Despite its reputation as a cold and unfeeling technology, artificial intelligence (AI) is demonstrating a heartwarming potential, as researchers at Osaka Metropolitan University unveil an extraordinary application that evaluates cardiac functions and precisely identifies valvular heart disease using chest X-rays.
In an impressive integration of medical science and technology, the team has introduced an innovative AI-based method to categorize heart functions and effectively pinpoint valvular heart diseases. Their pioneering achievement is featured in a recent publication in The Lancet Digital Health.
Valvular heart disease, a leading cause of heart failure, is typically diagnosed through echocardiography, a technique demanding specialized expertise and thus facing a shortage of skilled technicians. On the other hand, chest radiography, commonly known as chest X-rays, is a prevalent diagnostic tool primarily used to identify lung diseases. Although chest radiographs incidentally capture the heart, its capability in detecting cardiac function or disease remained largely unexplored.
Chest radiographs are extensively conducted in hospitals due to their accessibility and rapidity. Recognizing this, Dr. Daiju Ueda and his research team from the Graduate School of Medicine at Osaka Metropolitan University’s Department of Diagnostic and Interventional Radiology conceived the notion that chest radiographs could supplement echocardiography by determining cardiac function and disease.
The team, under Dr. Ueda’s guidance, developed an AI model capable of precisely categorizing cardiac functions and valvular heart diseases using chest radiographs. To counter potential biases that may emerge when AI is trained on a single dataset, the researchers aimed for a multi-institutional approach.
They collected a substantial dataset comprising 22,551 chest radiographs and an equivalent number of echocardiograms from 16,946 patients across four facilities between 2013 and 2021. Utilizing the chest radiographs as input and the echocardiograms as output data, the AI model was meticulously trained to learn the connections between the two datasets.
The outcome was a highly accurate AI model that adeptly categorized six specific types of valvular heart disease, with Area Under the Curve (AUC) scores ranging from 0.83 to 0.92. AUC, a performance indicator for AI models, employs a scale from 0 to 1, with higher values indicating superior performance. Notably, the model achieved an AUC of 0.92 when detecting left ventricular ejection fraction, a critical metric for monitoring cardiac function, at a 40% cut-off point.
Dr. Ueda expressed his belief in the significant impact of their research: “It took us a very long time to get to these results, but I believe this is significant research. In addition to improving the efficiency of doctors’ diagnoses, the system might also be used in areas where there are no specialists, in night-time emergencies, and for patients who have difficulty undergoing echocardiography”.
The team’s achievement underscores the immense potential of AI in revolutionizing medical diagnosis and patient care, showing that technology can indeed have a heart.
