Pain Detection Through Automated Video Analysis

Overview

The study team has developed an algorithm for pain assessment based on automated video facial and body pose analysis. The investigators aim to assess the sensitivity of this algorithm in detecting pain in post-surgical patients and refine the algorithm to increase the sensitivity of pain detection in patients.

Post-surgical pain, if inadequately controlled, has deleterious short and long term consequences for the patient. Although most patients are able to report their pain scores, a minority are unable to do so and assessing their pain can prove to be a challenge for healthcare professionals. In recent years, facial recognition tools have been developed based on the premise that subtle facial variations signifies pain. However, changes in body, and head posture can also represent pain. As such, these tools are with their limitations and are only validated on certain groups of patients, thus may not be sensitive enough to detect pain in post-surgical patients. The first stage of the study will be conducted on 40 patients presenting for major gynaecological surgery, with the obtained data used to fine tune the algorithm. The patients will be video-taped pre-surgically in the pre-evaluation anaesthetic clinic and post-surgically in the ward. They will be asked to rate their pain scores on the numerical rating scale and fill in questionnaires on their psychological and quality of health status. The pain scores will be correlated with the results obtained from the pain assessment algorithm. The second phase will improve and enhance the model by (1) analysing body pose to improve the model performance; (2) validating the improved model by recruiting 200 patients undergoing surgical and pain procedures, inpatient and outpatient consultations to collect their videos before and after surgery and inpatient and outpatient pain consultations; (3) integrate the model into a standalone electronic application to improve its usability in both inpatient and outpatient settings. The third phase will recruit 130 male paediatric patients presenting for circumcision surgery to improve algorithm by i) Adding body posture analysis and other physiological measurement to further improve the performance of our model; ii) Developing our model for use in the pediatric population; and iii) Improving its usability in both clinical and non-clinical settings. Deidentified keypoints will be extracted from the videos to further validate the model.

Conditions

Interventions

Eligibility

Locations (2)

Outcomes