Strengthening Accuracy of Lung Cancer Diagnosis Through A Novel Multimodal Bayesian Convolutional Neural Network

Lung cancer is a global epidemic, attributed to the general lack of early and effective screening, leading to low survival rates. Machine learning techniques have previously been used in diagnosing lung cancer, yet no comprehensive, integrated method of detecting lung cancer currently exists. This study proposes a novel method of lung cancer diagnosis through a multimodal process by combining both histopathological tissue images and CT scan images to diagnose lung cancer, while also utilizing bayesian layers within the neural network to account for prediction uncertainty. The proposed method achieved a test accuracy of 99.43%, performed better than traditional baseline methods, and was upheld as a robust method for strengthening current lung cancer diagnosis. The scope of the proposed method extends beyond just lung cancer, to other forms of cancer diagnosis and the entire biomedical imaging field.