Enhanced Heart Rate in Motion Accuracy with the Evie Ring Using Advanced Deep Learning Algorithms
Evie Ring Research
7 min minutes read

Enhanced Heart Rate in Motion Accuracy with the Evie Ring Using Advanced Deep Learning Algorithms

Study Director – Michael Leabman,

Founder and CTO, Movano Health



The Evie Ring is designed to provide both medical grade data related to vital signs, as well as wellness information, such as activity tracking and sleep monitoring. There is a strong clinical need for wearable medical devices which can automatically track heart rate, blood oxygen saturation (SpO2), and other biomarkers to improve the health of individuals living with chronic conditions or by anyone interested in maintaining good health. For medical applications in healthcare, accuracy of the data is paramount. The Evie Ring uses advanced algorithms to process signals collected by its sensors, turning data into reliable measurements and actionable insights. Previously, the accuracy of the Evie Ring in monitoring of heart rate at rest was evaluated and demonstrated accuracy within 1 beat per minute (BPM). This study was designed to validate the performance of the ring's novel heart rate (HR) monitoring algorithm under conditions of motion or activity, which synchronizes data from a three-axis accelerometer and PPG sensors to provide precise heart rate readings during any user activity.

The Challenge

Measuring heart rate from PPG signals in wearables presents several significant challenges. One major issue is low blood perfusion, especially in peripheral areas like the finger base, which can lead to weak PPG signals. Additionally, motion artifacts caused by both periodic and non-periodic movements can distort the PPG signal, making it difficult to accurately identify heart rate data. The fit of the ring on the finger also plays a critical role; a loose or too-tight fit can introduce noise or alter the signal. Ambient light interference is another challenge, as external light sources can affect the PPG readings. Furthermore, variations in melanin levels and skin color can impact the absorption and reflection of the light used in PPG, complicating the measurement process. Addressing these challenges requires sophisticated algorithms and careful sensor design to ensure accurate and reliable heart rate monitoring.

The Evie Solution

To overcome the challenges of measuring heart rate from PPG signals in wearables, the Evie HR algorithm employs several advanced techniques. By leveraging both PPG and 3D accelerometer data, the algorithm effectively removes motion artifacts from the PPG signal, ensuring more accurate heart rate tracking even during significant movement.

During the study, a large dataset was collected from individuals engaging in a variety of activities. PPG with 3D accelerometer data from the Evie Ring was collected in tandem with an industry standard chest strap reference heart rate device. Using this data, a deep-learning solution was developed to optimally filter out motion artifacts and accurately track heart rate.

To address the challenge of low perfusion and darker skin tones, the Evie Ring heart rate monitoring (EHRM) algorithm enhances the signal-to-noise ratio (SNR) of the PPG signal and judiciously controls the analog front-end (AFE) settings. This includes the implementation of Automatic Gain Control (AGC), which adjusts the gain to maximize the SNR.  To address light interference, the EHRM algorithm incorporates techniques to suppress ambient light noise, ensuring that external lighting conditions do not affect the accuracy of the readings. To address motion noise, the Evie Ring uses multiple wavelengths of PPG, combining them to optimize the PPG signal further. These multi-wavelength readings also allow the algorithm to adapt to various skin tones and enhance overall measurement accuracy. Through these sophisticated methods, the EHRM algorithm delivers reliable and precise heart rate monitoring, overcoming the inherent challenges of wearable PPG-based heart rate measurement.

Study Design and Results

A study was conducted to validate the performance of the EHRM algorithm. The evaluation protocol was designed to encompass various heart rate levels, ensuring a comprehensive assessment of the algorithm under many possible conditions.


Data Collection

Sixty-five (65) participants were enrolled in the study and represented a diverse range of ages, skin colors, blood perfusion levels, and physical conditions (34 females, 31 males, 17–64 years of age, mean = 34.4, racial breakdown: 7 African American, 15 Asian, 9 Hispanic, 9 Native American, and 25 Caucasian-White). Each of the 65 subjects completed 7 to 10 sessions, encompassing various activities, resulting in a total of 551 sessions. The activities were sleep, resting (sitting), daily activities, walking, running, climbing stairs, cycling, working out at the gym and swimming.  During activity sessions, participants wore an Evie Ring and a Polar H7 chest strap, considered the gold standard for heart rate.


Data from the Evie Ring test devices was compared to the Polar H7 chest strap.

The histogram in Figure 1 below shows the distribution of the HR values that were calculated by the EHRM algorithm in this study. The range of heart rate values was from 35-210 bpm.  The mean and standard deviation were 119 and 31 bpm, respectively.

Histogram of heart rate value in the study as a percentage


The team analyzed the data and determined the ± X bpm error-band accuracy, where X= 5 bpm, 10 bpm, and 15 bpm, and determined the proportion of samples that fell into the X band for that activity.  This represents the percentage of valid data points that have an absolute error of ± X bpm compared to the reference device output. This metric is calculated at a frequency of 1Hz.

Combined performance results for Evie Heart Rate vs. Polar under various activities

As shown in Table 1 and Figure 2, the EHRM algorithm shows a high correlation with the Polar H7 chest strap outputs across a diverse data set, encompassing various melanin/skin colors, perfusion levels, ages, and physical conditions. Overall, the results indicate that the algorithm reliably reports heart rate across all activities.

The most challenging activities in the study were gym sessions (91.1% within ± 5 bpm) and all-day activities (93.4% within ± 5 bpm), due to non-harmonic motion artifacts and low peripheral perfusion. Walking, running and cycling posed significant challenges for the algorithm; yet despite the nature of these activities, the percentage of sessions within the ± 5 bpm band was at the higher range of 94.6%, 94.5%, and 93.5%. During walking and running, heart rate frequency often aligns closely with motion frequency, and the motion artifact component in the PPG signal is significantly stronger than the heart rate component. Figure 2 below shows the time-domain heart-rate values of four sessions.

4 graphs of the time-domain heart-rate values

The spectrogram below demonstrates the algorithm's ability to remove motion artifacts from the original PPG signal (Figure 3). In the spectrogram, any signal that occurs with a high frequency component at a given frequency will present itself as a white line. In Panel A, the original PPG signal is shown.  Notice there are multiple white lines spanning the time chart indicative of both a heart rate and a motion artifact signal. The heart rate and motion frequencies overlap in this example. In Panel B, the acceleration signal is shown.  This signal is only due to the motion artifact. In Panel C, the motion artifact signal from Panel B is removed from the PPG signal in panel A, to show the resultant heart rate.  It can be seen that the algorithm effectively removed the motion artifact leaving the true heart rate signal.

Spectrogram of the algorithm's ability to remove motion artifacts from the original PPG signal


In summary, the Evie Ring algorithm offers a robust solution for accurate heart rate monitoring using PPG and 3D accelerometer data from the Evie Ring. This innovation has the potential to enhance the reliability of wearable health monitors, providing users with more accurate and consistent heart rate measurements. The results of these tests demonstrate the algorithm's effectiveness in various real-world scenarios, making it a valuable tool for all health applications.

Back to blog
1 of 3