, and low efficiency.

Images