A hybrid AI method for lung cancer classification using explainable AI techniques.
Resham Raj Shivwanshi1, Neelam Shobha Nirala1
1Department of Biomedical Engineering, National Institute of Technology Raipur, 492010, India.
Summary
This study introduces a hybrid AI model for lung cancer detection in CT scans, achieving high accuracy. Explainable AI helps understand model decisions for improved computer-assisted diagnosis systems.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Oncology
Background:
- Computer-assisted diagnosis (CAD) systems for lung cancer (LC) analysis of CT images face challenges due to complex image structures and abnormality locations.
- Extracting relevant information for LC CAD systems is difficult, hindering effective model development.
Purpose of the Study:
- To present a hybrid AI method for lung cancer malignancy classification to aid researchers in engineering model performance.
- To improve the decision-making process of LC CAD systems through explainable AI.
Main Methods:
- The IncCat-LCC: Explainer methodology combines handcrafted radiomic features (HcRdF), InceptionNet CNN features (INCF), and Vision Transformer features (ViTF) for extraction.
- Feature selection is performed using XGBOOST (XGB), followed by GPU-based CATBOOST (CB) classification.
Main Results:
- The framework achieved high performance in lung nodule multiclass malignancy classification.
- Key metrics include accuracy (96.74%), precision (93.68%), recall (96.74%), F1-score (95.19%), specificity (98.47%), and AUC (99.76%) for the normal class.
Conclusions:
- Explainable AI (XAI) provides insights into model performance and statistical outcomes, aiding reader comprehension.
- This work can enhance existing LC CAD systems by using XAI to identify factors contributing to improved performance and reliability.
Keywords:
CATBOOST classifierExplainable AI (XAI)Lung cancer CADRadiomic featuresReliable cancer assessmentVision transformer
