Radiomics approach to distinguish between benign and malignant soft tissue tumors on magnetic resonance imaging.

Objective: To build a radiomics signature based on MRI images and evaluate its capability for preoperatively identifying the benign and malignant Soft tissue neoplasms (STTs). Materials and methods: 193 patients (99 malignant STTs and 94 benign STTs) were at random segmented into a training cohort (69 malignant STTs and 65 benign STTs) and a validation cohort (30 malignant STTs and 29 benign STTs) with a portion of 7:3. Radiomics features were extracted from T2 with fat saturation and T1 with fat saturation and gadolinium contrast images. Radiomics signature was developed by the least absolute shrinkage and selection operator (LASSO) logistic regression model. The receiver that operated characteristics curve (ROC) analysis was used to assess radiomics signature's prediction performance. Inner validation was performed on an autonomous cohort that contained 40 patients. Results: A radiomics was developed by a total of 16 radiomics features (5 original shape features and 11 were wavelet features) achieved favorable predictive efficacy. Malignant STTs showed higher radiomics score than benign STTs in both training cohort and validation cohort. A good prediction performance was shown by the radiomics signature in both training cohorts and validation cohorts. The training cohorts and validation cohorts had an area under curves (AUCs) of 0.885 and 0.841, respectively. Conclusions: A radiomics signature based on MRI images can be a trustworthy imaging biomarker for identification of the benign and malignant STTs, which could help guide treatment strategies.