Smartphone based atrial fibrillation detection in horses via build in accelerometer sensors and artificial intelligence

Atrial fibrillation (AF) is the most common clinically important arrhythmia in horses and a frequent cause of poor performance. In some horses, AF may remain undetected for longer periods of time. In humans the camera of the smartphone is frequently used to acquire a photoplethysmography signal of the heart rhythm for AF detection with medically certified apps. This is not possible in horses due to the hair and pigmented skin. As an alternative, the built-in accelerometer of a smartphone has also been used for heart rate detection in humans and dogs. The goal of this study was to asses if a custom designed algorithm could extract the heart rhythm of a horse from the built-in accelerometer of a smartphone by the use of an app. For this study 3 recordings of at least 30 seconds were made in 9 horses with AF and 9 horses in sinus rhythm. Recordings were made using an android phone (SAMSUNG GALAXY S20) while recording the raw accelerometry data at 100Hz to a comma separated text file. The smartphone was held with its base perpendicular the skin in the 5th intercostal space 5cm above the height of the left olecranon. Only the y-axis of the accelerometer was used. The algorithms for signal filtering, heart rate detection and classification were custom made and implemented in MATLAB R2021b. In brief, out of the 3 recordings of each horse the recording with the best signal-to-noise ratio was selected by the algorithm, after which the heart rate was extracted with a band pass filter (5-25Hz) followed by a local maxima detection. For classification into AF and sinus rhythm, a pretrained neural network was used that has been previously described1. The final classification obtained a 94% accuracy, 89% sensitivity and 100% specificity for AF detection. One horse with AF could not be classified by the algorithm because all 3 recordings had a too low signal-to-noise ratio due to skin twitching upon contact with the smartphone. Further validation with different smartphone types and more horses will be necessary for further optimizing the algorithm and making it more robust, but the current study resulted in a promising quick and easy screening tool for AF in horses