Artificial Intelligence-based Automatic Echocardiographic Quantification in Advanced Heart Failure (AIED Study)

Heart failure (HF) is a clinical complication. About half of HF patients have heart failure with normal systolic fraction (HFpEF), and most of them are elderly women. The other type is systolic heart failure, characterized by a left ventricular ejection fraction of less than 40 (LVEF<40). The clinical symptoms of HFpEF are very similar to those of low systolic fraction heart failure (HFrEF) with abnormal left ventricular ejection fraction. Generally speaking, the morbidity and severity of HFrEF are higher, and the survival rate is lower. HFpEF is generally difficult to diagnose, so it is critical to find a method to accurately diagnose HFpEF. HFpEF is most commonly diagnosed by echocardiography and biomarkers. In a cardiac ultrasound examination, it is impossible to diagnose HFpEF based on a single parameter of the results. We need multiple examination parameters to gather enough evidence to confirm the existence of HFpEF. These parameters include the mitral inflow velocity pattern, the pulmonary vein flow pattern, changes in flow velocity from the left atrium to the left ventricle, tissue Doppler measurements, and M-mode ultrasound measurements. We train artificial intelligence to distinguish between normal and abnormal cardiac ultrasound images, measure or evaluate all the above parameters, and analyze all the data. We hope that, with the help of artificial intelligence, we can improve the prediction and diagnosis rate of HFpEF.

Simply diagnosing HFrEF requires an LVEF of less than 40%. Diagnosing HFpEF poses significant clinical challenges because no single tool or method can reliably confirm the condition or predict associated hospitalizations. Consequently, diagnosis depends heavily on physician judgment, requiring the synthesis of considerable clinical data and information. Recognizing the heterogeneity of the HFpEF phenotype, phenomapping integrates comprehensive data (clinical history, physiological measurements, biomarkers, ECG, echocardiographic parameters) to stratify patients into distinct subtypes, thereby optimizing classification for improved prognostic prediction. It can be seen from this that HF will rely heavily on artificial intelligence in the future to assist in patient data management and classification diagnosis and further develop clinical prediction models. This research project will implement a multi-center design to collect ultrasound images from patients with heart failure and perform relevant analyses using artificial intelligence.