TROIKA

Dataset Overview

Dataset Name

• TROIKA
• Cardiac Arrhythmia Suppression Trial (CAST)

Dataset Description

TROIKA

• Source: Dataset proposed in Zhang and collaborators' 2015 paper for heart rate monitoring.
• Content: PPG and three‑dimensional (x, y, z) accelerometer signals from 12 male participants (ages 18–35). Data were collected while subjects performed rest, walking, running, and cooling on a treadmill. Signal sampling frequency is 125 Hz.
• Additional Information: Includes reference heart rate values calculated from ECG signals as ground truth.
• Limitations: Data are only from males, with a limited age range, and do not consider skin tone variations, which may not reflect the broader population characteristics.

CAST

• Source: Data collected from the Cardiac Arrhythmia Suppression Trial sponsored by the National Institute of Cardio‑Pulmonary and Blood Research.
• Content: 24‑hour heart rate recordings from patients who have experienced myocardial infarction. The data were smoothed and resampled to simulate PPG pulse‑rate data obtained from wrist‑worn devices.
• Citation: Study by Stein PK et al., 2000.

Dataset Applications

TROIKA

• Development and testing of pulse‑rate algorithms that estimate pulse rate from PPG signals and tri‑axial accelerometer data.

CAST

• Clinical applications aimed at computing clinically meaningful metrics and discovering medical trends.

Algorithm Description

• Principle: Detects blood volume changes in vessels via a PPG sensor based on blood motion.
• Function: Estimates pulse rate from PPG and accelerometer signals, outputting a confidence measure.
• Output: Includes the mean absolute error between predicted and true heart rates and a confidence score.
• Limitations: The algorithm may mistakenly interpret user movement as pulse rate, leading to estimation errors.