New AI Model Detects Heart Defects Early with ECG Screening
A team of investigators from Brigham and Women’s Hospital and Keio University in Japan have made a breakthrough in detecting heart defects early using artificial intelligence (AI). Their newly developed deep learning AI model screens electrocardiograms (ECG) for signs of atrial septal defects (ASD), a condition that can lead to heart failure if not treated in a timely manner. The results of their study, published in eClinicalMedicine, highlight the potential benefits of deploying this model on a population level to identify patients at risk before irreversible damage occurs.
ASD is a common adult congenital heart disease characterized by a hole in the heart’s septum, which allows blood to flow between the left and right atriums. While it is diagnosed in only about 0.1% to 0.2% of the population, it is believed to be underreported due to the lack of noticeable symptoms before complications arise. Symptoms, when present, are typically mild or may not appear until later in life, making early detection challenging. These symptoms can include an inability to engage in strenuous exercise, abnormal heart rhythms, heart palpitations, and an increased susceptibility to pneumonia.
Even in the absence of symptoms, ASD can strain the heart and raise the risk of various complications such as atrial fibrillation, stroke, heart failure, and pulmonary hypertension. Once these complications manifest, they become irreversible, even if the defect is surgically corrected later on. Detecting ASD early allows for minimally invasive surgery to correct the issue, improving life expectancy and minimizing potential complications.
At present, there are several methods to detect ASD. The most common approach is listening to the heart using a stethoscope, which can identify the largest defects in approximately 30% of cases. Another method is echocardiography, a time-consuming test that is not suitable for screening purposes. Electrocardiography (ECG), on the other hand, is fast and can potentially be used as a screening tool. However, human analysis of ECG readouts to identify ASD has limited sensitivity.
To overcome the limitations of human analysis, the research team investigated whether an AI model could more accurately detect ASD from ECG data. They trained a deep learning model using ECG data from over 80,000 patients aged 18 and above who had undergone both ECG and echocardiogram tests. Out of these patients, 857 were diagnosed with ASD. The data was collected from three hospitals: Brigham and Women’s Hospital and Keio University in Japan, which are large teaching institutions, and Dokkyo Medical University, Saitama Medical Center in Japan, a community hospital.
The AI model was then tested using scans from Dokkyo, where patients were not specifically being screened for ASD. The model demonstrated higher sensitivity compared to using known abnormalities found on ECG readings to screen for ASD. Specifically, the model accurately detected ASD 93.7% of the time, while relying on known abnormalities only achieved a detection rate of 80.6%.
Lead author, Shinichi Goto, MD, Ph.D., has stated that the AI model outperformed human experts in identifying cases of ASD. However, one limitation of the study is that the model was trained using samples primarily from academic institutions that typically deal with rare diseases like ASD. Further research is needed to determine the model’s effectiveness in a general population.
Despite its potential, even using echocardiography to detect ASD does not guarantee identification of every defect, especially smaller ones that are less likely to require surgical closure. Furthermore, due to the nature of machine learning, it is challenging to determine which features the AI model identifies in ECG readouts. This lack of transparency hinders the ability to identify specific indicators of ASD from the model’s outputs.
The promising results of this study suggest that integrating AI-based ECG screening into routine medical appointments or leveraging existing ECG tests taken for other purposes could greatly enhance the early detection of ASD before it leads to irreversible heart damage. With ECG being a relatively low-cost and widespread procedure, implementing this technology on a population level could potentially save lives and prevent severe complications associated with untreated ASD.
