Contactless Blood Pressure Estimation based on Graph Convolution Network with Semi-supervised Learning
Oral presentation
Research approach
Extracted rPPG with POS and combined a representative pulse segment with 51 blood-pressure features. A two-layer GCN estimates systolic and diastolic pressure from a sample-similarity graph.
View full-size figure
Results
On Vital Videos VV-Medium, which contains 250 participants, rPPG inputs yielded SBP/DBP MAE of 17.42/10.94 mmHg. Skin-color inputs yielded 18.10/11.13 mmHg.
| Input | SBP | DBP |
|---|---|---|
| Skin color | 18.10 | 11.13 |
| rPPG | 17.42 | 10.94 |
| Contact PPG | 15.71 | 10.89 |