Panic Attack Prediction for Patients With Panic Disorder via Machine Learning and Wearable Electrocardiography...

Area of Science:

• Digital psychiatry
• Affective computing
• Artificial intelligence in mental health

Background:

• Panic attack prediction is challenging due to variable physiological responses and subjective assessments.
• Current methods lack accuracy in identifying panic episodes.
• There's a need for objective, real-time monitoring solutions.

Purpose of the Study:

• To develop a multimodal deep learning framework for enhanced panic attack prediction.
• To integrate real-time physiological signals (ECG) with psychological data.
• To improve the accuracy and reliability of panic attack detection systems.

Main Methods:

• Adapted ConvNetQuake architecture for temporal ECG pattern extraction.
• Pretrained and fine-tuned the model on extensive ECG datasets.
• Incorporated psychological assessments (DSM-IV, PDSS) as auxiliary inputs.
• Evaluated the multimodal framework using standard performance metrics.

Main Results:

• The model achieved 71.43% accuracy, 83.72% precision, 70.59% recall, and 76.60% F1 score.
• Detected heart rate variability anomalies linked to panic episodes.
• Multimodal integration significantly outperformed unimodal approaches in prediction reliability.

Conclusions:

• Wearable-based early warning systems for panic attacks are feasible.
• The approach supports just-in-time digital interventions.
• Wearable AI holds significant potential for advancing digital psychiatry and affective computing.