Date of Award
Medical Doctor (MD)
Cancers of the head and neck are particularly burdensome in volume, mortality, and morbidity, and thus pose significant challenges in clinical management. Advances in imaging modalities and machine learning (ML) methods offer tremendous promise diagnosing, prognosticating, and predicting tumor behavior. In this work, we offer an overview of ML-driven developments in head and neck cancer care and present three distinct methods: a diagnostic classifier evaluating cystic neck masses (CNMs) from imaging, a prognostic regressor for disease recurrence using clinicopathologic data, and a prognostic regressor for disease progression using imaging data. In classifying CNMs, the convolutional neural network (CNN) outperformed radiomics-based models, achieving an accuracy of 73.3%. Using selected clinicopathologic data, oral cancer recurrence prediction was enhanced with random survival forest model (concordance index of 0.761) when compared to traditional cox proportional hazard regression and a neural network-based regressor. A CNN-based hazard regressor achieved a concordance index of 0.582 when predicting disease progression from isolated PET-CT scans. Ultimately, we establish and evaluate novel applications to enhance diagnostic and prognostic evaluations in head and neck cancer with the aim to guide future work to advance clinical management.
Bourdillon, Alexandra Tan, "A Deep Dive In Head & Neck Cancer: Machine Learning Applications In Diagnostic And Prognostic Evaluations" (2022). Yale Medicine Thesis Digital Library. 4055.