Cardiomyopathy is a disease of the heart muscle. It causes the heart to have difficulties to pump blood to the whole body, which can lead to symptoms of heart failure. Cardiomyopathy also can lead to sudden cardiac death. In China, it is estimated in 2023 that there are about 1,000,000 patients suffering hypertrophic cardiomyopathy and about 120,000 patients suffering dilated cardiomyopathy. According to “2024 heart disease and stroke statistics: a report of US and global data from the American Heart Association”, 410,000 deaths were estimated for cardiomyopathy and myocarditis based on 204 countries and territories in 2021.
Since genetic testing is costly and has a long report generation period, the phenotypic cardiac MRI (CMR) imaging is usually considered as the necessary examination utility for cardiomyopathy diagnosis, as it can be used to measure the cardiac structure and function, evaluate the severity of cardiomyopathy. However, the analysis of cardiac MRI imaging is a heavy workload to physicians as it requires a lot of time to contour heart anatomy and requires long time training to improve their accuracy. For the similar reasons, only two commercial software tools are available over the world for CMR analysis for a long time.
Beyond the CMR imaging, clinical data such as baseline and lab test variables are also important for cardiomyopathy diagnosis. So given the amazing progress in deep learning techniques, we aim to apply deep learning on multi-modal medical data to diagnose various forms of cardiomyopathies.
However, in the medical application scenario, large deep learning models are inherently black boxes, and hard to gain the trust of medical professionals. We break the large black-box into 3 pieces and have medical knowledge embedded in the framework, so that we not only take advantage of the deep learning advances, but also provide an opportunity for medical professionals to trust AI.
Furthermore, our knowledge oriented deep learning framework achieved high diagnosis accuracy to benefit cardiomyopathy patients and could run in portable or small computing devices.
Through our research project, the public attention may be directed to other possibilities than the overwhelming large AI models, especially in the medical AI region. Our knowledge oriented approach perfectly avoid the illusion pitfalls that most large AI models currently face.
