DAIMonD (Diastolic Artificial Intelligence Modelling on cardiovascular Data)

Echocardiography (heart ultrasound scanning) is the most widely used imaging test to assess how the heart relaxes and fills with blood - termed diastolic function. Diastolic dysfunction can cause heart failure with preserved ejection fraction, which represents between a third to a half of all heart failure.

The current clinically used echocardiographic method of assessing diastolic function relies on expert-consensus based recommendations. These rely on certain key measurements from the scan (which cannot always be obtained), frequently result in an “indeterminate” grading, and are less accurate than the diagnostic benchmark of invasively measured heart pressures (not routinely performed in the majority of patients). Furthermore, certain heart conditions or blood flow states can confound certain diastolic parameters by affecting the measurements independently to diastolic dysfunction, making robust non-invasive diagnosis even harder.

We have developed a new method of assessing diastolic function, which uses routinely acquired echocardiography parameters and machine-learning to generate a diastolic function model. Our model is based upon thousands of scans from the EchoVision study, and is used to derive a personalised diastolic disease score for each scan.

The primary objective of this study is to clinically validate the diastolic disease score in data from patients without diastolic-confounding factors. We will do this by retrospectively collecting pairs of data from patients who have undergone an echocardiogram and heart catheterisation procedure. The invasive pressure measurement will be used as the reference-standard for identifying advanced diastolic dysfunction.

Secondary objectives are to clinically validate the diastolic disease score in the presence of diastolic-confounding factors such as pulmonary hypertension, valvular heart disease, and atrial fibrillation.

Clinical echocardiogram report data, clinical heart catheterisation haemodynamic data and electronic medical record data will be collected retrospectively from patients already seen in the NHS as part of their routine clinical care. Data will not be collected from patients who have already opted out of having their data used for medical research through the NHS ‘national patient opt-out scheme’.

This research is funded by the NIHR Oxford Biomedical Research Centre, University of Oxford, UK.

This research has been approved by the UK Health Research Authority (IRAS Project ID: 310486).