Data-Driven Decision Support for the Diagnosis and Prognosis of Critical Heart Failures Based on 3D Echocardiography Data

Objectives: Real-time 3D Echocardiography (RT3DE) is a fast, noninvasive and accurate method to measure cardiac left ventricular (LV) volumes and function. However, the diagnosis and prognosis based on these parameters are challenging because the data handling requires expertise for data interpretation and knowledge of reference values. The objective of our project is to develop a data-driven and computer aided procedure to diagnose the severity of heart failure in children based on reference data on healthy subjects. Furthermore, we establish a method to predict the cardiovascular condition of a patient for a projection into the future.

Methods: RT3DE data of the LV in 600 healthy subjects are used as reference and compared with values of pathological LVs. These data include end-diastolic and end-systolic volumes (EDV, ESV) and systolic and diastolic parameters derived from volume changes during the cardiac cycle. For the diagnosis, we train a classifier on the data and apply cross-validation for model selection. For the prognosis, we use a regression model to predict the cardiac parameters for a chosen time period. Customized cardiovascular computer models are used to fit the models individually to each subject. We can simulate LV pressure-volume loops to estimate the patient´s medical condition. In addition, we vary cardiac parameters in the model to simulate the change for the cardiovascular system.

Results: RT3DE data of the LV in 600 children and adolescents aged from 0–18 years were compared with data of children with aortic stenosis, borderline and hypoplastic left ventricle and cardiomyopathy. First, we computed the resulting parameters EDV and ESV and fitted regression models. These models can be plotted as percentiles and used for visual assessments. We developed a cardiovascular computer model of LV to simulate the cardiac cycle based on the volume data. Relevant parameters such as the resistance, elastance and compliance of the LV are computed. Recent results with newborn children match closely with literature values.

Conclusion: This approach may help clinicians to estimate the cardiac function of an individual patient in a fast procedure by using a computer estimation. Furthermore, it may give a prognosis, so that critical conditions can be predicted. Interventions and treatments can be planned accordingly. As precondition, this data-driven approach has to be applied in clinical use tests to prove its validity for routine examinations.