2019/09/06 by Priyam Das, Das, Priyam, Debsurya De +13
Computer Science · Medicine · #Computation (stat.CO) #FOS: Computer and information sciences #Head and Neck Cancer Studies #Medical Image Segmentation Techniques #Methodology (stat.ME) #Radiomics and Machine Learning in Medical Imaging
paper · pdf · doi:10.48550/arxiv.1909.04024
openalex publication_date 2019/09/06 · openalex created_date 2025/10/10 · openalex updated_date 2026/08/01
In the context of a binary classification problem, the optimal linear\ncombination of continuous predictors can be estimated by maximizing an\nempirical estimate of the area under the receiver operating characteristic\n(ROC) curve (AUC). For multi-category responses, the optimal predictor\ncombination can similarly be obtained by maximization of the empirical\nhypervolume under the manifold (HUM). This problem is particularly relevant to\nmedical research, where it may be of interest to diagnose a disease with\nvarious subtypes or predict a multi-category outcome. Since the empirical HUM\nis discontinuous, non-differentiable, and possibly multi-modal, solving this\nmaximization problem requires a global optimization technique. Estimation of\nthe optimal coefficient vector using existing global optimization techniques is\ncomputationally expensive, becoming prohibitive as the number of predictors and\nthe number of outcome categories increases. We propose an efficient\nderivative-free black-box optimization technique based on pattern search to\nsolve this problem. Through extensive simulation studies, we demonstrate that\nthe proposed method achieves better performance compared to existing methods\nincluding the step-down algorithm. Finally, we illustrate the proposed method\nto predict swallowing difficulty after radiation therapy for oropharyngeal\ncancer based on radiation dose to various structures in the head and neck.\n