Zurich/Regensburg – Researchers at the Swiss Federal Institute of Technology in Zurich and the KUNO Clinic St. Hedwig in Regensburg have developed an algorithm that can detect pulmonary hypertension in newborns. Artificial intelligence also suggests the right diagnosis.
Researchers at the Swiss Federal Institute of Technology in Zurich (ETH) and the KUNO Clinic St. Hedwig in Regensburg have created an artificial intelligence (AI) that is able to recognize pulmonary hypertension, a heart disease that is difficult to diagnose in newborns, according to a report by ETH. The severe disease causes the pulmonary arteries to remain narrowed after birth or to close again in the first days or weeks after birth.
This restricts the blood flow to the lungs and reduces oxygen saturation in the blood. However, this requires a very specific ultrasound examination. If this heart disease is present, rapid action is required. However, the necessary experience and expertise is often not available outside the large perinatal centers.
The team led by data scientist and ETH professor Julia Vogt, together with neonatologists at the Regensburg clinic, has developed a computer model that can provide reliable support in diagnosing the disease in newborn children. In their study, they first trained their algorithm with many hundreds of video recordings of cardiac ultrasound examinations of 192 newborns. The data set also contained the respective diagnoses.
How well the algorithm can interpret the images was then tested using a data set with ultrasound images of 78 newborns. The result: the model suggested the correct diagnosis in 80 to 90 percent of cases and determined the correct severity of the disease in 65 to 85 percent of cases. According to Vogt, the AI merely provides support. "The key thing for us, however, is that in the end it is always a person, a doctor, who decides." ce/mm