Lung cancer is among the leading causes of cancer-related mortality worldwide, and early detection from CT images is critical for improving patient outcomes. Manual interpretation of CT scans is time-consuming, subjective, and susceptible to inter-observer variability, necessitating automated and reliable computer-aided diagnosis systems.

This study proposes HALI-Net, a hybrid attention-based deep learning framework for multiclass lung cancer classification from CT scan images. The framework integrates adaptive Gaussian-based preprocessing, handcrafted texture feature extraction using Gray Level Co-occurrence Matrix (GLCM) and Local Binary Pattern (LBP), and deep feature extraction via InceptionV3 enhanced with a Multi-Scale Channel Attention (MSCA) mechanism. A hybrid feature fusion strategy combines deep and handcrafted representations, which are subsequently classified by an MLP classifier. Gradient-weighted Class Activation Mapping (Grad-CAM) is incorporated to provide visual interpretability. The model is evaluated on the IQ-OTH/NCCD lung cancer dataset comprising three diagnostic classes: Benign, Malignant, and Normal.

HALI-Net achieves a classification accuracy of 98.55%, with strong precision, recall, F1- score, and a ROC-AUC of 99.90%. Ablation studies confirm the individual contribution of each proposed component, and Grad-CAM visualizations demonstrate consistent focus on clinically relevant pulmonary regions across all diagnostic categories.

The proposed HALI-Net framework delivers an accurate, computationally efficient, and clinically interpretable solution for automated lung cancer detection from CT images. Its hybrid architecture and explainability mechanisms make it well-suited for integration into computer-aided diagnostic workflows in clinical environments.