Grad-CAM Enhanced Explainable Deep Learning for Multi-Class Lung Cancer Classification Using DE-SAMNet Model.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Lung cancer (LC) is a leading cause of cancer mortality globally.
• Manual interpretation of chest CT scans for LC diagnosis is time-consuming and variable.
• Automated diagnostic tools are needed to improve accuracy and efficiency.

Purpose of the Study:

• To develop and evaluate DE-SAMNet, a hybrid deep learning framework for automated multi-class lung cancer classification from CT scans.
• To assess the model's performance on public and private clinical datasets.
• To enhance the interpretability of the automated classification using explainable AI (XAI).

Main Methods:

• DE-SAMNet integrates DenseNet121 and EfficientNetB0 for multi-scale feature extraction.
• A Spatial Attention Module (SAM) refines feature representation by focusing on clinically relevant regions.
• A compact fusion mechanism combines features for final classification.

Main Results:

• The model achieved high performance on a public dataset (99.54% accuracy) and a private dataset (95.96% accuracy).
• DE-SAMNet outperformed existing approaches in lung cancer classification.
• XAI techniques (Grad-CAM) visualized decision-making, highlighting lesion-specific regions for transparency.

Conclusions:

• DE-SAMNet provides a highly accurate and interpretable solution for automated lung cancer detection.
• The explainability features enhance trust and demonstrate clinical potential for early diagnosis.
• The framework addresses challenges in manual CT scan interpretation, improving diagnostic consistency.