A Novel Hybrid Deep Learning Method for Early Detection of Lung Cancer Using Optimized Deep Neural System

Abstract

Lung cancer remains one of the deadliest cancers in worldwide, largely because it is often diagnosed late, when treatment options are limited. Computed Tomography (CT) imaging supports early screening, but manual interpreting these scans manually is slow and produces inconsistent results across different radiologists. A hybrid deep learning framework that pairs a 3D convolution neural networks (3D-CNN), responsible for extracting spatial patterns, with an recurrent neural networks (RNN) for modeling contextual dependencies across CT slices, capturing both shape-level structure of a nodule and how that structure evolves through the scan sequence. The model is carried out on a set of 3,200 grayscale CT images drawn from a Federated Learning dataset, with every scan resized to 224×224 pixels and standardized ahead training. The hybrid method is performance with standalone 3D-CNN, RNN, 3D-CNN+RNN, vision transformer (ViT), and s upport vector machine (SVM), with Accuracy, Precision, Recall, and F1-Score as the comparison metrics . The results show that the combined 3D-CNN+RNN model reaches the top classification accuracy is 95%, outperforming the baseline models.