A prostate magnetic resonance T2 weighted image quality standardization enhancement device and method

Abstract

The application is a prostate magnetic resonance T2 weighted image quality standardization enhancement device and method, which comprises an image acquisition module, a pretreatment module, a model acquisition module, an image enhancement module and a quality evaluation module; the core is a PST-CyGAN network, which integrates a spatial adaptive normalization module SPADE, a pyramid spatial adaptive normalization module Pyramid-SPADE and a Transformer module in the generator on the basis of the CycleGAN framework, which are respectively used for anatomical perception spatial alignment, multi-scale feature fusion and global context modeling, realizing non-paired unsupervised domain conversion of cross-center prostate T2 weighted images from standard resolution to high resolution; the network training adopts a quality stratification strategy based on PI-QUAL score, which is high in reliability; a quality trigger deployment strategy based on PI-QUAL score is proposed, and intelligent allocation of computing resources is realized; the application is significantly superior to existing comparative methods, has strong generalization ability and near real-time inference efficiency, effectively reduces the quality difference of cross-institutional images, and supports consistent application of PI-RADS score.