Multimodal assessment of arterial stiffness using photoplethysmography and laser Doppler flowmetry

Figures

• 

  Fig. 1. Photoplethysmography (PPG) signals acquired from healthy, intermediate, and unhealthy vessel–tissue phantoms using a multi-wavelength sensor, incorporating red, infra-red (IR), and green channels. The blood-mimicking fluid (BMF) used an intralipid formulation prepared as a 30% oil-in-water emulsion.

• 

  Fig. 2. PPG signals acquired from healthy, intermediate, and unhealthy vessel–tissue phantoms using a multi-wavelength sensor, incorporating red, IR, and green channels. The BMF used LPFS at a concentration of 0.25% of the total volume.

• 

  Fig. 3. Box plot of the PPG features extracted from phantoms with the intralipid-based BMF. The red, IR, and green features are shown across phantoms arranged by decreasing elasticity: healthy (0.82 MPa), intermediate (1.48 MPa) and unhealthy (2.06 MPa).

• 

  Fig. 4. Box plot of the PPG features extracted from phantoms with the LPFS-based BMF. The red, IR, and green features are shown across phantoms arranged by decreasing elasticity: healthy (0.82 MPa), intermediate (1.48 MPa) and unhealthy (2.06 MPa).

• 

  Fig. 5. Correlation of PPG features with vessel elasticity using intralipid-based BMF. Pearson correlation coefficients (p < 0.05) are presented for amplitude, area under the curve (AUC), median end-datum difference, and median upslope–downslope ratio across red, IR and green wavelengths. Stronger correlations were observed in intralipid, compared to LPFS-based features, particularly for amplitude and AUC, while slope-derived features displayed weaker or inconsistent associations.

• 

  Fig. 6. Correlation of PPG features with vessel elasticity using LPFS-based BMF. Pearson correlation coefficients (p < 0.05) are presented for amplitude, AUC, median end-datum difference, and median upslope–downslope ratio across red, IR and green wavelengths.

• 

  Fig. 7. Laser Doppler flowmetry (LDF) direct current (DC) flux signals. The features are acquired from healthy (green), intermediate (yellow), and unhealthy (orange) vessel–tissue phantoms using the intralipid-based BMF (left panel) and LPFS-based BMF (right panel). The raw DC signals are displayed over the full 4-min recording.

• 

  Fig. 8. Polynomial regression fits of DC mean flux extracted from LDF signals. The features are acquired from healthy (green), intermediate (yellow), and unhealthy (orange) vessel–tissue phantoms using the intralipid-based BMF (top panel) and LPFS-based BMF (bottom panel).

• 

  Fig. 9. Confusion matrix for intralipid-based multiclass vessel classification. Classifications of healthy, intermediate and unhealthy phantoms are compared PPG and multimodal (PPG and LDF) algorithms.

• 

  Fig. 10. Confusion matrix for LPFS-based multiclass vessel classification. Classifications of healthy, intermediate and unhealthy phantoms are compared using PPG and multimodal (PPG and LDF) algorithms.

• 

  Fig. 11. Fabricated vessel–tissue phantoms. The phantom on the left represents a healthy femoral artery, with a vessel characterised by a Young’s modulus of 0.82 MPa. The central phantom corresponds to an intermediate pathological stage, incorporating a vessel with a Young’s modulus of 1.48 MPa. The phantom on the right represents an atherosclerotic femoral artery, with a vessel modelled to have a Young’s modulus of 2.06 MPa.

• 

  Fig. 12. PPG and LDF sensor placement. Configuration of the LDF (A) and PPG (B) sensors on the vessel–tissue phantom. The PPG sensor and LDF probe were alternated to obtain sequential readings and prevent optical interference.