An Ensemble Framework for Automatic Quantification of Image Distortion in Prostate Diffusion-Weighted Imaging

Motivation: Diffusion-weighted imaging(DWI) is essential for prostate cancer imaging; however, its susceptibility to image distortion presents challenges in radiology interpretation. Automatic quantification of the level of image distortion could provide immediate feedback to scan operators, potentially reducing patient recalls. Goal(s): This study aimed to develop an ensemble machine-learning framework for the automatic assessment of distortion in prostate DWI. Approach: The framework integrates distortion factors derived from both segmentation-based and registration-based computational methods, employing ensemble learning to improve classification accuracy. Results: The proposed ensemble model demonstrated superior performance in accurately classifying distortion ranks, achieving AUCs of 1.0, 0.93, 0.93 in distinguishing the three distortion levels. Impact: The developed ensemble framework for automatic assessment of image distortion may assist in acquiring high-quality prostate DWI and reducing patient recalls.