(주) 뷰노

ESC congress 2025
-

AI-driven analysis of treadmill test ECG: clinical utility and performance evaluation

  • Sep. 2025
  • by Young Hoon Ji et. al.

Background/Introduction
The Treadmill Test (TMT) is a widely utilized non-invasive diagnostic tool for identifying exercise-induced cardiac ischemia in patients presenting with angina. Accurate interpretation of TMT data is essential for detecting subtle ECG changes indicative of ischemia. However, clinician's diagnostic process can be time-intensive and subject to variability due to the complexity of ECG patterns and differences in clinician expertise. Recent advancements in artificial intelligence (AI) have demonstrated promising solutions for addressing these challenges by enabling rapid and accurate analysis of ECG data.

Purpose
This study aims to evaluate the performance of an AI model in analysing ECG changes during TMT and to compare its diagnostic accuracy with that of cardiologists' interpretations as ground truth.

Methods
We used the PTB-XL dataset, which includes ECG data labeled for ST-segment changes (ST-T changes and non-specific ST changes), to train a model for detecting these changes (positive: 2,954; negative: 18,883). The dataset was split into train, validation, and test sets in a 7:1:2 ratio, and the model was trained using the SE-ResNet152. Evaluation was conducted on TMT data acquired from 10 patients (positive: 5; negative: 5). The AI model classifies TMT-positive cases when the average output value during the exercise phase exceeds a cutoff value determined by the maximized Youden's J statistics from the validation set during model development. The model performance was evaluated using sensitivity, specificity, positive predictive value, and negative predictive value. The results were evaluated using cardiologists' interpretations as ground truth.

Results
The AI model achieved a sensitivity of 1.0, a specificity of 0.8, a positive predictive value of 0.83, and a negative predictive value of 1.0.

Conclusion
This study presents the clinical utility of an AI-based ECG analysis model in interpreting TMT data for detecting ST-segment changes in patients with angina. When compared with cardiologists' interpretations, the AI model demonstrates high sensitivity and comparable specificity. The AI model demonstrates the potential to streamline the interpretation process, minimizing time demands and reducing variability inherent to human analysis. Furthermore, these findings suggest the potential to enhance the efficiency of the traditionally clinician-dependent diagnostic process by integrating an AI model into the TMT procedure. Further validation using larger and more diverse datasets is necessary to confirm its generalisability and enhance its clinical applicability.

Link

-

Author

Young Hoon Ji¹, Yeongyeon Na¹, Sunghoon Joo¹, Gihyeon Seo¹, Jin Hwa Lee², Myoung Soo Park²

Tag in

#biosignal