Logo image
Optimized feature selection of nodular and peri‑nodular radiomics for lung cancer risk prediction with ensemble artificial neural networks
Conference proceeding

Optimized feature selection of nodular and peri‑nodular radiomics for lung cancer risk prediction with ensemble artificial neural networks

Kevin Knoernschild, Kimberly E. Schroeder, Jacob Kitzmann, Chrissy Lusk, Christine Neslund-Dudas, Ann G. Schwartz and Jessica C. Sieren
Vol.13926, pp.139261Y-139261Y-5
Progress in Biomedical Optics and Imaging
04/02/2026
DOI: 10.1117/12.3085592

View Online

Abstract

Lung cancer is responsible for 22% of all cancer related deaths, highlighting the need for screening high-risk individuals using Low Dose Computed Tomography (LDCT) to enhance early detection. Around 97% of nodules identified during lung cancer screening are benign, meaning there is a need for effective tools to differentiate malignant from benign cases. Additionally, incorrectly flagging nodules as suspicious during evaluation of LDCT images can lead to downstream complications from repeated imaging, invasive biopsy, and overall stress to the patient. We hypothesized that a machine learning method utilizing quantitative features extracted from the nodule and surrounding parenchymal tissue of LDCT images can assist in classifying nodule malignancy. Our approach explores predictive performance of ensembles of Artificial Neural Networks trained on important radiomic features using LASSO, ELASTICNET, and MRMRE feature selection methods.

Details

Metrics

8 Record Views
Logo image