DeepXplainer: An interpretable deep learning based approach for lung cancer detection using explainable artificial...

Area of Science:

• Artificial Intelligence in Medicine
• Deep Learning for Medical Diagnosis
• Explainable AI (XAI)

Background:

• Artificial intelligence (AI) offers significant potential in healthcare, including diagnosis and forecasting.
• A major limitation of current AI in healthcare is its 'black box' nature, hindering trust and adoption.
• There is a critical need for interpretable AI models that provide both accurate predictions and clear explanations.

Purpose of the Study:

• To introduce DeepXplainer, a novel interpretable hybrid deep learning technique for lung cancer detection.
• To develop a model that not only predicts lung cancer but also explains its predictions.
• To address the distrust in AI by providing transparency in medical AI applications.

Main Methods:

• Developed DeepXplainer, a hybrid model combining a convolutional neural network (CNN) for feature learning and XGBoost for classification.
• Utilized the SHAP (SHapley Additive exPlanations) method for generating local and global explanations of model predictions.
• Trained and evaluated the model on the open-source 'Survey Lung Cancer' dataset.

Main Results:

• DeepXplainer achieved superior performance compared to existing methods on key metrics.
• The model demonstrated high accuracy (97.43%), sensitivity (98.71%), and F1-score (98.08%).
• Explanations were generated for each prediction, enhancing model interpretability at both local and global levels.

Conclusions:

• The proposed DeepXplainer, a hybrid deep learning model (ConvXGB), effectively detects lung cancer with high accuracy.
• The model integrates feature learning, classification, and prediction explanation components.
• DeepXplainer's interpretability can aid clinicians in lung cancer detection and treatment, fostering greater trust in AI tools.