Back to all papers

Validation of ST-RADS calculator: AI web tool for soft tissue-RADS implementation in de novo non-operated tumors and tumor-like lesions.

September 4, 2026pubmed logopapers

Authors

Ramachandran S,Chhabra R,Kharbat Y,Lodhi S,Xi Y,Amini B,Ahlawat S,Subhawong T,Samet J,Chhabra A

Affiliations (8)

  • Texas A&M College of Medicine, Bryan, TX, USA.
  • UT Austin, TX, USA.
  • Department of Radiology, UT Southwestern Medical Center, Dallas, TX, USA.
  • Diagnostic Imaging Division, MD Anderson, USA.
  • Department of Radiology, Johns Hopkins, Baltimore, MD, USA.
  • Department of Radiology, University of Miami, Miami, FL, USA.
  • Department of Radiology, Northwestern University, Chicago, IL, USA.
  • Department of Radiology, UT Southwestern Medical Center, Dallas, TX, USA; Department of Orthopedic Surgery, UT Southwestern Medical Center, Dallas, TX, USA. Electronic address: [email protected].

Abstract

Soft-tissue RADS (ST-RADS) is an ACR-sponsored MRI-based scoring system for risk-stratification of soft tissue tumors and tumor-like lesions. Despite strong interobserver agreement and diagnostic accuracy, clinical adoption is limited by reader fatigue and workflow burden. We developed and validated an AI-based web tool (ST-RADS calculator) using ST-RADS algorithms and GPT API to facilitate routine implementation. The ST-RADS calculator offers three input modes: image-only, text features-only, and combined. A retrospective study of 134 histopathology-confirmed soft tissue tumors compared ST-RADS scores from four fellowship-trained musculoskeletal (MSK) radiologists against the calculator using all three approaches. Sensitivity and specificity were compared using generalized estimating equations with pairwise contrasts. Sensitivity was similar across all methods: 85% for radiologists, 86% for image-only, 81% for text features-only, and 86% for combined (p > 0.90). Specificity differed significantly: 76.5% for radiologists, 69.4% for text features-only, 61.2% for combined, and 25.5% for image-only. Text features-only did not differ significantly from radiologists (p = 0.46). Dichotomized accuracy (benign 1-3 vs. malignant 4-5) was 71% for radiologists, 72% for text features-only, 61% for combined, and 42% for image-only. Full-scale accuracy ranged from 19% (image-only) to 52% (combined). The ST-RADS calculator with text-based input achieved accuracy comparable to expert MSK radiologists, while image-only approaches require further refinement. This tool can support routine ST-RADS implementation in clinical practice.

Topics

Journal Article

Ready to Sharpen Your Edge?

Subscribe to join 11k+ peers who rely on RadAISlice. Get the essential weekly briefing that empowers you to navigate the future of radiology.

We respect your privacy. Unsubscribe at any time.