Multi-modal medical image coding and generating method based on generative adversarial network

Abstract

The invention discloses a multi-modal medical image coding and generating method based on a generative adversarial network. The method comprises the following steps: preprocessing data and obtaining a training data set, simulating a scene of clinical MRI sequence deletion, constructing a generative adversarial neural network, and training the generative adversarial neural network; according to the method, one network is trained at a time, feature extraction of multiple MRI sequences is achieved at the same time, compared with an existing model, the quality of translated images is higher, the capacity of flexibly translating the input MRI images into any needed MRI sequences is achieved, and the method is suitable for being applied to various MRI sequences. According to the method, feature coding of MRI images of various modes and mutual generation among MRI multi-mode images can be solved, an unsupervised training mode is adopted, a generation network can still be trained under the condition that training data modes are missing, clinical scenes are better met, and image generation among any multiple modes can be achieved.

Description

technical field [0001] The invention relates to the technical field of image processing, in particular to a method for encoding and generating multimodal medical images based on confrontation generation networks. Background technique [0002] Magnetic resonance imaging (MRI) can acquire a variety of different sequences (eg, T1-weighted, T2-weighted, T1-enhanced contrast (T1c), T2 fluid-attenuated inversion recovery (T2FLAIR), etc.), each of which can provide a different view of tissue contrast and spatial resolution, the combination of sequences provides more complementary information, enabling doctors to make more accurate diagnoses and treatments, and many times specific sequences work best for specific disease diagnoses, for example, for glioblastoma In terms of tumor diagnosis, T1 and T2 FLAIR sequences can clearly show the edema area of ​​the tumor, and T1c sequence can clearly show the enhancement area around the tumor. [0003] However, due to inter-institutional dif...

AI Technical Summary

Problems solved by technology

[0003] However, due to inter-institutional differences in acquisition protocols, time constraints, and the presence of image artifacts, it is often difficult to obtain enough MRI sequences for patients, resulting in frequent missing sequences in MRI examinations. Variability among personnel, machines, patient conditions, and other factors limits cross-center comparisons and studies of MRI sequences

[0004] The above problems not only interfere with the diagnosis of doctors, but also hinder the analysis of many downstream data, which usually assume the existence of a specific set of pulse sequences to perform their tasks, although deep learning-based brain tumor segmentation with clear clinical needs has achieved Although considerable progress has been made, translating state-of-the-art computational methods into tools for routine use in the clinic remains a major challenge, as most methods rely on specific sequences being fed into the model,

Images