High-Throughput, Contact-Free Detection of Atrial Fibrillation From Video With Deep Learning
Why the study?
Existing approaches for atrial fibrillation detection can screen only 1 patient at a time, leaving the feasibility of concurrent high-throughput screening unknown.
Can a deep convolutional neural network accurately detect atrial fibrillation from facial photoplethysmographic signals captured via video in multiple patients concurrently?
Population
20 patients with permanent AF and 24 control individuals in sinus rhythm
Comparison
Video-based facial photoplethysmography via DCNN vs reference electrocardiogram traces
Design
Prospective proof-of-concept study
Authors
Discussion
Loading...
Member takes
Implication
Should not yet change practice; supports prospective validation trials to confirm accuracy in clinical settings.
Key Points
- This study aims to evaluate the feasibility of high-throughput atrial fibrillation detection using facial photoplethysmographic signals from video.
- Prospective evaluation involving 20 patients with permanent atrial fibrillation and 24 control individuals.
- Utilized a Canon 50D digital camera to capture 64 videos of multiple patients concurrently from 150 cm away.
- Applied a pretrained deep convolutional neural network to analyze the facial photoplethysmographic signals.
- Successfully recorded videos capturing 5 patients simultaneously in various heart rhythm permutations.
- Deep learning model effectively analyzed the facial signals, differentiating between AF and sinus rhythm, confirmed by reference ECG traces.
Structured PICO
Can a deep convolutional neural network accurately detect atrial fibrillation from facial photoplethysmographic signals captured via video in multiple patients concurrently?
P
Population
44 individuals, comprising 20 patients with permanent atrial fibrillation (mean age 76.6, 60% male) and 24 control individuals in sinus rhythm (mean age 56.8, 58.3% male).
I
Intervention
Contact-free high-throughput atrial fibrillation detection analyzing facial photoplethysmographic (FPPG) signals from multiple patients concurrently using a single digital camera and a pretrained deep convolutional neural network (DCNN).
C
Comparator
Reference electrocardiogram (ECG) traces.
O
Outcome
Feasibility of high-throughput AF detection (binary AF/SR classification).surrogate
A deep learning approach using facial photoplethysmography from a digital camera demonstrates the feasibility of contact-free, high-throughput screening for atrial fibrillation in multiple patients simultaneously.
Cite This Study
Yan et al. (2019) studied this question.