• Lakeside Upper School, Seattle, Washington
• Radcliffe Department of Medicine, University of Oxford, Oxford, United Kingdom

Lung cancer remains the most lethal form of cancer, primarily due to late-stage diagnoses. Early detection significantly improves survival rates, yet it remains challenging. This study aims to enhance early lung cancer diagnosis by developing and evaluating three models: a Multi-Layer Perceptron (MLP) for clinical data, a Convolutional Neural Network (CNN) for imaging data, and a hybrid model combining both data types. We hypothesized that integrating clinical and imaging data would yield higher diagnostic accuracy than single-modality approaches. Using the U.S. National Institute of Health’s Prostate, Lung, Colorectal, and Ovarian Cancer Screening Trial dataset, which includes over 100,000 chest X-rays and associated clinical records, we preprocessed the data and balanced the distribution of positive and negative samples to train and evaluate our models. The hybrid model achieved the highest accuracy (71.58%), performing slightly better than the MLP (70.88%) and more notably better than the CNN (58.25%), suggesting that multi-modality integration may offer added value under certain conditions. Future research should consider adding other data sources such as genetic and environmental factors to enhance the model's performance further. These f indings underscore the promise of multi-modality approaches in transforming lung cancer diagnostics, potentially leading to earlier detection and improved patient outcomes.