Influence of co-registration on lesion characterization in diffusion-weighted breast MRI

To evaluate if co-registering Diffusion-Weighted Imaging (DWI) before generating Apparent Diffusion Coefficient (ADC) maps can improve differentiating benign and malignant breast lesions in MRI based on the A6702 thresholds.

This IRB-approved study involved an in-house dataset and the publicly available ACRIN-6698 dataset, both including multi b-value DWI. In phase one, 16 ANTs library-based co-registration methods were evaluated on a subset of n = 138 cases from our in-house cohort. The quantitative assessment included mean ADC values of manually segmented lesions (diagnostic metrics using individual and A6702-defined thresholds) and coefficient of Variation. In the second phase, the best-performing methods were tested for generalizability on an unseen set of 40 cases (20 from in-house and 20 from external dataset). Three blinded readers segmented lesions on co-registered and non-co-registered ADC maps. Agreement and consistency were evaluated via Bland-Altman, segmentation distance, and intraclass correlation coefficient.

Rigid co-registration using DWI at b = 750 s/mm² as reference (b750-Rigid) improved accuracy of both optimal/conservative A6702 trial thresholds with sensitivity/specificity increasing from 93%/10% to 97%/30% and 100%/30% respectively. Mean ADC values were not significantly different after co-registration (p > 0.05).

Co-registration of DWI images before ADC map generation, particularly using b750-Rigid registration, seems promising for improving lesion classification in breast MRI. Further validation is warranted.