Radiogenomic analysis of lung cancer

Abstract

This thesis focuses on the development of a machine-learning model for classifying the two primary histological subtypes of lung cancer: adenocarcinoma and squamous cell carcinoma. The proposed model integrates radiomic features extracted from patients’ computed tomography (CT) images, genomic data, and clinical characteristics to evaluate their combined potential in achieving accurate and reliable classification. The primary aim is to evaluate the effectiveness of a radiogenomic approach in classifying lung cancer subtypes and to determine if radiomic features alone can achieve comparable performance. Additionally, the study investigates whether genomic data alone can serve as a reliable input for classification tasks. To identify the most suitable machine-learning approach, three classifiers—Random Forest (RF), Support Vector Machine (SVM), and Logistic Regression—are initially applied to the dataset. The classifier with the highest performance is then used in subsequent analyses involving feature reduction. Three advanced feature reduction methods—LASSO (Least Absolute Shrinkage and Selection Operator), PCA (Principal Component Analysis), and mRMR (Minimum Redundancy Maximum Relevance)—are employed to refine the dataset by selecting the most relevant features and further optimizing classification accuracy. The performance is assessed across three scenarios: radiomic, genomic, and radiogenomic approaches. The results indicate that the radiogenomic model outperforms the radiomic approach with AUC 95,1% against 52,7% after normalization. Furthermore, combining three feature reduction methods significantly enhances the classification accuracy, confirming their utility in identifying the most informative attributes from the dataset with the radiogenomic approach reaching an AUC of 98,9% and the radiomic an AUC of 73,6%. The genomic approach showcases AUC 97,3% highlighting the importance of the RNA-Sequencing data and proving that genomic data can be a standalone input. This dual investigation contributes not only to identifying the optimal approach for histological classification in lung cancer but also recommends an improved methodology for feature selection in complex datasets. The findings have significant implications for advancing the role of radiogenomics in precision oncology and highlight the potential of using only genomics in machine learning to enhance cancer diagnostics and patient stratification.