Quantum-enhanced multimodal prognostic transformer for skin disease progression prediction and visualization

Area of Science:

• Artificial Intelligence
• Dermatology
• Quantum Computing

Background:

• Accurate classification and staging of skin diseases like monkeypox, chickenpox, and measles are crucial for effective clinical management, particularly in resource-limited settings.
• Current diagnostic methods may lack the precision needed for timely intervention, highlighting a need for advanced AI solutions.

Purpose of the Study:

• To introduce a proof-of-concept Quantum-Enhanced Multimodal Prognostic Transformer (Q-MPT) for joint disease classification and staging.
• To integrate dermoscopic images with patient metadata for enhanced diagnostic and prognostic capabilities.
• To explore the potential of quantum-inspired computation in multimodal AI for dermatology.

Main Methods:

• Developed Q-MPT integrating a Vision Transformer with a metadata fusion pathway and a quantum layer.
• Employed long short-term memory (LSTM) for latent trajectory prediction and a quantum-inspired generative module for simulating disease progression.
• Utilized attention rollouts, Integrated Gradients, and variational autoencoders for model explainability.

Main Results:

• Q-MPT achieved 89.4% accuracy in disease classification and 87.3% in stage prediction on a custom dataset.
• The model outperformed conventional convolutional neural networks (CNNs) and Vision Transformer baselines.
• Demonstrated the feasibility of combining quantum-inspired methods with multimodal learning for dermatological applications.

Conclusions:

• Q-MPT represents a novel framework bridging diagnostic and prognostic AI in dermatology.
• The study highlights the potential of quantum-inspired AI for improving skin disease assessment.
• Further validation on diverse datasets is necessary to establish clinical readiness.