The challenge
Cardiovascular disease (CVD) is the leading cause of death in Europe and costs the economy nearly
€ 200 billion each year. Consequently, there is an urgent need for reliable diagnostic tests which identify patients at risk of CVD, so they can be given the appropriate treatment at an earlier stage, increasing their chance of survival and reducing the cost burden of administering unnecessary treatment. Perfusion, the flow of blood and hence oxygen, is essential to the functioning of the heart, and reduced perfusion, or ischemia, is an early marker of CVD. Accurately measuring perfusion can indicate areas with inadequate blood supply, and therefore patients at risk.
However, the current ‘gold standard’ for accurately quantifying cardiac perfusion is through invasive measurements with catheters which is both costly and has undesirable side-effects. There are non-invasive medical imaging modalities, such as Positron Emission Tomography (PET), Magnetic Resonance Imaging (MRI) and Computed tomography (CT), from which a perfusion measurement can be determined. From the image acquired, a cardiac perfusion map is derived which is used to make clinical decisions. The results can vary significantly between imaging techniques and indeed within the same imaging technique applied at different centres. In addition it can take years of experience to accurately diagnose a patient based on a perfusion map as it is hard to detect small changes in blood flow visually.
The solution
The EMPIR project Perfusimaging (15HLT05) developed physical standards and data analysis tools, applicable to a range of imaging modalities to support traceability and reliability of clinical data. NPL led the part of the project on data analysis and uncertainty quantification, which included developing a method to classify patients into diseased or not-diseased groups accounting for uncertainty associated with perfusion measurements. Accounting for measurement uncertainty can help make clinical decisions with more confidence, especially in the case of borderline patients where the diagnosis is not clear, and to quantify the risk associated with the misdiagnosis of a patient.
There are a number of steps involved in generating a perfusion map – from patient preparation to data analysis – each associated with potential sources of uncertainty. Information about the uncertainties associated with perfusion measurements is not usually taken into account when patients are diagnosed, which can be detrimental to the decision-making process. Using uncertainty information can help to quantify the risk of misdiagnosis of patients, especially in borderline cases. If the risk is found to be high, a clinician can then order further tests rather than relying on a single measured value of blood perfusion where the uncertainty is unquantified.
A risk-based decision-making framework was developed which looked at the extent of myocardial perfusion defects and reflected the uncertainty in the myocardial perfusion values. By combining patient data and expert insights, this framework could:
- Help less experienced clinicians make better decisions regarding patient health
- Serve as a starting point for further clinical investigation
- Be used as a screening to categorise patients so that the most severe cases can be prioritised on the clinical list
The impact
Working with senior scientists and clinicians, a decision-making framework based on the general principles of conformity assessment was developed. It was used to classify patients as diseased or not diseased with a level of confidence reflecting the measurement uncertainty associated with the perfusion measurements. If a decision is considered to be high risk or inconclusive, a clinician can then order further tests to support and improve the diagnosis of the patient.
The decision-making framework was applied to clinical data from patients participating in a cardiac PET perfusion study at the Turku PET Centre in Finland, as a first step towards applying metrology practices to improve confidence in clinical decision making.