Background
Even though immune checkpoint inhibitors have improved survival rates for patients with recurrent and/or metastatic head and neck squamous cell carcinoma (HNSCC), a significant number of patients do not show durable responses. Previous studies have described correlation between overall survival and features derived from both collagen fibers and spatial arrangement of tumor-infiltrating lymphocytes (TILs) in patients with HNSCC. In this study, we developed and evaluated an AI-informed integration of spatial architecture of collagen and TILs on digitized Hematoxylin and Eosin (H&E) slides from patients with HNSCC to predict their response to immunotherapy (IO).
Methods
Whole slide images (WSIs) from 71 HNSCC patients treated with IO were collected from University Hospitals (D1, n=41) and Emory University (D2, n=30). IO response was defined per RECIST v1.1. D1 trained a predictive model, while D2 was used as an independent validation cohort. Computer algorithms identified 2 types of nuclei (TILs & non-TILs) and built proximity-based clusters. Metrics on density, intersection, and neighborhood were computed. The minimum Redundancy Maximum Relevance algorithm identified TIL features most correlated with response in D1. Additionally, an image processing model identified stromal collagen fibers and computed 5 features: fragmentation, bundling, rigidity, anisotropy, and density. A random forest classifier was trained using TIL and collagen features to differentiate responders from non-responders and was then validated on D2.
Results
The model based on the combination of features from collagen fibers and spatial arrangement of TILs was able to differentiate between those patients treated with IO who will respond from those who will not with an area under the receiver operating characteristic curve of 0.92, specificity of 1.00, and sensitivity of 0.50. table 1 presents the corresponding confusion matrix.
Conclusions
A computerized image analysis model based on measurements of collagen fibers and spatial arrangement of TILs from digitized primary tumor H&E slides was found to be predictive of IO response in patients with HNSCC. With validation in a larger cohort, this approach has the potential to help in better identifying patients who will benefit from IO while sparing some patients from unnecessary therapy and allowing to direct them to alternative treatments.
Acknowledgements
Research reported in this publication was supported by the National Cancer Institute under award numbers R01CA249992-01A1, R01CA202752-01A1, R01CA208236-01A1, R01CA216579-01A1, R01CA220581-01A1, R01CA257612-01A1, 1U01CA239055-01, 1U01CA248226-01, 1U54CA254566-01, National Heart, Lung and Blood Institute 1R01HL15127701A1, R01HL15807101A1, National Institute of Biomedical Imaging and Bioengineering 1R43EB028736-01, National Center for Research Resources under award number 1 C06 RR12463-01, VA Merit Review Award IBX004121A from the United States Department of Veterans Affairs Biomedical Laboratory Research and Development Service the Office of the Assistant Secretary of Defense for Health Affairs, through the Breast Cancer Research Program (W81XWH-19-1-0668), the Prostate Cancer Research Program (W81XWH-15-1-0558, W81XWH-20-1-0851), the Lung Cancer Research Program (W81XWH-18-1-0440, W81XWH-20-1-0595), the Peer Reviewed Cancer Research Program (W81XWH-18-1-0404, W81XWH-21-1-0345, W81XWH-21-1-0160), the Mayo Clinic Breast Cancer SPORE grant P50 CA116201 from the NIH, the Kidney Precision Medicine Project (KPMP) Glue Grant, and sponsored research agreements from Bristol Myers-Squibb, Boehringer-Ingelheim, Eli-Lilly and AstraZeneca. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health, the U.S. Department of Veterans Affairs, the Department of Defense, or the United States Government.
Ethics Approval
The study was approved by the IRB Committee at Emory University.
Abstract 1205 Table 1