Explainable Decision Support System for Lung Cancer Diagnosis from CT Images Using Hybrid AI Models

Abstract

Lung cancer is one of the main causes of death around the world, so early detection is crucial. Diagnosing small lung nodules can take a lot of time and require skilled radiologists. This research presents an AI-based system for classifying lung cancer using CT images, focusing on its performance and ease of understanding. The study combines carefully crafted features—including texture, shape, intensity, and wavelet descriptors—with deep features from various deep learning models, such as CNNs, CoAtNet, and EfficientNet. It highlights the most important attributes through feature selection and fusion strategies. Different machine learning and hybrid deep learning models were tested. In the end, the best-performing model with explainable AI was combined to give radiologists insights into how the model makes decisions. Experiments were conducted on 30,020 CT images from the BOWL2017 dataset, comprising 790 patients. Up to 98% accuracy was noted, with 98.4% precision and 98.0% recall. Radiologists confirmed that the model's attention maps match clinically relevant patterns, which improve efficiency. This framework demonstrates how interpretable AI can improve CT-based lung image classification and support radiologists in their decision-making.