In patient care, particularly in ICUs, monitoring patients on a 24-hour basis to detect signs of state deterioration or imminent complications is a critical and, in acute cases, a life-saving task.
Manual monitoring by specially trained and highly experienced ICU personnel complements devices such as EEGs that monitor vital signs in order to address issues such as false alarms, misdetections and overall patient state assessment.
Full-time bedside care is straining and cost-ineffective. Severe burnout syndrome is present in about 50% of critical-care physicians and in one third of critical-care nurses. ICU personnel rely on the alarm systems of the devices that monitor vital signs, but such devices tend to be weighted toward sensitivity rather than specificity, thus generating numerous false alarms. ICU personnel are trained to filter those alarms subconsciously, but they may inevitably miss critical alarms due to conflicting priorities or improper filtering.
Video monitoring can complement existing monitoring systems by covering the idle times of manual inspection and thus detecting signs of patient state deterioration that can be missed either by the devices that monitor vital signs or by the ICU personnel themselves.
Tele-ICU solutions based on video surveillance have been introduced in numerous ICUs across the globe. They have been proved to reduce patient mortality and length of stay as well as improve personnel process compliance, in addition to saving costs.
Most video-monitoring solutions available today provide remote visualization of the patient bed and are easily observable by ICU personnel. Beyond patient-status monitoring, several other patient-care benefits are starting to emerge, including medication management and review or care plans based on collection and evaluation of patient video streams .
However, video monitoring of patients generates petabytes of data per day, adding a significant additional burden to the Big Data problems of the ICU cockpit framework.
Nevertheless, this data is rich and typically contains critical information that can — in itself but even more so in combination with vital signs and patient information — enhance the system’s capability to detect critical states in an early, robust and personalized way. This stands to improve patient care and personnel effectiveness significantly.
Extracting critical information from a video stream, detecting patterns of patient state deterioration and linking those patterns to the basic principles that underlie state changes and intervention-critical cases are aspects of cognitive computing that lead to data-driven insights.
In this project, we focus on detecting epileptic seizures in patients suffering from status epilepticus by analyzing patient-monitoring video streams.
Previous work detected objects and motion in the patient bed scene and analyzed motion trajectories, motion patterns and motion segmentation to detect seizures. The use of markers or color amenities to make objects more clearly detectable have also been studied.
Alternative methods use optical flow or spatio-temporal interest points. The main technology in this context is to classify patient states. Previous studies focused on a subset of epileptic seizures, based on short video segments of infants or well-defined and marked adults.
In this project, we focus on a neuro-ICU where epileptic seizures occur with high frequency and in a controlled environment. The goal is to detect eminent seizures in patients with known epilepsia in a patient bed scene equipped with numerous devices to monitor vital signs and an active scene where ICU personnel intervene regularly to check the patient and the monitoring devices.
We employ sparse-coding and machine-learning technologies to model neuro-ICU patient states to detect false alarms and epileptic seizure episodes. The technology we are developing addresses and complies with patient privacy issues.