Federated learning enables big data for rare cancer boundary detection

Although machine learning (ML) has shown promise across disciplines, out-of-sample generalizability is concerning. This is currently addressed by sharing multi-site data, but such centralization is challenging/infeasible to scale due to various limitations. Federated ML (FL) provides an alternative paradigm for accurate and generalizable ML, by only sharing numerical model updates. Here we present the largest FL study to-date, involving data from 71 sites across 6 continents, to generate an automatic tumor boundary detector for the rare disease of glioblastoma, reporting the largest such dataset in the literature (n = 6, 314). We demonstrate a 33% delineation improvement for the surgically targetable tumor, and 23% for the complete tumor extent, over a publicly trained model. We anticipate our study to: 1) enable more healthcare studies informed by large diverse data, ensuring meaningful results for rare diseases and underrepresented populations, 2) facilitate further analyses for glioblastoma by releasing our consensus model, and 3) demonstrate the FL effectiveness at such scale and task-complexity as a paradigm shift for multi-site collaborations, alleviating the need for data-sharing.

NIMG-54. RADIOMIC FEATURES FROM LESION HABITAT PREDICT RESPONSE TO COMBINATION OF NIVOLUMAB AND BEVACIZUMAB IN PATIENTS WITH RECURRENT GLIOBLASTOMA: A FEASIBILITY STUDY

PURPOSE

The use of immunotherapy in glioblastoma management is under active investigation. Glioblastomas are “cold” tumors, meaning that they have inactivated or fewer tumor infiltrative lymphocytes in addition to substantial tumor necrosis, attributing to their poor response to immunotherapy. A significant challenge is the apriori identification of Glioblastoma patients who will respond favorably to immunotherapy. In this work, we evaluated the ability of computerized MRI-based quantitative features (radiomics) extracted from the lesion habitat (including enhancing lesion, necrosis, and peritumoral hyperintensities) to predict response and progression-free survival (PFS) in recurrent GBM patients treated with combination of Nivolumab and Bevacizumab.

METHODS

Immunotherapy response assessment in neuro-oncology (iRANO) criteria along with PFS were used to analyze n=50 patients enrolled in a randomized clinical trial where patients received Nivolumab with either standard or low dose Bevacizumab. These patients were assessed to see if they had complete response, partial response, stable disease (i.e. responders, n=31), or disease progression (i.e. non-responders, n=19). Lesion habitat constituting necrotic core, enhancing tumor, and edema were delineated by expert radiologist on Gd-T1w, T2w and FLAIR MRI scans. COLIAGE radiomic features from each of the delineated regions were selected using minimum redundancy maximum relevance (mRMR) via cross-validation, to segregate non-responder patients from responders. A multivariable cox proportional hazard model was used to predict survival (PFS).

RESULTS

CoLlAGe correlation, sum average, and sum variance features (capture local heterogeneity) from the lesion habitat, were found to segregate non-responder patients from responders with an accuracy of 86%, followed by 80% using features from peritumoral hyperintensities and 78% from enhancing tumor. In our survival analysis, C-index of 0.688 was obtained using features from the entire lesion habitat, followed by peritumoral hyperintensities (0.675) and enhancing tumor (0.656).

CONCLUSION

Radiomic features from the lesion habitat may predict response to combination of Nivolumab and Bevacizumab in recurrent Glioblastomas.

NEIM-01. PREDICTION OF RESPONSE TO COMBINATION OF NIVOLUMAB AND BEVACIZUMAB IN PATIENTS WITH RECURRENT GLIOBLASTOMA VIA RADIOMIC ANALYSIS ON CLINICAL MRI SCANS

PURPOSE

The use of immunotherapy in glioblastoma management is under active investigation. Glioblastomas are “cold” tumors, meaning that they have inactivated or fewer tumor infiltrative lymphocytes in addition to substantial tumor necrosis, attributing to their poor response to immunotherapy. A significant challenge is the apriori identification of Glioblastoma patients who will respond favorably to immunotherapy. In this work, we evaluated the ability of computerized MRI-based quantitative features (radiomics) extracted from the lesion habitat (including enhancing lesion, necrosis, and peritumoral hyperintensities) to predict response and progression-free survival (PFS) in recurrent GBM patients treated with combination of Nivolumab and Bevacizumab.

METHODS

Immunotherapy response assessment in neuro-oncology (iRANO) criteria along with PFS were used to analyze n=50 patients enrolled in a randomized clinical trial where patients received Nivolumab with either standard or low dose Bevacizumab. These patients were assessed to see if they had complete response, partial response, stable disease (i.e. responders, n=31), or disease progression (i.e. non-responders, n=19). Lesion habitat constituting necrotic core, enhancing tumor, and edema were delineated by expert radiologist on Gd-T1w, T2w and FLAIR MRI scans. COLIAGE radiomic features from each of the delineated regions were selected using minimum redundancy maximum relevance (mRMR) via cross-validation, to segregate non-responder patients from responders. A multivariable cox proportional hazard model was used to predict survival (PFS).

RESULTS

CoLlAGe correlation, sum average, and sum variance features (capture local heterogeneity) from the lesion habitat, were found to segregate non-responder patients from responders with an accuracy of 86%, followed by 80% using features from peritumoral hyperintensities and 78% from enhancing tumor. In our survival analysis, C-index of 0.688 was obtained using features from the entire lesion habitat, followed by peritumoral hyperintensities (0.675) and enhancing tumor (0.656).

CONCLUSION

Radiomic features from the lesion habitat may predict response to combination of Nivolumab and Bevacizumab in recurrent Glioblastomas.