Table 1 Summary of datasets, modalities, and corresponding real-world applications

From: Transforming label-efficient decoding of healthcare wearables with self-supervised learning and “embedded” medical domain expertise

Dataset

Modality

Downstream task

# Class

# Sample

# Channel

Bandpass

Win (s)

Freq (Hz)

CinC1729

ECG

Cardiovascular Disease Classification

4

8528

1

0.5–40

10

125

CPSC30

ECG

Rhythm/Morphology Abnormality Identification

9

6877

1

0.5–40

10

125

MIMIC-III-WDB31

ECG

–

–

111,619

1

0.5–40

10

125

SleepEDF8

EEG

Sleep Staging

5

41,509

2

0.4-30

30

100

Capture2424

IMU

Activity Recognition

6

45,553

3

–

10

100

Simband33,34

PPG

Cardiovascular Disease Classification

4

7590

1

0.5–20

10

50

  1. This table outlines the datasets used in our experiments, the type of wearable modality they represent, and the associated downstream classification tasks. These tasks, based on the expert annotations provided with each dataset, span clinically and behaviorally relevant applications, including cardiac rhythm classification (ECG, PPG), sleep stage detection (EEG), and physical activity recognition (IMU). All tasks are formulated as multi-class classification problems.
Back to article page