E1.M2

DATASET DESCRIPTION

Disease Area

Thoracic Oncology

Specific Research Focus

Lung Cancer Early Detection in Lifelong Never-Smokers

Dataset Theme / Edition 

Never-Smoker Lung Cancer — Cross-Sectional + Longitudinal Multimodal Dataset

Short Dataset Summary

10,000 cross-sectional patient records + 120,000+ longitudinal visit rows for binary lung cancer classification in never-smokers (pack-year < 1). Encodes CT imaging findings, Lung-RADS risk scores, EGFR/ALK/ROS1 driver mutations, liquid biopsy biomarkers, environmental carcinogen exposures (radon, air pollution, cooking fuel), and 5-archetype temporal disease trajectories. Prevalence calibrated to 10%, consistent with never-smoker LC incidence in mixed environmental/pulmonology screening populations. 

Intended Research Applications

LC risk classification · Early nodule detection AI · Lung-RADS model benchmarking · EGFR/ALK/ROS1 biomarker research · Environmental exposure analytics · Temporal sequence modelling (LSTM/Transformer) · TimeGAN synthetic data generation · SHAP attribution studies 

Total Number of Patients

10,000 cross-sectional · 120,114 longitudinal visit rows

Number of Features / Parameters

51 total — 32 cross-sectional (demographics · exposures · symptoms · CT imaging · genomics · outcomes) + 19 temporal sequence features per visit 

Data Modalities Included

Structured EHR-style tabular · CT imaging metadata (findings, nodule size, Lung-RADS) · Genomic/molecular markers (EGFR, ALK, ROS1) · Liquid biopsy composite score · Environmental exposure indices · Longitudinal visit sequences · Padded NumPy tensors (10,000 × 18 × 17) with attention masks 

Recommended AI Tasks

 Binary classification · Sequence modelling (LSTM, Transformer) · TimeGAN temporal synthetic generation · Survival analysis · Multi-label biomarker prediction · Federated oncology model training 

Dataset Size

34 MB longitudinal CSV · 14 MB dynamic features CSV · 1.4 MB static features CSV · 12 MB NumPy tensor (.npy) · 3 KB metadata JSON 

Release Date

May 2026 

Why This Dataset Exists

 No publicly available synthetic dataset exists specifically for never-smoker lung cancer with Lung-RADS-aligned CT features, EGFR/ALK/ROS1 mutation profiles with ethnic enrichment, environmental carcinogen pathways, and longitudinal temporal trajectory sequences — the combination required to train clinically meaningful early detection AI for this underserved population. 

Real-World Problem Addressed

Never-smoker lung cancer accounts for 10–25% of all lung cancer cases globally and disproportionately affects women and East Asian populations through distinct molecular pathways. These patients are frequently missed by standard smoking-history-based screening criteria and lack dedicated AI tools trained on their specific clinical presentation — leading to late-stage diagnosis in a population where early detection dramatically improves survival.