Pretreatment Radiation Esophagitis Prediction Using Quantum Machine Learning in Patients With Esophageal Cancer.

Area of Science:

• Quantum computing applications in medicine
• Machine learning for predictive diagnostics
• Radiotherapy and oncology

Background:

• Radiation esophagitis (RE) is a common side effect of radiotherapy for esophageal cancer (EC).
• Accurate prediction of RE is crucial for optimizing treatment plans and improving patient outcomes.
• Current predictive models may not fully leverage complex dosimetric and clinical data.

Purpose of the Study:

• To introduce and validate a novel hybrid quantum-classical machine learning (QML) approach.
• To assess the feasibility and accuracy of QML for predicting pretreatment radiation esophagitis (RE) in esophageal cancer (EC) patients.
• To integrate quantum features with dosiomic and clinical data for enhanced prediction.

Main Methods:

• Developed a hybrid QML model incorporating quantum features (Q), dosiomic features (D), and clinical factors (C).
• Extracted quantum features using quantum convolutional neural networks (Q-CNN) with quantum attention mechanisms.
• Fused features (Q+D+C) and used a Random Forest classifier for RE prediction in 218 training/internal validation and 55 external validation patients.

Main Results:

• The hybrid QML model (Q-CNN+Q-Attention+D+C) achieved optimal performance with an external validation accuracy of 0.83 and AUC of 0.83.
• Feature fusion significantly improved prediction accuracy compared to using quantum features alone.
• The model demonstrated high sensitivity and specificity across training, internal, and external validation sets.

Conclusions:

• The proposed QML method offers a novel and accurate approach for pretreatment RE prediction.
• Integrating quantum amplitude encoding, quantum attention, and multimodal feature fusion enhances prediction reliability.
• This QML model shows significant potential for clinical application in personalized radiotherapy for EC patients.