Machine learning (ML) is a tool for working with huge amounts of data that would otherwise be a difficult task. Its use is surging and it is now also being used in medicine. Due to technology being ever-present in medicine, there is already an abundance of data ready to be used in ML. This thesis works with data of patients with hypertrophic cardiomyopathy, a cardiovascular disease. A system was developed that observes consecutive measurements of a patient and alerts us in case of significant worsening. The severity of the patient's condition is estimated using a pretrained prediction model, which predicts the chances of patient soon having a key event such as a cardiac arrest. The changes in estimations are detected using the Page-Hinkley method. Incremental learning was also implemented. The system was tested using various prediction models. The results also demonstrate the importance of different components of the system for its overall usefulness. The best result achieved an F1 score of 0.357, with a sensitivity of 0.426, a precision of 0.308 and a specificity of 0.976. This result was achieved using logistic regression.
|
