Explainable Quantum Machine Learning: A New Paradigm for Liver Disease Detection Integrating XAI Technologies

This article introduces an open-source project that implements liver disease detection using hybrid quantum-classical neural networks and integrates explainable AI (XAI) technology to address the black-box problem in medical AI. The project uses feature ablation experiments to verify the scientific validity of model explanations, providing a new paradigm for the clinical implementation of AI in healthcare.

Artificial intelligence has great potential in medical diagnosis, but the black-box nature of deep learning hinders its clinical implementation. Quantum machine learning (QML) leverages quantum parallelism to process complex data, yet its own complexity exacerbates the interpretability challenge. Balancing quantum advantages with model transparency has become a core issue.

Hybrid Quantum-Classical Architecture

• Classical Layer: TensorFlow/Keras handles data preprocessing and feature extraction
• Quantum Layer: PennyLane implements parameterized quantum circuits to explore high-dimensional feature patterns

Multi-Dimensional XAI Technologies

• Kernel SHAP: Calculates marginal contributions of features
• Integrated Gradients: Quantifies the impact of features on predictions
• Permutation Feature Importance: Verifies the actual importance of features Cross-validation ensures the reliability of explanations.

Feature Ablation Experiments

Experimental Condition	Accuracy
Baseline Model	~85.62%
After Removing Key Features	~58.90%
The accuracy dropped by 27%, confirming that the model relies on real medical features.

Key Predictive Factors

Identified direct bilirubin and alkaline phosphatase as core factors, which are highly consistent with clinical medical indicators, enhancing the model's credibility.

Model Generalization

The training and validation loss curves are consistent, with low overfitting risk and good generalization ability.

Technology Stack

Component	Technology Selection	Purpose
Language	Python 3.10	Development Language
Classical DL	TensorFlow 2.15	Model Construction
Quantum Computing	PennyLane	Quantum Circuit Implementation
XAI	SHAP/Integrated Gradients	Explanatory Analysis

Project Structure

Modular design: data/ (dataset), models/ (model definitions), xai/ (explanation modules), etc., with a clear and reproducible workflow.

Current Limitations

1. Small dataset size (ILPD has about 580 entries)
2. Based on quantum simulators, not verified on real hardware
3. Requires more clinical validation

Future Directions

• Expand large-scale medical datasets
• Explore implementation on real quantum processors
• Integrate imaging/genomic data
• Develop interactive explanation interfaces for doctors

This project demonstrates the potential of combining quantum ML and XAI, ensuring the scientific validity of explanations through feature ablation validation. This "verifiable interpretability" provides an important reference for the clinical implementation of AI in healthcare and is of reference value to researchers and developers.