Towards a ‘resolution limit’ for DW‐MRI tumor microstructural models: A simulation study investigating the feasibility of distinguishing between microstructural changes

• Published in: Magnetic resonance in medicine 2019-04, Vol.81 (4), p.2288-2301
• Main Authors: McHugh, Damien J., Hubbard Cristinacce, Penny L., Naish, Josephine H., Parker, Geoffrey J. M.
• Format: Article
• Language: English
• Subjects: Algorithms; Apoptosis; Axons - pathology; Biomarkers; Biomarkers - metabolism; Cell death; Cell density; Cell size; Change detection; Computer Simulation; Density; Diffusion Magnetic Resonance Imaging; diffusion‐weighted MRI; Feasibility Studies; Full Papers—Imaging Methodology; Humans; Magnetic resonance imaging; Mathematical models; microstructural changes; Microstructure; Mimicry; Neoplasms - diagnostic imaging; Neoplasms - pathology; Normal Distribution; optimum design; Parameter estimation; Reproducibility of Results; resolution limit; Shrinkage; Signal-To-Noise Ratio; tumor microstructure; Tumors
• ISSN: 0740-3194; 1522-2594
• DOI: 10.1002/mrm.27551

Purpose To determine the feasibility of extracting sufficiently precise estimates of cell radius, R, and intracellular volume fraction, fi, from DW‐MRI data in order to distinguish between specific microstructural changes tissue may undergo, specifically focusing on cell death in tumors. Methods Simulations with optimized and non‐optimized clinical acquisitions were performed for a range of microstructures, using a two‐compartment model. The ability to distinguish between (i) cell shrinkage with cell density constant, mimicking apoptosis, and (ii) cell size constant with cell density decreasing, mimicking loss of cells, was evaluated based on the precision of simulated parameter estimates. Relationships between parameter precision, SNR, and the magnitude of specific parameter changes, were used to infer SNR requirements for detecting changes. Results Accuracy and precision depended on microstructural properties, SNR, and the acquisition protocol. The main benefit of optimized acquisitions tended to be improved accuracy and precision of R, particularly for small cells. In most cases considered, higher SNR was required for detecting changes in R than for changes in fi. Conclusions Given the relative changes in R and fi due to apoptosis, simulations indicate that, for a range of microstructures, detecting changes in R require higher SNR than detecting changes in fi, and that such SNR is typically not achieved in clinical data. This suggests that if apoptotic cell size decreases are to be detected in clinical settings, improved SNR is required. Comparing measurement precision with the magnitude of expected biological changes should form part of the validation process for potential biomarkers.