Artificial Intelligence: Improving Early Detection of Pulmonary Hypertension by Transthoracic Echocardiography: ADOPT
Artificial Intelligence: Improving Early Detection of Pulmonary Hypertension by Transthoracic Echocardiography: ADOPT
Pulmonary Hypertension (PH) is a condition caused by high blood pressure in the blood vessels that carry blood to the lungs. It can cause severe breathlessness and failure of the right side of the heart. Sadly it is often fatal, and life expectancy ranges from months to years. For some subtypes of PH, effective treatments exist which can improve life expectancy and quality-of-life. Accurate tools for the assessment of PH are therefore essential so that life-saving medications can be started earlier.
In existing diagnostic pathways, evidence for the suspicion of PH is frequently overlooked, significantly delaying the time to diagnosis. Echocardiography (echo) is a quick, safe and well-tolerated test requested to investigate breathless patients, and which can provide useful information about the suspicion of PH. However, outside of specialist PH centres, doctors may not routinely look for and comment on the presence of clues to possible PH.
The investigators think that using Artificial Intelligence (AI) techniques to read echo's could make their interpretation faster and more reliable. There may also be subtle clues to the presence or severity of PH on echo, less recognisable to the human eye, which AI can identify.
In this study the investigators will gather echo images from 5 specialist PH hospitals across the UK which have all been anonymised (patient's name and personal details removed). These will all be historic scans (i.e. have already taken place) and will be grouped into those with PH present (including PH sub-type) or absent. These anonymised echo images will be used to develop and train an AI tool to identify scans where PH is present, including which specific type of PH may be present. The developed AI tool will then be tested on a separate group of scans (not used in the training stage) to validate its performance.
In this study the investigators will gather retrospective echo images from 5 specialist PH hospitals across the UK (Royal Free Hospital NHS FT; Sheffield Teaching Hospitals NHS FT; Royal Papworth Hospital NHS FT; NHS Golden Jubilee National Hospital Glasgow; Royal United Hospitals Bath NHS FT).
These will all be historic scans (i.e. have already taken place) and will be grouped into those with PH present (including PH sub-type) or PH absent. Inclusion criteria involve patients aged ≥18 who have undergone both a transthoracic echo (TTE) and a right heart catheter (RHC) as part of their clinical care. Exclusion Criteria will involve patients aged <18, known or suspected congenital heart disease and patients who have opted out of allowing their information to be used for research and planning (via the national data opt-out choice). A clinical case report form (CRF) will be used to capture patient demographics, clinical data with regards to the PH assessment including previous TTE results. Where available, mortality data will be recorded within 5 years of the RHC.
These anonymised echo images will be collated and labelled centrally in a core lab at the RUH Bath, who will work with Janssen to develop and train an AI tool to identify scans where PH is present, including which specific type of PH may be present.
AI tool training will be based on 5 groups (each group anticipated to contain 415 echocardiograms): mild pre-capillary PH; moderate pre-capillary PH; severe pre-capillary PH; post capillary PH; no PH. The tool will then be validated in a separate pool made up of 425 echocardiograms (a combination of pre-capillary, post capillary PH and no PH). The validation cohort will not have been used in the training stage.
Inclusion Criteria:
Exclusion Criteria: