(P 408) CHANGES IN COMPUTED TOMOGRAPHY FEATURES INDUCED BY BRIGIMADLIN (BI 907828) IN PATIENTS WITH LIPOSARCOMA: INSIGHTS FROM PHASE I STUDIES

Objective: Traditional response assessment criteria such as RECIST are limited in evaluating treatment response in patients with liposarcoma receiving systemic therapy. The field of drug development for liposarcoma is witnessing a paradigm shift with the potential use of radiomics, a quantitative analysis of imaging features, for developing predictive and prognostic signatures. We aim to externally validate previously validated radiomic features identified in multicenter clinical trials to prove the concept of the potential of radiomics to provide valuable insights in liposarcoma drug development.

Methods: The cohort consisted of 22 patients, with 8 diagnosed with well-differentiated liposarcoma and 14 with dedifferentiated liposarcoma, treated with brigimadlin (BI 907828) in Phase 1 clinical trials NCT03449381 and NCT03964233. These patients had a total of 40 measurable tumor lesions based on CT scans. Radiologists performed semi-automatic segmentation of the tumor lesions at all available time points and extracted tumor features using both Columbia Image Feature Extractor (CIFE) and pyRadiomics software. Ten pre-identified prognostic and/or predictive radiomic features extracted at baseline and Week 6 were used to compute delta features reflecting changes between these time points, with the objective of differentiating responders (patients with a partial or complete response based on RECIST 1.1 criteria) from non-responders through per-patient analysis (n=22) and per-lesion analysis (n=40). As an ancillary study, independent radiologists assessed qualitative features such as solidity, necrosis, shape, calcification, vascularization, and septations.

Results: A majority of patients (16 out of 22) had tumors located in the retroperitoneum. The dose regimen for brigimadlin in this study varied among the enrolled patients. In the per-patient analysis, at baseline, we observed that responders (P-value of patient-based/lesion-based analyses) had higher tumor volume (P=0.002/ < 0.0001) and a distinct radiomics phenotype (Figure 1). The analysis of early changes in tumor phenotype demonstrated that responders had a decrease in both tumor volume (P=0.003/0.001) and in tumor heterogeneity (P=0.006/0.003). In line with the quantitative radiomics analysis, treatment response was more likely to be observed in qualitatively predominantly fatty lesions and in those with fewer and thinner septae.

Conclusion: Based on a small patient cohort, the radiomic analysis hints at the potential of the selected features to predict treatment response at both patient and lesion levels, with the most significant feature being the reduction in tumor volume and heterogeneity on CT scans at Week 6. The validation of these radiomic features, previously identified in multicenter trials, indicates their potential in identifying patients who will benefit from treatment; this was further supported by the qualitative assessment performed by radiologists.