Advanced lung tumor diagnosis using a 3D deep neural network based CAD system

Abstract

In the realm of interdisciplinary research, Computer-Aided Diagnosis (CAD) emerges as a pivotal tool, aiding radiologists in comprehending lung computer tomography scans (both 2D and 3D) for enhanced prediction of lung malignancy and benign cases with heightened accuracy. The objective of clinical examination is to furnish medical experts with more effective scientific insights and treatment strategies. As conventional computer-aided design frameworks prove to be increasingly challenging and time-consuming, this study delves into the contemporary landscape of clinical image analysis in computer-aided diagnosis, with a specific focus on the architecture of the Convolutional Neural Network (CNN). Employing the Kaggle Data Science Bowl dataset, encompassing CT scan images of lung tumors, our proposed CNN architecture has exhibited remarkable performance. Notably, the CNN Net deep learning algorithm achieved a peak classification accuracy of 94%, coupled with a substantial sensitivity and specificity of 90.2%. The extensive 15-fold cross-validation procedure further highlighted the robustness of the CNN network, reporting a minimal log loss of 0.54. This outcome underscores the superior diagnostic capabilities of the CNN architecture in the context of lung tumor diagnosis. In essence, this paper not only contributes to advancements in radio diagnosis and precision medication but also provides a valuable addition to the field by presenting a meticulous figure for computer-aided design utilization.