Table 4 A comparison with state of art.

From: Application and accuracy analysis of different facial regions based on deep learning in the diagnosis of hypertension

Study/Method

Modality

Model/Approach

Dataset size

Reported metrics

Notes

Facial complexion using L\ and a\ (Bayesian) ([PMC][1])

Facial color (CIELAB)

Bayesian & LASSO-based statistical models

1,099

subjects

AUC ≈ 0.82–0.83

Non-DL, TCM context

PPG morphology classification ([MDPI][2])

PPG signals

ML classification on waveform features

359 recordings

Accuracy ≈ 0.73

Contact-based wearable tech

Proposed (this study)

Facial images

U-Net + ResNet-18 (deep learning)

506 images

Acc ≈ 0.83, F1 ≈ 0.75, AUC ≈ 0.84

Non-contact, interpretable via facial zones

  1. Note: [1]: Ang L, Lee B J, Kim H, et al. Prediction of hypertension based on facial complexion[J]. Diagnostics, 2021, 11(3): 540.; [2]: Evdochim L, Dobrescu D, Halichidis S, et al. Hypertension detection based on photoplethysmography signal morphology and machine learning techniques[J]. Applied Sciences, 2022, 12(16): 8380.
Back to article page