Quantum AI for psychiatric diagnosis: enhancing dementia classification with quantum machine learning.

Area of Science:

• Quantum Machine Learning (QML)
• Medical Imaging Analysis
• Artificial Intelligence in Healthcare

Background:

• Early dementia detection is crucial for effective patient management, necessitating highly accurate classification methods.
• Deep learning (DL) models handle large datasets, while quantum machine learning (QML) offers enhanced computational speed and data storage using qubits.
• QML leverages quantum computing principles to potentially improve machine learning model efficiency and accuracy, especially for complex pattern recognition in imaging.

Purpose of the Study:

• To propose and evaluate a hybrid quantum-classical convolutional neural network (QCNN) for accurate dementia classification using MRI data.
• To investigate the efficacy of integrating quantum feature extraction with classical deep learning and knowledge distillation (KD) for improved performance.
• To develop a scalable and efficient framework for dementia classification applicable in clinical settings.

Main Methods:

• A hybrid QCNN framework was developed, processing MRI images through pre-processing, region of interest (ROI) extraction, and quantum feature mapping.
• Pixel values from image patches were encoded as qubits and processed by a parameterized quantum circuit (PQC) to generate quantum features.
• A knowledge distillation (KD) framework was employed, using a deeper CNN (teacher) to guide the QCNN (student) for enhanced generalization and feature learning.

Main Results:

• The QCNN without KD achieved high accuracies: 0.9523 (ADNI-1), 0.9611 (ADNI-2), and 0.9412 (OASIS-2).
• With KD, the student QCNN model demonstrated improved sensitivity, reaching an accuracy of up to 0.9978, surpassing existing state-of-the-art methods.
• The hybrid approach showed superior performance compared to traditional ML/DL methods for dementia classification.

Conclusions:

• The proposed hybrid quantum-classical CNN framework offers a highly accurate and efficient method for dementia classification from MRI data.
• The integration of quantum feature extraction and knowledge distillation significantly enhances classification performance and model generalization.
• This QML-based approach shows great promise for advancing early dementia detection and improving clinical patient management.