Claudia I. Henschke

Maria Mahbub, Robert J. Klein, Myvizhi Esai Selvan et al.

Lung cancer (LC) is the third most common cancer and the leading cause of cancer deaths in the US. Although smoking is the primary risk factor, the occurrence of LC in never-smokers and familial aggregation studies highlight a genetic component. Genetic biomarkers identified through genome-wide association studies (GWAS) are promising tools for assessing LC risk. We introduce HEMERA (Human-Explainable Transformer Model for Estimating Lung Cancer Risk using GWAS Data), a new framework that applies explainable transformer-based deep learning to GWAS data of single nucleotide polymorphisms (SNPs) for predicting LC risk. Unlike prior approaches, HEMERA directly processes raw genotype data without clinical covariates, introducing additive positional encodings, neural genotype embeddings, and refined variant filtering. A post hoc explainability module based on Layer-wise Integrated Gradients enables attribution of model predictions to specific SNPs, aligning strongly with known LC risk loci. Trained on data from 27,254 Million Veteran Program participants, HEMERA achieved >99% AUC (area under receiver characteristics) score. These findings support transparent, hypothesis-generating models for personalized LC risk assessment and early intervention.

Sergei V. Fotin, David F. Yankelevitz, Claudia I. Henschke et al.

Candidate generation, the first stage for most computer aided detection (CAD) systems, rapidly scans the entire image data for any possible abnormality locations, while the subsequent stages of the CAD system refine the candidates list to determine the most probable or significant of these candidates. The candidate generator creates a list of the locations and provides a size estimate for each candidate. A multiscale scale-normalized Laplacian of Gaussian (LoG) filtering method for detecting pulmonary nodules in whole-lung CT scans, presented in this paper, achieves a high sensitivity for both solid and nonsolid pulmonary nodules. The pulmonary nodule LoG filtering method was validated on a size-enriched database of 706 whole-lung low-dose CT scans containing 499 solid (>= 4 mm) and 107 nonsolid (>= 6 mm) pulmonary nodules. The method achieved a sensitivity of 0.998 (498/499) for solid nodules and a sensitivity of 1.000 (107/107) for nonsolid nodules. Furthermore, compared to radiologist measurements, the method provided low average nodule size estimation error of 0.12 mm for solid and 1.27 mm for nonsolid nodules. The average distance between automatically and manually determined nodule centroids were 1.41 mm and 1.43 mm, respectively.