Impact of the calculation algorithm on biexponential fitting of diffusion-weighted MRI in upper abdominal organs

Abstract

Purpose To compare the variability, precision, and accuracy of six different algorithms (Levenberg-Marquardt, Trust-Region, Fixed-Dp, Segmented-Unconstrained, Segmented-Constrained, and Bayesian-Probability) for computing intravoxel-incoherent-motion-related parameters in upper abdominal organs. Methods Following the acquisition of abdominal diffusion-weighted magnetic resonance images of 10 healthy men, six distinct algorithms were employed to compute intravoxel-incoherent-motion-related parameters in the left and right liver lobe, pancreas, spleen, renal cortex, and renal medulla. Algorithms were evaluated regarding inter-reader and intersubject variability. Comparability of results was assessed by analyses of variance. The algorithms' precision and accuracy were investigated on simulated data. Results A Bayesian-Probability based approach was associated with very low inter-reader variability (average Intraclass Correlation Coefficients: 96.5-99.6%), the lowest inter-subject variability (Coefficients of Variation [CV] for the pure diffusion coefficient Dt: 3.8% in the renal medulla, 6.6% in the renal cortex, 10.4-12.1% in the left and right liver lobe, 15.3% in the spleen, 15.8% in the pancreas; for the perfusion fraction Fp: 15.5% on average; for the pseudodiffusion coefficient Dp: 25.8% on average), and the highest precision and accuracy. Results differed significantly (P < 0.05) across algorithms in all anatomical regions. Conclusion The Bayesian-Probability algorithm should be preferred when computing intravoxel-incoherent-motion-related parameters in upper abdominal organs.