A Multi-Objective Hybrid Deep Learning Framework for Pulmonary Nodule Detection in Lung Cancer CT Imaging with Enhanced Diagnostic Reliability

Abstract

The detection of pulmonary nodules on the CT imaging of lung cancer is a critical but difficult issue because of high variability in nodule size, shape and appearance and false positive in the traditional computer-aided diagnosis systems. The present study suggests a multi-objective hybrid deep learning model capable of improving the accuracy of detection, the reliability of diagnostics and clinical applicability. The aim is to jointly optimize the sensitivity, specificity and computational efficiency based on an integrated architecture that combines Convolutional Neural Networks (CNN), Vision Transformers (ViT) and Long Short-term Memory (LSTM) networks. The framework involves extraction of multi-scale features, attention, and temporal context modeling and optimizes the hyperparameters with a hybrid metaheuristic algorithm combining a combination of both the Grey Wolf Optimizer (GWO) and Particle Swarm Optimization (PSO) algorithms. They are tested on benchmark Lung Nodule Dataset, and have a comprehensive preprocessing and augmentation methods. The proposed model has an accuracy of 97.6, sensitivity of 96.8, specificity of 98.1 and AUC of 0.985 which is higher than the baseline CNN (93.2% accuracy) and standaloneViT (94.5% accuracy) models. The probability of false positives is lowered by a factor of 21.4% and the latency of detection is elevated by 18.7% in comparison with current methods. The newness is the multi-objective optimization approach along with the hybrid deep learning and feature fusion based on attention which makes it powerful in terms of its performance in heterogeneous datasets. The framework has good generalization and scalability to real-life clinical implementation.The proposed solution has a great potential of improving pulmonary nodule detection through providing high accuracy, reliability and computationally efficient diagnostic support in the early detection of lung cancer.