Related Experiment Video
Updated: Jun 6, 2025
08:10
Estimating Bilateral Atrial Function by Cardiovascular Magnetic Resonance Feature Tracking in Patients with Paroxysmal Atrial Fibrillation
Published on: July 20, 2022
1.6K
Using Machine Learning to Predict the Duration of Atrial Fibrillation: Model Development and Validation.
Satoshi Shimoo1, Keitaro Senoo1,2, Taku Okawa1
1Department of Cardiac Arrhythmia Research and Innovation, Graduate School of Medical Science, Kyoto Prefectural University of Medicine, Kyoto, Japan.
JMIR Medical Informatics
|November 22, 2024
Summary
Machine learning models accurately predict atrial fibrillation (AF) duration, improving cardiologist diagnosis. However, clinicians showed limited reliance on AI predictions, even when aware of their diagnostic limitations.
Area of Science:
- Cardiology
- Artificial Intelligence
- Medical Diagnostics
Background:
- Atrial fibrillation (AF) is a progressive condition with clinical types based on duration: paroxysmal, persistent (PeAF; <1 year), and long-standing persistent (≥1 year).
- AF duration is a critical risk factor for recurrence after catheter ablation, influencing treatment strategies for PeAF.
Purpose of the Study:
- To enhance cardiologists' accuracy in diagnosing AF duration.
- To develop and validate a machine learning (ML) model for predicting AF duration.
Main Methods:
- A dataset of 189 patients with PeAF was used, with 145 for training an ML model and 44 for testing.
- Two groups of cardiologists (A and B) evaluated AF duration in test data, first independently, then with ML model predictions provided.
- Group B was also informed about diagnostic limitations before the second evaluation stage.
Main Results:
- The ML model achieved 81.8% accuracy in predicting AF duration on test data (72% sensitivity, 89% specificity).
- Cardiologists' correct diagnosis rates improved significantly from phase 1 to phase 2 in both groups (Group A: 63.9% to 71.6%; Group B: 59.8% to 68.2%).
- Despite improvements and awareness of limitations, cardiologists' disagreement with ML predictions remained substantial (17.3% in Group A, 20.9% in Group B).
Conclusions:
- ML models can significantly improve cardiologists' diagnostic accuracy for AF duration.
- Cardiologists demonstrated a tendency to not fully trust ML model predictions, even when aware of their own diagnostic limitations.
