Oxynet

AI-Powered Exercise Threshold Detection

Harness the power of deep learning to identify respiratory-based exercise intensity thresholds with expert-level accuracy. Trained on over 1,600 crowdsourced cardiopulmonary exercise tests from accredited exercise physiology experts.

Get Started with Oxynet

Upload your own data files or try it out with our prepared example datasets to explore Oxynet's capabilities.

I already have my CPET file

Upload your own cardiopulmonary exercise testing data files to analyze with Oxynet's AI-powered threshold detection.

Go to Analyze Page

I do not have a CPET file

Select from our prepared example datasets to explore Oxynet's capabilities and see how it identifies exercise thresholds.

Key Features of Oxynet

Discover what makes Oxynet a powerful tool for exercise threshold detection.

Instant threshold detection

Get automated identification of respiratory-based exercise intensity thresholds in seconds

Expert-level accuracy

Trained on over 1,600 cardiopulmonary exercise tests analyzed by experienced exercise physiologists

Confidence scores

Receive probability estimates for each threshold to guide your clinical and research decisions

Educational tool

Compare AI predictions with your own assessments and expert annotations to enhance learning

Research-validated

Performance validated in peer-reviewed research with mean bias <50 mL·min⁻¹ compared to expert consensus

How Oxynet Works

Oxynet uses a deep learning convolutional neural network trained on crowdsourced expert interpretations to automatically identify ventilatory thresholds from cardiopulmonary exercise test data.

Deep Learning Architecture

Oxynet employs a convolutional neural network that processes 40-second sliding windows of respiratory data. The algorithm analyzes patterns across five key variables—oxygen uptake (V̇O₂), carbon dioxide output (V̇CO₂), minute ventilation (V̇_E), end-tidal O₂ (P_ETO₂), and end-tidal CO₂ (P_ETCO₂)—to classify exercise intensity into three domains: moderate (below first threshold), heavy (between thresholds), and severe (above second threshold).

V̇O₂

V̇CO₂

V̇_E

P_ETO₂

P_ETCO₂

Probability of being in the moderate domain (< θ_LT)

Probability of being in the heavy domain (θ_LT-RCP)

Probability of being in the severe domain (> RCP)

Probability Distribution by Intensity Domain

Oxynet outputs the probability of each data point to be in one of the three exercise intensity domains: moderate (below first threshold), heavy (between thresholds), and severe (above second threshold). These probability distributions are visualized across the exercise intensity spectrum in the chart below.

The estimated lactate threshold (θ_LT) and respiratory compensation point (RCP) are identified at the points where the probability curves intersect—specifically, where the probability of being in the moderate domain drops below 50% (θ_LT) and where the probability of being in the severe domain exceeds 50% (RCP).