A Non-Invasive Hemoglobin Detection Device Based on Multispectral Photoplethysmography

Figures

• 

  Figure 1. Schematic representation of PPG signal generation.

• 

  Figure 2. The process of non-invasive hemoglobin detection.

• 

  Figure 3. (a) Shows the emission spectrum. (b) Shows the detector spectral range sensitivity. (c) Shows the optical bandpass filter.

• 

  Figure 4. The structure of our finger clip: (A) Traditional Finger Clip: PD light-receiving surface changes with the opening and closing of the finger clip. (B) Our Finger Clip: The PD light-receiving surface does not change with the opening and closing of the finger clip. The pressure exerted by the finger and the distance from the light source to the sensor can be monitored. (C) Structural drawing of the finger clip we designed: Employing 3D printing techniques, we fashioned this device, allowing for assembly without the need for adhesives or screws.

• 

  Figure 5. (a) Shows physical views of the non-invasive hemoglobin instrument. (b) Demonstrates a user’s application of the homemade testing finger grips.

• 

  Figure 6. Scheme of PPG primary-wave extraction.

• 

  Figure 7. (a) Original PPG signal; sampling rate, 500 Hz; number of samples, 6400. (b) PPG signal after mean filtering. (c) PPG signal after filtering with an FIR filter. (d) The result of a single data acquisition; each PPG signal is followed by the pressure sensor data and displacement sensor data recorded in real time during the measurement.

• 

  Figure 8. Heatmap of the correlation analysis.

• 

  Figure 9. Diagram of input features. To facilitate the presentation of the obtained data, nine features were normalised. And some of the data are presented in the figure, with increased darkness of the line corresponding to elevated hemoglobin content.

• 

  Figure 10. (a) R2 of the three models in the cross-validation; (b) MSE of the three models in the cross-validation; (c) MAE of the three models in the cross-validation.

• 

  Figure 11. Comparison of (a) the AdaBoost regression model, (b) the BPNN regression model, and (c) the RF regression model with regard to predicting hemoglobin concentration.

• 

  Figure 12. (a) Bland–Altman diagram of the AdaBoost regression model for predicting hemoglobin concentration. (b) Bland–Altman diagram of the BPNN regression model for predicting hemoglobin concentration. (c) Bland–Altman diagram of the RF regression model for predicting hemoglobin concentration. The horizontal coordinate of the Bland-Altman diagram is the mean of the predicted and actual values, the vertical coordinate is the difference between the predicted and actual values, and the diagram in-cludes three lines, with the middle line indicating the mean of the difference, and the top and bottom lines being the upper and lower limits of the 95 per cent consistency boundaries. The Bland-Altman diagram allows visualisation of the measurement deviation and degree of consistency between two observations. If most of the points are distributed near the mean difference and within the standard deviation, the two data sets are in good agreement.