Learning to Distort: Weakly-Supervised Image Quality Transfer for Prostate DWI Correction

2026-06-17

AI Summary

This study addresses the challenge of geometric distortion in prostate diffusion-weighted imaging (DWI), which compromises diagnostic reliability, by proposing a weakly supervised image quality translation framework. Lacking access to pixel-aligned distorted and undistorted clinical images, the method leverages only image-level quality labels (distorted vs. undistorted) to construct quality prototypes in a pretrained feature space. It then combines prototype-based flow matching with generative modeling to synthesize realistic distorted images paired with their undistorted counterparts, enabling the training of a forward correction model. By integrating image quality assessment signals into weakly supervised learning, the approach avoids reliance on pixel-wise correspondences or purely unpaired strategies. Experiments demonstrate that the generated distortions are highly realistic and that the corrected images significantly improve performance on downstream clinical tasks, including PI-RADS assessment and Gleason scoring, outperforming state-of-the-art unpaired methods such as CycleGAN, UNIT-DDPM, and OT-FM.

Abstract

Single-shot echo-planar prostate diffusion-weighted imaging (DWI) is frequently complicated by geometric distortions, which impact the ability to derive reliable diagnoses from such images. Developing automated correction methods is challenged by the absence of paired distorted and undistorted clinical scans. In this paper, we first propose a novel weakly-supervised image quality transfer (IQT) framework from undistorted to distorted images that utilizes image quality assessment (IQA) signals to supervise the transfer process. Unlike traditional methods that require expensive, voxel-wise paired data or resort to developing unpaired algorithms, our approach utilizes image-level quality labels (here, distorted vs. undistorted) to establish latent quality prototypes within a pre-trained feature space. Recognizing that simulating realistic distortions is more reliable than direct unpaired correction, we describe a weakly-supervised prototype flow matching algorithm to explicitly regularize generative trajectories towards distorted prototypes, producing realistic susceptibility artifacts that mimic clinical degradations. By synthesizing these realistic pairs, we enable a second IQT model to be trained in the forward direction for distortion correction. Experimental results demonstrate that our generated images successfully mimic the diagnostic interference of real-world artifacts, which leads to more capable distortion correction IQT models. In addition to qualitative comparisons, we also conduct exhaustive quantitative evaluations that compare our approach with existing unpaired approaches (e.g., CycleGAN, UNIT-DDPM, and OT-FM) - as either forward or reverse alternatives - by assessing clinical downstream task performance in PI-RADS and Gleason score classification, using both in-distribution and external data sets.

Problem

• geometric distortion
• prostate DWI
• image quality transfer
• weakly-supervised learning
• distortion correction

Innovation

• weakly-supervised learning
• image quality transfer
• prototype flow matching
• distortion correction
• diffusion-weighted imaging