Survival Prediction for Oral Cancer Patients: A Machine Learning Approach

Abstract

There is a high incidence of oral cancer in Brazil, with 150,000 new cases estimated for 2020-2022. In most cases, it is diagnosed at an advanced stage and are related to many risk factors. The Registro Hospitalar de Câncer (RHC), managed by Instituto Nacional de Câncer (INCA), is a nation-wide database that integrates cancer registers from several hospitals in Brazil. RHC is mostly an administrative database but also include clinical, socioeconomic and hospitalization data for each patient with a cancer diagnostic in the country. For these patients, prognostication is always a difficult task a demand multi-dimensional analysis. Therefore, exploiting large-scale data and machine intelligence approaches emerge as promising tool for computer-aided decision support on death risk estimation. Given the importance of this context, some works have reported high prognostication effectiveness, however with extremely limited data collections, relying on weak validation protocols or simple robustness analysis. Hence, this work describes a detailed workflow and experimental analysis for oral cancer patient survival prediction considering careful data curation and strict validation procedures. By exploiting multiple machine learning algorithms and optimization techniques the proposed approach allowed promising survival prediction effectiveness with F1 and AuC-ROC over 0.78 and 0.80, respectively. Moreover, a detailed analysis have shown that the minimization of different types of prediction errors were achieved by different models, which highlights the importance of the rigour in this kind of validation.

Prediction of Skin Tumor Invasiveness: A National Analysis Through Explainable Artificial Intelligence (XAI)

“…In this scenario, the INCA, which operates under the aegis of the Brazil's Ministry of Health, assumes a pivotal role in the ongoing surveillance and support of individuals diagnosed with cancer through the utilization of the RHC. As detailed by [Lopes et al 2021], this comprehensive system, which has been integrated into hospital infrastructure, is designed to collect, store, efficiently process, and rigorously analyze patient-specific data.…”