Abstracts

RETURN TO ABSTRACT LISTING


E1500. MRI-Based Delta-Radiomics for the Assessment of Chemotherapeutic Response in Soft-Tissue Sarcomas
Authors
  1. Brandon Fields; Keck School of Medicine of University of Southern California
  2. Natalie Demirjian; Keck School of Medicine of University of Southern California
  3. Darryl Hwang; Keck School of Medicine of University of Southern California
  4. Bino Varghese; Keck School of Medicine of University of Southern California
  5. Steven Cen; Keck School of Medicine of University of Southern California
  6. Vinay Duddalwar; Keck School of Medicine of University of Southern California
  7. George Matcuk; Cedars-Sinai Medical Center
Objective:
Estimates of treatment response in soft-tissue sarcomas (STS) that rely on size-based criteria, including RECIST, may fail to appreciate satisfactory biologic response to neoadjuvant chemotherapy (NAC). Though Choi (CT) and modified Choi (MRI) criteria that account for decreased density of enhancement may better predict NAC response, our goal was to validate quantitative features of the underlying radiomics metrics in MRIs of patients taken pre- and post-NAC that could provide an earlier and more accurate assessment of treatment response in STS using a delta-radiomics approach.

Materials and Methods:
44 subjects with pathologically proven STS that received NAC at our institution were retrospectively identified. Inclusion criteria included only subjects who had both a baseline MRI study prior to NAC initiation, as well as a post-treatment MRI at least 2 months following chemotherapy initiation and prior to surgical resection. The most frequent pathologic diagnoses were undifferentiated pleomorphic sarcoma (n=17), synovial sarcoma (n=6), myxoid liposarcoma (n=4), and leiomyosarcoma (n=4). An experienced musculoskeletal radiologist supervised manual segmentations of the 3D regions-of-interest using Synapse 3D software. Subsequently, automated data processing, co-registration, and parameter extraction were performed across multiple different sequences using custom MATLAB code. Delta radiomics was applied to 1708 features that were extracted on 88 pre- and post-NAC scans. Univariate analyses were conducted using Exact test with Wilcoxon score. The machine-learning algorithm was constructed using Random Forest (RF) and Real Adaptive Boosting (Adaboost), with a leave-one-out cross validation. Prediction accuracies were quantified by Area-Under-The-Curve (AUC).

Results:
RF and Adaboost were unable to detect NAC response, with AUCs of 0.40 and 0.44, respectively. Univariate analysis concordantly revealed only 4.75% of parameters with p<0.05.

Conclusion:
While preliminary models have suggested a role for MRI-based delta-radiomics in predicting NAC response using machine-learning algorithms, we do not find these results to be reproducible. Results from published literature with strong prediction accuracies in similar cohorts cannot be validated.