A Multicompartmental Diffusion Model for Improved Assessment of Whole-Body Diffusion-weighted Imaging Data and Evaluation of Prostate Cancer Bone Metastases

• Published in: Radiology. Imaging cancer 2023-01, Vol.5 (1), p.e210115-e210115
• Main Authors: Conlin, Christopher C, Feng, Christine H, Digma, Leonardino A, Rodríguez-Soto, Ana E, Kuperman, Joshua M, Rakow-Penner, Rebecca, Karow, David S, White, Nathan S, Seibert, Tyler M, Hahn, Michael E, Dale, Anders M
• Format: Article
• Language: English
• Subjects: Aged; Bayes Theorem; Bone Neoplasms - diagnostic imaging; Bone Neoplasms - secondary; Diffusion Magnetic Resonance Imaging - methods; Humans; Magnetic Resonance Imaging - methods; Male; Prospective Studies; Prostatic Neoplasms - diagnostic imaging; Prostatic Neoplasms - pathology
• ISSN: 2638-616X; 2638-616X
• DOI: 10.1148/rycan.210115

Purpose To develop a multicompartmental signal model for whole-body diffusion-weighted imaging (DWI) and apply it to study the diffusion properties of normal tissue and metastatic prostate cancer bone lesions in vivo. Materials and Methods This prospective study ( : NCT03440554) included 139 men with prostate cancer (mean age, 70 years ± 9 [SD]). Multicompartmental models with two to four tissue compartments were fit to DWI data from whole-body scans to determine optimal compartmental diffusion coefficients. Bayesian information criterion (BIC) and model-fitting residuals were calculated to quantify model complexity and goodness of fit. Diffusion coefficients for the optimal model (having lowest BIC) were used to compute compartmental signal-contribution maps. The signal intensity ratio (SIR) of bone lesions to normal-appearing bone was measured on these signal-contribution maps and on conventional DWI scans and compared using paired tests (α = .05). Two-sample tests (α = .05) were used to compare compartmental signal fractions between lesions and normal-appearing bone. Results Lowest BIC was observed from the four-compartment model, with optimal compartmental diffusion coefficients of 0, 1.1 × 10 , 2.8 × 10 , and >3.0 ×10 mm /sec. Fitting residuals from this model were significantly lower than from conventional apparent diffusion coefficient mapping ( < .001). Bone lesion SIR was significantly higher on signal-contribution maps of model compartments 1 and 2 than on conventional DWI scans ( < .008). The fraction of signal from compartments 2, 3, and 4 was also significantly different between metastatic bone lesions and normal-appearing bone tissue ( ≤ .02). Conclusion The four-compartment model best described whole-body diffusion properties. Compartmental signal contributions from this model can be used to examine prostate cancer bone involvement. Whole-Body MRI, Diffusion-weighted Imaging, Restriction Spectrum Imaging, Diffusion Signal Model, Bone Metastases, Prostate Cancer Clinical trial registration no. NCT03440554 © RSNA, 2023 See also commentary by Margolis in this issue.