Advanced multicompartment diffusion model for noninvasive grading of endometrial cancer: comparative analysis with apparent diffusion coefficient (ADC) histogram parameters
To evaluate the predictive ability of restricted spectrum imaging (RSI) model-derived parameters for grading endometrial cancer (EC) and compare their performance with the histogram parameters of the mono-exponential model. A total of 63 patients with EC were enrolled. Regions of interest were manually delineated, and voxel-wise fitting was performed using both a mono-exponential model and 2-4-compartments RSI models. The optimal model was determined based on the Bayesian Information Criterion. Histogram parameters of the apparent diffusion coefficient (ADC) (mean, variance, 10/25/50/75/90th percentile, minimum, maximum, kurtosis, skewness) were extracted. One-way analysis of variance (ANOVA) or Kruskal-Wallis tests were employed to analyse differences in magnetic resonance imaging (MRI) parameters across histological grades. Receiver operating characteristic curve analysis was used to assess their diagnostic performance. The four-compartment RSI model was identified as the optimal model for characterising EC lesions. RSI4-F1/F3 exhibited significant differences between G1/G2 or G1+G2/G3 lesions, and their combination achieved the highest diagnostic performance (AUC = 0.872, 0.759), outperforming all ADC histogram parameters. Significant differences in RSI4-F1/F2/F3 were observed between G1/G3 lesions, with their combination yielding an AUC of 0.922, comparable to ADCmin (AUC = 0.918). Only RSI4-F1/F3 effectively differentiated between G2/G3 or lesions, with their combination yielding the highest AUC (0.729). Incorporating tumour size further enhanced diagnostic performance across all grades (AUC = 0.913, 0.946, 0.757, 0.791 for G1/G2, G1/G3, G2/G3, G1+G2/G3, respectively). The four-compartment RSI model provides valuable insights into the component weights of tumour microenvironment across EC grades. This approach enhances the noninvasive grading of EC lesions.
