Abstract: - Correct and prompt performance of cardiopulmonary resuscitation yields improvements in mortality

and social return rates. Chest compression, a vital cardiopulmonary resuscitation technique, requires regular re-

education for skill maintenance. Training with a manikin is feasible for chest compression, but assessing

proficiencies without an expert presents challenges. This study aims to facilitate autonomous chest compression

training even without expert supervision based on pose estimation. Twenty subjects were recruited for the

training and successive performance evaluation of chest compression on a sensor-equipped training manikin,

and the corresponding videos were recorded simultaneously. A system was developed to analyze chest

shooting at a position shifted 45 degrees from the front. The proposed method may serve as a tool for

completely automated CPR chest compression training, anticipating an increase in citizens proficient in

cardiopulmonary resuscitation.

Key-Words: - cardiopulmonary resuscitation, chest compression, compression count, deep learning, pose

Received: May 27, 2024. Revised: August 9, 2024. Accepted: September 14, 2024. Published: October 21, 2024.

arrives. Data from the Japan Ministry of Internal

Affairs and Communications reveals insights into

the nationwide occurrences of cardiac and

respiratory arrest injuries and diseases in Japan,

indicating that CPR doubles the survival rate one

month later while tripling the social return rate, [1].

recommended chess compression rate is 100–120

compressions per minute, ensuring complete chest

recoil after each compression, [2]. Additionally, it is

CPR improves survival from cardiac arrest,

including adequate rate and depth of chest

compressions, full chest recoil between

compressions, minimum interruptions in chest

compressions, and the avoidance of excessive

ventilation, [4]. In a study on CPR during out-of-

hospital cardiac arrest [5], chest compressions were

neglected half of the time, and a significant portion

of compressions were too shallow. Consequently,

maintaining CPR skills necessitates regular chest

emergency lifesavers. While it is possible to assess

the tempo and depth of chest compressions using a

training manikin with a pressure sensor, its

expensive nature poses a challenge to widespread

use. Although chest compression training with

inexpensive manikins is conducted in places like

driving schools, visual evaluation limitations and

compression evaluation system using pose

estimation. This system will facilitate the

assessment and verification of chest compression

skill proficiency even without expert oversight, with

the expectation that recursive self-training can lead

to more efficient chest compression skill

different chest compression interruption procedures.

They found significant differences between experts

and non-experts [6]. Similarly, an RGB-D (Kinect)

sensor was employed for motion capture, evaluating

compression tempo and depth using an approximate

sine wave model from skeletal data, [7]. Meinich-In

[8], an experiment with chest compression modeling

was performed using Kinect, demonstrating

acceptable measurements of sternum compression

depth with a smartphone's depth camera and

were recorded at frame per second (FPS) =25 by

two GoPro HERO10 Black cameras made by GoPro

Inc., USA.

Fig. 1: Experimental scene, GoPro HERO10 Black

camera, and Resusci Anne QCPR manikin

Chess compression can be well analyzed using pose

estimation models by tracking keypoints of hands to

get the features of chest compression. However,

automatically analyzing human posture (pose) from

detect peak clusters within the coordinate data.

Peaks were used for successive analysis including

interruption detection, and the evaluation of

function. By setting the time lag (k) to the average

tempo, the stability of the posture and tempo during

compression was evaluated.

and N is the length of the x-coordinate waveform in

a key point.

by the elbow was calculated using the cosine

theorem for the x and y coordinates of the key

points at the right shoulder, elbow, and wrist.

are shown in (5) and Figure 5.

Fig. 5: Definitions of α and β

of compressions (CC), Count of appropriate tempos

(CT), Count of appropriate depths (CD), and Count

of appropriate recoils (CR). It should be noted that

for the comparison, considering the discrepancy

between the recorded video and the Skill Report, the

comparison was made separately for each video of

chest compressions performed by 20 subjects. The

accuracy for each item was evaluated using an

accuracy defined as follows.

Table 2. Accuracies of the assessment items by each

In the evaluation of posture, the average of all

subjects was taken for the stability of the

compression posture as shown in Figure 6, the angle

of the elbow, and the angle against the training

manikin for each type of chest compression

performed.

posture.

4 Discussion

In this study, evaluations were performed for each

video. Therefore, if the results of the proposed

evaluation system and the Skill Report, i.e., the

golden standard, do not match, it is considered

unacceptable. Consequently, a slightly lower

accuracy is inevitable, but the evaluation of tempo

manikin is possible, thus it is deemed a viable

method for evaluating chest compression posture.

elbow and the angle towards the manikin improved

malignant ventricular fibrillation and tachycardia,

and therefore improve their survival rate and social

return rate.

Acknowledgement:

We thank the subjects’ efforts in the experiments.

References:

[13] Ultralytics, Pose - Ultralytics YOLOv8 Docs,

[14] Google, Pose detection | ML Kit | Google

[19] SciPy, scipy.signal.find_peaks — SciPy