Explorative Investigation of Whole-Lesion Histogram MRI Metrics for Differentiating Uterine Leiomyomas and Leiomyosarcomas.

OBJECTIVE The purpose of this study is to assess the utility of texture analysis of multiple MRI sequences for the differentiation of uterine leiomyomas and leiomyosarcomas. MATERIALS AND METHODS Seventeen leiomyosarcomas and 51 leiomyomas undergoing MRI before resection were included. Whole-lesion volumes of interest were placed on T2-weighted images, contrast-enhanced T1-weighted images, and apparent diffusion coefficient (ADC) maps. The diagnostic performance of histogram metrics was assessed. RESULTS For T2-weighted images, significant differences were observed for mean, skewness, entropy, mean of the bottom 10th percentile (mean0-10), mean of the 10th through 25th percentiles (mean10-25), and mean of the 25th through 50th percentiles (mean25-50) (p ≤ 0.010). For T1-weighted contrast-enhanced images, significant differences were observed for mean0-10, mean10-25, and mean25-50 (p ≤ 0.045). For the ADC maps, no metrics showed a significant difference (p ≥ 0.067). Metrics with AUC greater than 0.8 were the mean0-10 (0.875), mean10-25 (0.863), mean25-50 (0.839), and mean (0.802) for T2-weighted imaging. The mean0-10, mean10-25, and mean25-50 for T2-weighted imaging all achieved greater AUCs than did the standard mean (p ≤ 0.038). Patients with leiomyosarcoma were significantly older than those with leiomyoma (p < 0.001; AUC = 0.866). At multivariable regression, significant independent predictors of leiomyosarcoma were patient age (p = 0.002) and T2-weighted imaging mean0-10 (p = 0.004), with a combined AUC of 0.955. Patient age achieved sensitivity of 82.4% and specificity of 92.2%; T2-weighted imaging mean0-10 achieved sensitivity of 82.4% and specificity of 74.5%. CONCLUSION For whole-lesion histogram metrics obtained on various MRI sequences, T2-weighted images provided the highest, and ADC maps the lowest, performance for differentiating uterine leiomyomas and leiomyosarcomas. Metrics reflecting percentiles from the bottom half of the histogram distribution outperformed the standard mean. Models combining the T2-weighted imaging whole-lesion metrics and patient age achieved particularly high diagnostic performance. Although these findings require validation in larger studies, they have implications for facilitating improved treatment selection for these two entities.