OncoScan3D: Lung Tumor Detection with ResUNet

The diagnosis and treatment of lung diseases in Computed Tomography images represent an essential task, particularly concerning the segmentation of the lung and its tumors. The challenges in this domain arise from the uneven distribution, indistinct borders, varied densities, as well as diverse shapes and sizes of lesions. Addressing these complexities is pivotal for minimizing the time and energy expended in the diagnosis of lung diseases. The current study primarily concentrates on employing deep learning algorithms for the segmentation of the lung and its tumors from abdominal Computed Tomography scan images. The algorithm utilized is rooted in the modified Deep Residual UNET architecture and automated semantic segmentation and Convolutional Neural Networks to separate the lung from Computed Tomography data and detect lesions within the segmented lung region. The overall methodology to utilize this architecture performs quite well with a Dice Similarity Coefficient of 96.35% and a lung segmentation rate of 89.28% resulting in an overall accuracy of 99%. These significant results from abdomen Computed Tomography help in early stage detection and allow more precise clinical attention of lung disorders.