Detection of Lung Cancer Mutation Based on Clinical and Morphological Features Using Adaptive Boosting Method

Abstract

Lung cancer is one of the leading causes of cancer-related mortality worldwide, with mutation detection playing a critical role in personalizing treatment strategies. Identifying Epidermal Growth Factor Receptor (EGFR) mutations non-invasively remains challenging due to the complex clinical and morphological patterns in patients. This study aimed to develop an Adaboost-based machine learning model to detect lung cancer mutations using clinical and morphological data from patients. Our contribution includes a novel application of the Adaboost algorithm to analyze clinical and morphological features, providing an efficient, non-invasive alternative for mutation detection. The dataset included clinical attributes and morphological data from 80 patients, processed through various preprocessing techniques such as imputation, outlier removal, and feature selection. Data were split into training and testing sets with an 80/20 ratio, and Adaboost was trained with optimized hyperparameters to maximize accuracy and robustness. The experimental results showed that Adaboost outperformed other machine learning algorithms, achieving high accuracy and stability across all preprocessing scenarios. After feature selection using ANOVA, Adaboost achieved an accuracy of 83% and an AUC of 0.90, indicating its robustness and sensitivity in mutation detection. The model was effective even when outliers were removed, and it maintained superior cross-validation scores compared to Naive Bayes, Decision Tree, KNN, and SVM. In conclusion, the Adaboost algorithm proved to be a reliable approach for detecting lung cancer mutations based on clinical and morphological data, offering potential as a supportive tool in clinical decision-making.