Abstract LB100: Automated feature reduction approaches for lung cancer detection through radiomic analysis

Abstract Over the years, cancer has increasingly become a global health issue, with early detection and diagnosis being critical for successful treatment. Radiomics, which utilizes imaging techniques such as CT, PET, MRI, or Ultrasound, involves extracting features from image-based data and applying ma- chine learning for quantitative analysis and disease prediction. These extracted features are divided into three categories namely first, second and higher order radiomics resulting in a massive number of features that necessitate the use of feature engineering to address the curse of dimensionality. This research focuses on exploring dimension reduction techniques to decrease the number of features and make the data more manageable. The radiomics workflow includes imaging, segmentation, feature extraction, and analysis. In this study, a large-scale CT dataset for lung cancer diagnosis, collected by scholars from a medical university in China, is utilized to demonstrate dimension reduction techniques using R, SAS, and Python software. The techniques employed include PCA, clustering, and manifold-based algorithms. These methods successfully reduced the number of features from 110 to 39 using PCA and to 21 with clustering. Overall, the analysis highlighted that texture-based features, particularly those in the second-order radiomics category, are the most significant in the lung cancer data. Citation Format: Mostafa Zahed. Automated feature reduction approaches for lung cancer detection through radiomic analysis [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2025; Part 2 (Late-Breaking, Clinical Trial, and Invited Abstracts); 2025 Apr 25-30; Chicago, IL. Philadelphia (PA): AACR; Cancer Res 2025;85(8_Suppl_2):Abstract nr LB100.