Can we use big data to help accurately diagnose and predict the outcome of stroke?

Assessment of modern machine learning methods and conventional statistical regression techniques in diagnosis and prediction of outcome after acute stroke using big data

A stroke happens when the blood supply to part of the brain is cut off, causing brain cells to become damaged or die. In a mini stroke, also known as a transient ischaemic attack (TIA), the disruption in blood flow is only temporary and people normally recover within 24 hrs. Diagnosis of stroke and TIA can be challenging, partly because the symptoms of other conditions (e.g. seizures, migraines) are similar. These ‘stroke mimics’ are frequently misdiagnosed as stroke. However, the early diagnosis of stroke and TIA can help healthcare professionals ensure optimal treatment. Predicting early complications and later outcomes is also challenging. But, predicting complications and the longer-term outcomes of stroke and TIA can help manage condition. Professor Philip Bath will test state-of-the-art machine learning and statistical approaches for diagnosing stroke/TIA vs. mimics and predicting outcomes after stroke. The mathematical models will be developed and tested using reliable data from a large number of clinical trials in stroke. Information such as brain scan data, gender, medical history and post-stroke outcomes from up to 90,000 people will be analysed. If the machine learning and/or statistical analyses provide accurate diagnoses and outcome prediction, then they could be integrated into routine clinical care as apps on phones and computers. Their use could help ensure people receive the most appropriate treatment and lead to substantial NHS savings.