Title: AutoIQ: An Ensemble Framework for Automatic Assessment of Geometric Distortion in Prostate Diffusion-Weighted Imaging

Abstract:

Geometric artifacts in prostate diffusion-weighted imaging (DWI) can compromise the precision of lesion localization and diminish the dependability of MRI-driven clinical evaluations. To address this, we introduce AutoIQ, a novel ensemble machine learning approach designed to automatically quantify and classify the severity of geometric distortion in DWI data. Our study retrospectively evaluated 140 prostate biparametric MRI examinations. Among these, 33 cases exhibited severe distortion necessitating a repeat scan, while 107 cases were deemed to have acceptable distortion levels according to expert radiologist judgment.

AutoIQ integrates two distinct strategies for quantifying distortion. The first is a segmentation-based technique that assesses the mismatch of prostate boundaries between T2-weighted imaging (T2WI) and DWI. The second is a registration-based approach that calculates the magnitude of deformation following the alignment of DWI with T2WI. These calculated distortion scores served as inputs for training both individual classifiers and a logistic-regression ensemble model.

Statistical analysis revealed that both computational methods effectively distinguished between severe and acceptable distortion cases (p < 0.001). When tested on an independent dataset, the ensemble model demonstrated superior performance compared to individual models, achieving an accuracy of 0.95, an F1-score of 0.93, and an Area Under the Curve (AUC) of 0.98. These findings indicate that AutoIQ offers a robust, automated, and quantitative method for assessing the quality of prostate DWI, potentially aiding in the identification of scans that need to be reacquired.

Source: arXiv Generated at: 2026-06-02 00:00:00 UTC