[ad_1]
Although intrigued by technology, one expert says it's hard to know "what it looks like" in clinical care.
Artificial intelligence (AI) can detect existing or emerging Fibula A signs in ECGs with normal sinus rhythm, researchers at Mayo Clinic have found. Their retrospective badysis, published online yesterday in the newspaper Lancet, indicates a high degree of accuracy with a single ECG, and this accuracy increases when the AI is applied to multiple ECGs from the same patient.
"A very common clinical scenario is that a person comes to the hospital with an ischemic stroke and we want to know if she is suffering from atrial fibrillation," TCTMD told lead author Paul A. Friedman, MD (Mayo Clinic, Rochester, MN). "We have done previous work using neural networks, machine learning, which revealed that it was extremely powerful for detecting subtle patterns [in ECG tracings]and we wondered: if a person had atrial fibrillation yesterday, is there a way to leave a trace of a discovery on an ECG today too subtle for a human to read, but a computer could detect it? "
To find out, the investigators, led by Zachi I. Attia, MSc, and Peter A. Noseworthy, MD, drew on an archive of the Mayo Clinic's digital data vault.
Friedman noted that once patients present with ischemic stroke, it is essential to know if they have already experienced Afib to be considered for anticoagulation. The implantable recorders or Holter monitors are bulky and expensive to deploy, he said, and they omit many cases of intermittent fibula.. The idea here was to exploit a standard ECG whose creation takes 10 seconds.
Yesterday, if a person was suffering from atrial fibrillation, could she leave a trace of a result on an ECG today too subtle for a human, but a computer could detect it? Paul A. Friedman
From an initial dataset of 1,000,000 ECGs collected at Mayo from 1993 to 2017, researchers focused on nearly 650,000 normal sinus rhythm ECGs from more than 180,000 adults. They divided the ECG records into three groups: 70% were used to form an AI system, 10% to validate the results of this training and the final 20% were used to test how well they were. IA identified A-fib on an ECG. .
The artificial intelligence system built by Mayo is known as the "convolutional neural network", which is a type of automatic learning optimized for searching visual imaging models. To create it, Attia, Noseworthy and his colleagues provided this network with an electrocardiogram representing the first episode recorded for each patient with atrial fibrillation, as well as all the electrocardiograms of the same patient in the 31 days preceding this episode.
"We chose this window of interest on the basis that the structural changes badociated with atrial fibrillation would be present before the first episode of recorded atrial fibrillation; we chose a relatively short time span as a conservative measure to avoid the use of ECG before structural changes develop, "explain the researchers in their article. For patients without A-fib, they provided to the neural network their ECG index.
Using a single ECG per patient, the neural network reached 79.0% sensitivity and 79.5% specificity for the detection of fibro-A. With several electronic ballasts available to the network, sensitivity has improved for 82.3% and specificity at 83.4%.
"Proof of concept"
In an editorial accompanying the article, Jeroen ML Hendricks, MD (South Australian Institute of Health and Medical Research, Adelaide, Australia) and Larissa Fabritz, MD (Institute of Cardiovascular Sciences, Birmingham, England) note that new ideas are welcome because "silent or undetected atrial fibrillation is common". the few screening methods available require a lot of time and resources.
The current hypothesis of the study, they explain, is "that the signature of atrial fibrillation due to structural changes in the atria can be identified by a trained network, using a standard 10-second ECG, recorded during sinus rhythm ".
This approach could be clinically important and could "lead to a paradigm shift in the recording of sinus rhythm rather than atrial fibrillation on an electrocardiogram, with a particular focus on identifying structural changes." However, false negative results could also be part of the results and prevent proper treatment, "note Hendricks and Fabritz, adding that, for this reason," the algorithm that supports the AI would require additional validation in a different cohort of patients, thus testing better health. hospital population, as well as a rigorous evaluation of prospective clinical trials. "
Warren J. Manning, MD (Beth Israel Deaconess Medical Center, Boston), comments on the study for the TCTMD and explains that it offers an intriguing "proof of concept" about the potential of AI for the detection of AA fib from an ECG. That said, Manning seemed somewhat underestimated by the details.
"Despite a million ECGs, they had only a sensitivity / specificity somewhere in the 80s, "he observed. The well-established risk factors for the development of Fibula A – such as a person's age, throbbing, or excessive caffeine intake – may have been comparable to those of the neural network, Manning warned, concluding, " It's hard to know where that's going [into] potential clinical care. "
Friedman and his colleagues are advancing research that hopefully will help clarify these early observations.
"Once we can tell if you've had a stroke and had atrial fibrillation yesterday, we hope we can then tell you that we can treat you with an anticoagulant and prevent another stroke," he said. Friedman. "We have not finished this study yet. We have just done the first; [it] shows that we can find atrial fibrillation that was present and has now stopped. "
Source link
