Early-Stage Multimodal Lung Cancer Detection Using Attention-Based Feature Fusion

Early-stage lung cancer diagnosis is often hindered by low contrast, subtle tumour morphology, and incomplete feature representation across imaging modalities. To overcome these limitations, we present the Multi-Stage Cross-Element Attention Quantum Convolutional Transformer with Group Aggregation Network (MSEGA-Net), an advanced multitask framework designed for segmentation and classification of multimodal lung cancer images. MSEGA-Net incorporates PNSTG-UNet++ for adaptive contrast enhancement and artefact suppression, Dual-Path Vision Transformers to capture comprehensive global–local dependencies, and a quantum-inspired convolutional layer for efficient high-dimensional feature encoding. The cross-element attention module facilitates fine-grained inter-modal fusion, while group aggregation enhances semantic consistency and long-range contextual reasoning essential for accurate early lesion identification. Extensive evaluations conducted on three benchmark datasets demonstrate that MSEGA-Net achieves existing performance with average classification accuracies ranging from 99.20% to 99.85%, as reported in the corresponding result tables, with statistical significance confirmed using ANOVA and Friedman tests (p ≤ 0.001). Furthermore, model interpretability is enhanced through Grad-CAM++–based visual explanations and LLM-assisted clinical reasoning, supporting transparent and trustworthy clinical decision support.Overall, MSEGA-Net offers a robust, generalizable, and explainable solution for early lung cancer detection, showcasing significant potential for real-world clinical integration.