Thanks to medical progress, the group of adult patients with congenital heart disease is growing constantly. However, primary diseases such as transposition of the great arteries (TGA) require lifelong medical care by specialized health professionals and centers. Take, for example, TGA patients, whose right heart chamber has to take over the function of the left heart chamber. They are at an increased risk of ventricular failure.

Avoiding such life-threatening secondary diseases requires an exact assessment of the right chamber’s function and its long-term development. Their altered nature that is due to both the illness and its treatment still presents a huge challenge to diagnostics. Researchers at the Competence Network for Congenital Heart Defects have set their hope on artificial intelligence (AI) for finding a remedy. An initial AI study that took place in collaboration with the renowned Royal Brompton Hospital London suggests that there is a certain potential of learning systems regarding the detection and classification of medical findings.

More Precise Than Humans

Under the direction of ACHD scientist Gerhard-Paul Diller, an international research team was able to prove the utility of new deep learning (DL) algorithms for diagnosing adult patients with severe congenital heart disease for the first time. With an accuracy level of 98%, the DL algorithm, which was specifically trained for analyzing patients’ ultrasound recordings, yielded a slightly better overall accuracy in identifying the correct diagnosis than the test group of specialized physicians.

The researchers had used the image data of three groups: 1. patients with transposition of the great arteries (TGA) after atrial switch surgery (Mustard or Senning procedure); 2. patients with so-called congenitally corrected transposition (ccTGA), in which also the heart chambers are reversed left to right; 3. heart healthy individuals.

Using AI for Improving Life Expectancy and Quality of Life

For the segmentation of the heart chambers, the U-Net architecture model developed by researchers at Freiburg University was used. Segmentation is the illustration of the heart chambers as single, individually definable segments for diagnostic evaluation. “The precise differentiation of the heart chambers’ findings has even exceeded our own expectations,” Gerhard-Paul Diller states, summarizing the results. He is the senior physician at the Department of Cardiology III: Adult Congenital and Valvular Heart Disease at the University Hospital Muenster. Angeborene Herzfehler (EMAH) und Klappenerkrankungen am Universitätsklinikum Münster.

The research team was able to build on experience gathered, for example, during successfully testing such artificial neuronal networks - which are similar to nervous systems - on identifying traffic signs. In the future, these methods of artificial intelligence are meant to facilitate the assessment of ventricular function and its development depending on the underlying disease and treatment. The research goal is an improved individualized preventive care and treatment in order to increase patients’ life expectancy and quality of life.