A reliable mobile-based solution to estimate basic body vitals can go a long way to facilitate patient-doctor interaction in the context of India. With this goal in mind, we at Eka Care, have developed a pulse rate monitor in collaboration with Father Muller (FM) Medical College and hospital. While the outcome of our study shall be published soon, we present some of our initial results here, highlighting the accuracy of our algorithm and the magnitude & diversity of our validation dataset.
Have you ever wondered how pulse oximeters and smartwatches detect your pulse rate, oxygen saturation, and respiratory rate? These devices use Photoplethysmography (PPG) [1], a technology that has revolutionized the way we interact with these body vitals and enabled their continuous monitoring through portable devices.
PPG utilizes a light sensor to measure variations in the intensity of different wavelengths of light as they pass through (or are reflected by) our body. The figure below illustrates this principle and shows how the pulsatile component of artery blood can be used for pulse rate estimation. If you are wondering that the camera set-up in your smartphone resembles the mode "a" in the figure below, you are absolutely right! Your phone camera can be used to compute PPG in a reflective mode (a).
EPRM uses the reflective mode principle for obtaining PPG signal as demonstrated above to compute your pulse rate. Currently, this feature can be accessed through our android mobile application. The following sections highlight the dataset on which EPM is validated.
To demonstrate the accuracy and reliability of EPRM we are building a diverse validation dataset of PPG comprising 10,000 patients in OPD and ICU settings, in collaboration with the FM hospital. So far, we have collected data of 5,700+ patients from 5 different centers of FM Hospital across its rural and urban centres.
Diversity across population characteristics
The graphs below demonstrate that our dataset is rich in terms of gender and age distribution, wherein we have patients with ages ranging from 12 to 90 years. We have also gathered information about the comorbidities present in these patients. Our dataset has a significant number of patients suffering from Hypertension, Diabetes Mellitus, and thyroid disorder.
Abbreviations used above mean - HTN: Hypertension, DM: Diabetes mellitus, TD: Thyroid disorder, BA: Bronchial Asthama, CAD: Coronary artery disease, CVA: Cerebrovascular accident, EP: epilepsy, CA: Cancer
Diversity across physiological state (body vitals)
It is critical that any vital monitoring tool works reliably across different physiological conditions. We ensure that our corpus comprises data encompassing this diversity. As shown in the graphs below, in our dataset, the pulse rate spans 45 to 160 beats per minute, respiration rate from 12 to 36 cycles/min, SpO2 from 90 to 99 % (naturally it's quite skewed), and there is a healthy distribution of systolic and diastolic blood pressure values.
Around 3% of the female patients reported wearing mehndi (heena) or nail paints. We analyse results for this cohort specifically since such factors can potentially impact the measurement of PPG.
To the best of our knowledge, there are no clear guidelines on permissible error margins for portable pulse rate monitors in the context of India. In international publications, authors have typically regarded ±10% as the error margin, which is a threshold for medical-grade ECG monitors [2].
Reference [2] is an interesting read that compares the accuracy of the Apple Watch and Fitbit with the gold standard.
We compare Eka pulse rate monitor results with the one obtained using Dr. Trust's pulse oximeter (reference pulse rate). In several instances, the reference pulse rate is also cross-checked using the Omron BP device. For comparison we use interclass correlation (ICC), mean absolute error (MAE), mean error (ME), the standard deviation of MAE as metrics, a norm across published studies on this topic [2, 3].
The plot below shows the comparison between the two types of measurements for each patient (EPRM vs reference)
The graph above shows good coherence across measurements of two separate devices. We obtain an interclass correlation (ICC) of 0.95
Mean absolute error (MAE) between the two measurements of the same patient is 3.47 beats per min with a standard deviation (SD) of 2.65
These numbers indicate that the EPRM measurements are well within the acceptable limits [2, 3].
The plot below shows the difference in the pulse rate value measured using two separate systems (Eka vs reference device) as a function of reference pulse rate. We obtain a mean error (ME) of 0.54 BPM, which is insignificant. The difference has a low correlation with the reference pulse rate, meaning that the error is almost independent of the pulse rate.
MAE for female patients wearing heena or nail pains came out to be 2.96 BPM with a standard deviation of 2.26. This shows that wearing heena or nail pain did not impact the PPG measurement.
We also analyzed results for cohorts with specific comorbidities such as hypertension and diabetes. The results are comparable to the ones reported above with no significant difference. This indicates EPRM measurement is robust across the body conditions such as varied blood pressure, and blood sugar levels.
Overall, the Eka pulse rate monitor performed reliably on a sizable diverse dataset with good accuracy
We will soon publish an update on a comprehensive evaluation over the entire 10,000 patients dataset, and peer-reviewed articles on the outcome of this study.
Download Eka Care android app to measure your pulse rate now!
We are grateful to Father Muller hospital for diligently collecting data samples.
We also informally compare our results with other devices
.jpg)
.png)