Program Information
MRI Radiomic Features Are Superior to Clinical Stage for Outcome Prediction in Sarcoma
M Spraker1*, L Wootton1 , D Hippe1 , A Chaovalitwongse2 , M Macomber1 , T Chapman1 , M Hoff1 , S Pollack1 , E Kim1 , M Nyflot1 , (1) University of Washington, Seattle, WA, (2) University of Arkansas, Fayetteville, Arkansas
Presentations
SU-K-605-16 (Sunday, July 30, 2017) 4:00 PM - 6:00 PM Room: 605
Purpose: Soft tissue sarcomas (STS) exhibit variable behavior and therapeutic decisions for any individual patient remain a critical challenge. We compared the accuracy of MRI radiomic features versus clinical stage for prediction of progression-free (PFS) and overall survival (OS) in STS.
Methods: A retrospective analysis of 120 patients with AJCC stage II-III STS was performed. 44 radiomic features were extracted from tumor regions defined on pretreatment T1 MRI by a radiation oncologist. The best predictor from each of four radiomic classes (intensity histograms, co-occurrence matrices, neighborhood difference matrices, zone size matrices) was selected based on standardized hazard ratios (HR) from univariate analyses. Cox regression models for PFS at 18 months and OS at 36 months were generated. Models evaluated were radiomic features alone, clinical stage alone, and radiomic features + stage. Harrell’s c-index was used to evaluate model performance following optimism adjustment with the bootstrap.
Results: At 18 months, 42 patients had progressed and at 36 months there were 27 deaths. Tumor volume (HR=2.7), dissimilarity (HR=0.4), busyness (HR=2.3) and large-zone/low-gray emphasis (HR=2.2) were selected for OS prediction while tumor volume (HR=1.9), correlation (HR=1.6), busyness (HR=1.8) and large-zone/low-gray emphasis (HR=1.7) were selected for PFS. Radiomic features alone had c-index=0.70 for OS and c-index=0.64 for PFS. Clinical stage was less predictive than radiomic features for both OS (c-index: 0.64 vs 0.70, p=0.002) and PFS (c-index: 0.58 vs 0.64, p=0.012). Additionally, combining radiomic features with clinical stage did not significantly improve prediction versus radiomic features for OS (c-index: 0.73 vs. 0.70, p=0.25) or PFS (c-index: 0.63 vs. 0.64, p=0.83).
Conclusion: These results suggest that models of radiomic features can provide good prediction of outcomes for patients with STS compared to clinical stage. Going forward, radiomic models validated in prospective trials may provide new strategies for precision treatment of sarcoma patients.
Funding Support, Disclosures, and Conflict of Interest: Research supported in part by a Research Scholar Grant from RSNA
Contact Email: