Opportunity
Stroke is the second leading cause of mortality and the third leading cause of disability worldwide. There are 13.7 million new stroke cases worldwide every year, and 5.5 million people die from stroke in a year. Stroke risk assessment is a fundamental component of disease prevention, yet few risk scores are available to estimate the risk of stroke, still fewer used a comprehensive set of medical data.
This invention will provide improved predictability on patient mortality, propose optimal treatment decision and LVO(Large Vessel Occlusion) prediction, through the use of machine learning and DNN (Deep Neural Network).
Technology
While existing predictive methods employ limited data sets, this invention will use independent variables (features) based on biological, clinical priori and statistical significance for stroke risk prediction model development. It also employs machine learning algorithms and DNN convergence to better improve prediction accuracy.
The method identifies and combines stroke related demographic, lifestyle risk factors and biomarkers data available through the clinical, image and text data using feature selection and machine learning methodologies. The training can be supervised and un-supervised. With this automated prediction algorithm , clinicians will be provided a well-informed decision support system towards different screening approaches including invasive procedures.
Mortality prediction, ICD-9 prediction and LVO prediction will also be covered by this invention.
Advantages
- Method is robust and cost-effective that makes use of comprehensive set of relevant patient related clinical, image and text data, yielding better predictive ability.
- This invention does not rely on any segmentation or selection of scans to facilitate model training.
- The machine learning (ML) based models together with DNN converge using supervised, unsupervised data has advantage over most existing technology which only targeted a few ML algorithms that only capable of providing individual performance result per ML method.
Applications
- The method can predict stroke risk, prognosis and classify stroke subtype clinically. It may someday be a warning App to person with high risk factors to be alerted to consult medical advices prior to actual onset of stroke.
