An electronic nose-based assistive diagnostic prototype for lung cancer detection with conformal prediction

Abstract

Lung cancer leads to high mortalities in various countries while the reliability of cancer diagnosis has not been paid enough attention. In this work, a novel application of conformal prediction in lung cancer diagnosis with electronic nose is introduced. The nonconformity measurement is based on k-nearest neighbors. In offline prediction, accuracies of 87.5% and 83.33% have been achieved by conformal predictors based on 1NN and 3NN respectively, outperforming those of simple k-nearest neighbor predictors. Additionally, conformal predictors provides confidence and credibility information of each prediction that could inform the patients of diagnostic risks. In online prediction, with increasing number of samples, the frequency of errors given by conformal predictions can gradually be limited by the significance level set by users. This project manifests that electronic nose promises to be an applicable cheaper analytic tool in assisting lung cancer diagnosis and conformal prediction provides a promising method to ensure reliability.