Video based heart rate detection in unrestrained laboratory rats: a feasibility analysis

Figures

• 

  Fig. 1. Recordings and recording setup. (A) Standstill recordings top view and side view. (B) Home Cage setup26.

• 

  Fig. 2. Graphical representation of the approach used to extract mean intensity signals from the acquired video frames. All steps were repeated for each video frame to receive the raw signal in time domain.

• 

  Fig. 3. Signal Processing Pipeline.

• 

  Fig. 4. Example of RoI selections. A: Example of RoI selection in top view video. B: Example of RoI selection in side view video.

• 

  Fig. 5. Bland-Altman plot comparing rPPG and ECG measurements. The solid line represents the mean difference (bias), and the dashed lines indicate the limits of agreement (mean ± 1.96 SD). Acceptable deviation of ± 30 bpm is highlighted by a shaded band.

• 

  Fig. 6. RoIs and resulting mean intensity signals. RoIs gained with extraction methods static (A,C,E), CSRT (B,D), and SAM2 (all) from top camera recordings (A,C,E) and side camera recordings (B,D). RoI sizes differ across body parts. The close ups contain the filtered rPPG \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\tilde{s}_{\textrm{BP}_i}(t)$$\end{document} and the ECG is scaled to the rPPG signals. A: Paw: SAM 2 vs. static. The whole paw is segmented with SAM2, only one part with static RoI. Heart rate could be detected in rPPG signal extracted with static RoI, not SAM2. B: Paw: SAM2 vs. CSRT. The whole paw is segmented with SAM2, only one finger with CSRT. Heart rate could be detected in rPPG signal extracted with CSRT, not SAM2. C: Ear: SAM2 vs. static. Clear RoI size differences between body parts and side/top recordings are visible. D: Ear: SAM2 vs. CSRT. Clear RoI size differences between body parts and side/top recordings are visible. E: Tail: SAM 2 vs. static. With SAM2 clear outlines of the…

• 

  Fig. 7. The frequency power spectrum of ECG and rPPG, and a segment of the respective signals in time domain. (A) The frequency power spectrum of ECG and rPPG. The plotted line indicates the frequencies with highest power within a timestep. HR of ECG: 355 bpm ± 1.5 bpm, HR of rPPG: 361 bpm ± 7 bpm (Mean and standard deviation). The rPPG signal is displayed in Fig. 6D. It should be noted that the last part of the spectrum, where the noise content is higher, originated from the last signal part, where the tracker failed. (B) A segment of the respective signals: ECG is scaled to rPPG, green dots indicate matching peaks within 0.05s tolerance.

• 

  Fig. 8. Time resolution in present dataset. Reference heart rate in 8.5 h timeframe and heart rates detected from rPPG. The shaded band marks the deviation of ± 30 bpm from the mean heart rate. (A) Top view, side view, static, SAM2, and CSRT. (B) Tail and ear.