Developing a multi-channel deep-learning model for automatically quantifying malignant bone disease from Multiparametric Whole-Body MRI

Motivation: Multiparametric Whole-Body MRI (mpWB-MRI) provides quantitative biomarkers such as relative fat content, Total Diffusion Volume (TDV), and Apparent Diffusion Coefficient (ADC) for cancer detection and assessment. Goal(s): To address the clinical need for algorithms that automatically detect tumours using mpWB-MRI data and measure TDV and ADC from delineated regions. Approach: Develop a multi-channel deep-learning model using b900, ADC, and relative fat fraction images. Results: Our model demonstrated promising performance in delineating malignant bone lesions in patients with Advanced Prostate Cancer and Multiple Myeloma achieving a relative difference of TDV and ADC below 6% and 3% compared with expert-defined delineations. Impact: Our multi-channel model can automatically quantify TDV and ADC from suspected malignant bone lesions across treatment with accuracy close to 70%, which can assist with clinical decision-making in patients with systemic cancer spread.