Encoding Cardiopulmonary Exercise Testing Time Series as Images for Classification using Convolutional Neural...

Area of Science:

• Cardiology
• Pulmonology
• Biomedical Engineering
• Artificial Intelligence

Background:

• Exercise testing, particularly Cardiopulmonary exercise testing (CPX), is a vital diagnostic tool for cardiovascular and pulmonary diseases.
• Traditional CPX analysis simplifies complex time-series data, potentially losing valuable trend information.
• Advancements in AI and technology offer new avenues for analyzing intricate physiological data.

Purpose of the Study:

• To develop a novel framework for analyzing multivariate exercise testing time-series data.
• To utilize image encoding techniques (Gramian Angular Field, Markov Transition Field) with convolutional neural networks (CNNs) for disease classification.
• To enhance the interpretability of AI-driven predictions in cardiovascular disease diagnosis.

Main Methods:

• Encoding time-series data from exercise testing into images using Gramian Angular Field (GAF) and Markov Transition Field (MTF).
• Applying a convolutional neural network (CNN) with attention pooling for classifying heart failure and metabolic syndrome patients.
• Utilizing GradCAMs to visualize and identify discriminative features contributing to model predictions.

Main Results:

• The proposed framework successfully processes multivariate exercise testing time-series data.
• Accurate prediction of cardiovascular diseases was achieved using the image-encoded data and CNN model.
• Interpretable Grad-CAMs were generated, highlighting the features driving the model's predictions.

Conclusions:

• The novel framework effectively analyzes complex CPX time-series data, overcoming limitations of traditional methods.
• This AI-driven approach offers accurate and interpretable predictions for cardiovascular diseases.
• The method holds significant potential for improving the diagnosis and management of patients with cardiovascular conditions.