An Information-Theoretic Paradigm in Modelling Diffusion-Weighted Imaging of Prostate Cancer: EDDIE (Entropy of Divergence of DWI Decay Curve)

Motivation: Typically, diffusion-weighted imaging (DWI) modelling is assumption-based using e.g. exponential models, but nonparametric (data-based) methods have not been explored. Goal(s): We propose a information-theoretic paradigm for DWI modelling which results in a novel radiomics for DWI of prostate cancer (PCa). Approach: The proposed radiomics, EDDIE (entropy of divergence of DWI decay curve) is formulated as entropy of information lost from approximating a reference by DWI decay curves. It is subjected to classification of clinically significant and insignificant PCa using test-retest DWI datasets of 78 patients. Results: EDDIE achieved an AUC score of 0.77 and an ICC (3,1) of 0.78 which indicates good repeatability. Impact: The proposed approach is nonparametric (assumption-free), interpretable (mathematically and physically meaningful) and complete (higher-order measurement). These may contribute towards more accurate and efficient DWI modelling. Besides, the associated novel radiomics could help ushering in more information-theoretic developments in this field.