Comparison of Continuous‐Time Random Walk and Fractional Order Calculus Diffusion Models for Preoperative Breast Lesion Characterization: Diagnosis, Biomarker Prediction and Molecular Subtyping

Abstract

ABSTRACT Background Accurate preoperative characterization of breast lesions is important for individualized management. The conventional monoexponential (Mono) model has limited ability to capture the complex microstructure of lesions. Non‐Gaussian diffusion models, including continuous‐time random walk (CTRW) and fractional‐order calculus (FROC), may overcome this limitation, but their value in predicting biomarkers and molecular subtypes remains unclear. Purpose To comprehensively evaluate the diagnostic utility of the Mono, CTRW, and FROC models and their derived parameters for preoperative breast lesion characterization. Study Type Prospective. Population 189 women (mean age, 47.4 ± 11.2 years) with 238 histopathologically confirmed breast lesions (138 benign, 100 malignant). Field Strength/Sequence 3.0‐T MRI; single‐shot spin‐echo echo‐planar imaging for multi‐ b ‐value diffusion‐weighted imaging (10 b ‐values ranging from 0 to 3000 s/mm 2 ). Assessment Diffusion parameters were extracted from whole‐lesion volumes generated from slice‐wise segmentations by two independent, blinded breast radiologists. Multiparameter joint models were constructed using logistic regression. Statistical Tests Model performance was evaluated using repeated 5‐fold cross‐validation and patient‐level cluster bootstrap for AUC comparisons. A two‐sided p < 0.05 was considered statistically significant, and Bonferroni‐adjusted thresholds were applied for multiple comparisons. Results For benign–malignant differentiation, the CTRW and FROC models yielded higher cross‐validated AUCs than the Mono model (0.928, 0.918, and 0.865, respectively). For prognostic biomarkers, the FROC model achieved optimal predictive performance for HER2 status (AUC = 0.836). For molecular subtyping, the CTRW model significantly outperformed the Mono model in distinguishing HER2‐positive lesions (AUC = 0.848 vs. 0.674); however, all models showed limited performance for triple‐negative lesions, with all AUCs ≤ 0.502. Data Conclusion Non‐Gaussian diffusion models, particularly the CTRW model, provide additional value beyond the Mono model for preoperative breast lesion characterization. Evidence Level 2. Technical Efficacy Stage 2.