Abstract 74P
Background
Evaluating radiological response in localized Soft Tissue Sarcomas (STS) presents challenges due to the disease’s rarity and heterogeneity. Standardized RECIST and MR-adapted Choi criteria frequently diverge in response categorization. The extent of inter- and intraobserver variability within these criteria remains poorly characterized for STS.
Methods
Five radiologists assessed pre- and post-radiotherapy T2 and T1 fat saturated post contrast MRIs of 17 STS patients by: diameter measurement, tumor segmentation and muscle region segmentation. All measurements were performed on the T1 sequence. Patients were assigned RECIST and Choi response categories. Interobserver agreement was quantified by calculating the percentage of complete agreement among radiologists. MRIs of two patients were digitally flipped (left-to-right) and re-assessed. Coefficient of Variation (CoV) of flipped images was computed for intra-reader reliability.
Results
Single-radiologist evaluation of baseline and follow-up scans showed complete agreement in 68% of cases for RECIST and 42% for Choi criteria, in other cases at least one radiologist assigned a different response category. When different radiologists measured baseline and follow-up, complete agreement rates dropped to 58% for RECIST and 5% for Choi. Semi-automatic diameter extraction from tumor segmentation substantially improved the agreement among radiologists: from 58% to 84% for RECIST and from 5% to 37% for Choi. In the reproducibility test using flipped images, complete agreement was observed for RECIST, while Choi showed disagreement in 4/40 observations. Mean CoV were 2.8% (range: 1.0% - 4.2%) for diameter and 5.0% (range: 1.5% - 10.9%) for intensity.
Conclusions
Our study demonstrates substantial variability in radiological response assessment of STS using RECIST and Choi criteria. Semi-automated measurements demonstrate potential for enhancing criteria reliability. These findings highlight a critical need for improved evaluation methods, especially in neoadjuvant treatment settings where pathological confirmation is delayed until after resection. Development of automated, standardized response evaluation methods may therefore be beneficial for this patient population.
Clinical trial identification
Editorial acknowledgement
Legal entity responsible for the study
The authors.
Funding
Has not received any funding.
Disclosure
All authors have declared no conflicts of interest.