Patients in the intensive care unit (ICU) can suffer severe clinical conditions such as seizure, delirium, etc. which are associated with motoric behaviors. From a clinical viewpoint, identification of motoric movements is needed for a prompt diagnosis. Video-based method provides an non-invasive solution for automatic and continuous monitoring of ICU patients. As the first step, we propose a video-based system to automatically detect patient movement events. Change detection is performed by means of both frame differencing and motion estimation using the video recordings of ICU patients. Post-processing procedures are investigated to remove spurious changes. Movement/nonmovement classifier distinguishes the change in each individual second using the Support Vector Machines (SVM). The optimal processing chain is derived based on leave-one patient-out crossvalidation. Frame differencing and motion estimation achieve comparable best motion detection performances of 87.24% and 87.31% accuracies, respectively. Finally, the movement events are obtained by merging and extending detected motions, and are compared with manually labeled events. A promising performance of 77.81% sensitivity is obtained.
|Date of Award||31 Dec 2014|
|Supervisor||Gerard de Haan (Supervisor 1), U. Großekathöfer (Supervisor 2), Esther van der Heide (External coach) & A. Heinrich (External coach)|