Measuring facial grimacing for quantifying patient agitation in critical care

Abstract

The effective delivery of sedation in critical care relies primarily on an accurate and consistent measure of a patient's agitation level. However, current methods for assessing agitation are subjective and prone to error, often leading to over sedation or cycles between agitation and oversedation. This paper builds on previous work developing agitation sensors based on heart rate and blood pressure variability, and overall whole body motion. In this research, the focus is on real-time measurement of high resolution facial changes that are observed to occur in agitation. An algorithm is developed that measures the degree of facial grimacing from a single digital camera.The method is demonstrated on simulated patient facial motion to prove the concept.A consistent measure is obtained that is robust to significant random head movement and compares well against visual observation of different levels of grimacing. The method provides a basis for clinical validation.

Pain Level Detection From Facial Image Captured by Smartphone

“…A Bayesian extension of SVM named relevance vector machine (RVM) is also used [31] to make high classification accuracy. Niese et al and Becouze et al used ICU/post-operative patients [32], [33] to measure pain experienced. Becouze et al used the idea based on the hypothesis that pain results in forming extra wrinkles [32].…”

“…Niese et al and Becouze et al used ICU/post-operative patients [32], [33] to measure pain experienced. Becouze et al used the idea based on the hypothesis that pain results in forming extra wrinkles [32]. Photogrammetric technique is used by Niese et al for finding features.…”

“BAM!” Depth-Based Body Analysis in Critical Care

“…Since the body is generally occluded by the blanket, bed occupancy is estimated by detecting the patient's head with skin color models [9] or markers [2]. Neither approach is practical (color is not available by night).…”

“…Current approaches require markers [2,8], active management by medical staff [5,16] and/or color [2,8]. We identified the three main problems for computer vision ICU monitoring:…”

“…Additionally, some sensors are intrusive and therefore not used for critical care. Computer vision monitoring systems are increasingly used [2,5,8,13,16] as they are easy to install, non-intrusive, and the versatility of the sensor allows them to handle a wide variety of tasks [8,11]. Agitation is the most common actigraphic cue, however there is no "golden standard" to quantify it [4] and most computer vision algorithms provide view-dependent custom measurements.…”