Researchers from Japan have successfully developed an AI model that can produce fast and accurate valvular heart disease diagnoses.
Artificial Intelligence (AI) is becoming increasingly integral to our healthcare ecosystem, with applications ranging from diagnosis to health monitoring. Just recently, scientists in Japan unveiled a novel and accurate AI method to detect valvular diseases and cardiac functions from chest radiographs. Their groundbreaking study was published in the highly influential journal The Lancet Digital Health.
Heart valves are extremely critical for blood circulation and health. They open and close to allow blood to flow from one area of the heart to another, thus ensuring that blood flows at the right time and in the correct direction. Unfortunately, infections, ageing, or congenital heart defects may result in damaged valves.
Valvular heart diseases are conventionally diagnosed and monitored using echocardiography, a test that uses sound waves to create pictures of the heart. By analysing the blood flow through the heart and heart valves, doctors are able to determine the condition of the patient’s heart and the severity of their valvular heart disease. Although echocardiography provides instant and reproducible results, this method of detection requires specialised skills and technicians.
Another common test that can be useful to determine valvular diseases is chest radiography using X-rays. Similar to echocardiography, chest X-rays are quick, accessible, and reproducible, making them an attractive choice to diagnose medical conditions in the chest area. However, despite the heart being visible in chest radiographs, the ability of chest radiography to determine heart disease is still poor. Hence, the research team, led by Dr. Jaidu Ueda from the Department of Diagnostic and Interventional Radiology at the Graduate School of Medicine of Osaka Metropolitan University, set out to develop an AI model that can accurately identify cardiac functions and valvular heart diseases from chest radiographs.
In their study, they used datasets from multiple healthcare institutions to develop a deep-learning AI model that can simultaneously detect valvular disease and cardiac functions from chest radiographs. The authors mentioned that it is essential to use datasets from multiple healthcare institutions so as to reduce overfitting, an undesirable machine learning behaviour where the final model works well only for images in the trained data sets. The AI model was trained to learn features relating to heart diseases by matching the chest radiograph input data to the echocardiography output reports.
Overall, the AI model was able to accurately categorise six types of valvular heart diseases, suggesting that chest radiographs have intrinsic features that can help identify cardiac functions and valvular diseases. Furthermore, the model has some advantages over echocardiography-based evaluations. For instance, the system requirements to run the AI model are very low, so any computer in daily clinical practise should be able to implement the model and rapidly acquire the results. In addition, the model can be used at any time, making it useful in areas where echocardiography specialists are not available or in nighttime emergencies when an echocardiography technician is not working.
Despite its many benefits, the authors warned that their model should only be used for classification purposes and not for estimating the precise value of echocardiographic results. Hence, they advise that the model should be used as a complement to transthoracic echocardiography for cardiac assessments. They also noted the need to incorporate data from more patients, especially those with other indications of heart disease, to further validate their model. All in all, the authors believe that the innovative model will be able to improve the efficiency of valvular heart disease diagnoses. [APBN]
Source: Ueda et al. (2023). Artificial intelligence-based model to classify cardiac functions from chest radiographs: A multi-institutional, retrospective model development and validation study. The Lancet Digital Health, S2589750023001073. https://doi.org/10.1016/S2589-7500(23)00107-3