An Artificial Intelligence Based Approach Toward Predicting Mortality in Head and Neck Cancer Patients With Relation to Smoking and Clinical Data

Abstract

Head and neck cancers are one of the most common cancers in the world which affects the mouth, throat, and tongue regions of the human body. Lifestyle factors such as smoking, and tobacco have been long associated with the generation of cancerous cells in the body. This paper is a novel approach towards extracting the correlation between these life factors and head and neck cancers, supported by crucial cancer attributes like the tumor-node-metastasis and human papilloma virus. Mortality prediction algorithms in cases of head and neck cancers will help doctors pre-determine the factors that are most crucial and help deliver specialized and targeted treatments. The paper used eight machine learning and four deep learning hyper-parameter tuned models to predict the mortality rate associated with head and neck cancer. The maximum accuracy of 98.8% was achieved by the gradient boosting algorithm in the paper. The feature importance of smoking and human papilloma virus positivity using the same classifier was approximately 4% and 2.5% respectively. The most influential factor in mortality prediction was the duration of follow-up from diagnosis to the last contact date, with 40.8% importance. Quantitative results from the area under the receiver operating characteristic curve substantiate the classifiers’ performance, with a maximum value of 0.99 for gradient boosting. This paper is bound to impact many medical professionals by helping them predict the mortality of cancer patients and aid appropriate treatments.

Keywords

• Head and neck cancer
• human papillomavirus
• machine learning
• smoking
• deep learning