Unveiling Brain Region Patterns in PET/CT scans for Lung Cancer Assessment: A Computational AI Framework

Author(s): Hakan Sat Bozcuk, Ahmet Eren Sen, Mehmet Artac, Bugra Kaya

This study aims to investigate potential differences in brain activity between lung cancer patients and healthy controls at the time of diagnosis, utilizing a computer vision artificial intelligence (AI) model. Participants undergoing evaluation for lung cancer (cases) and with benign pulmonary nodules (controls) underwent Positron Emission Tomography/ Computerized Tomography (PET/CT) scans. Specialized software reconstructed and labeled brain images. A computer vision AI model was developed using EfficientNet B0 through transfer learning, complemented by multivariate discriminant analysis. A total of 84 cases were recruited into the study. The constructed AI model exhibited robust accuracy (internal accuracy=1.0, external sensitivity=0.83) on a subset of cases (52 lung cancer patients, 22 controls). Notably, the right frontal lobe emerged as a crucial discriminator, displaying a 5% reduction in the ratio of frontal lobe to brainstem activity in lung cancer cases (Wilk’s Lambda=0.877, P=0.002). Based solely on PET/CT brain imaging data, our AI model accurately classified lung cancer patients. The distinct role of the right frontal lobe in this study underscores the broader significance, shedding light on brain function disparities at lung cancer diagnosis.