Objective: To quantify the diagnostic utility of imaging features in soft tissue masses (STMs) and to provide a ranked list of predictors for malignancy.
Subjects and methods: Imaging features in 260 cases of STMs with verified histology were assessed. Diagnostic properties including sensitivity, specificity, positive and negative predictive values, likelihood/odds ratios (OR) and normalized variance (NV) via random forest analysis were calculated. The diagnostic utility of an 8-item checklist consisting of the highest-ranked features was evaluated through a receiver-operating-characteristics (ROC) curve.
Results: The most predictive features (NV/OR in parentheses) were heterogeneous contrast-enhancement in ultrasound (297.9/15.1) and MRI (197.3/11.9), lesion roundness (209.8/5.5), diffusion restriction (175.8/9.3), cystic/necrotic intralesional areas (167.1/8.3), higher patient age (159.0/2.6), surrounding oedema (155.4/6.5) and intralesional Doppler hypervascularity (134.4/5.1). A simple 8-item checklist was highly predictive of malignancy in cases with at least 75% positive features (0.90 area under the ROC curve, 87.0% sensitivity, 84.5% specificity, 59.5% positive and 96.1% negative predictive value, 36.5 odds ratio) even in cases with only partial feature availability.
Conclusion: Features vary widely in their diagnostic value in STMs; an 8-item checklist based on the eight most decisive features can be a simple tool to assess the likelihood for malignancy in unknown soft tissue masses, even though a stratified approach is certainly still advisable when first confronted with an STM.
Keywords: Classification; Features; Prediction; Sarcoma; Soft tissue tumours.
Copyright © 2017 Elsevier B.V. All rights reserved.