Pain typically is measured by patient self-report or clinician impressions, either through clinical interview or the visual analog scale. While useful, self-report measures have several limitations that include idiosyncratic use, inconsistent metric properties across scale dimensions, efforts at impression management or deception, and differences between clinicians' and sufferers' conceptualizations of pain. Given individual differences among patients, their families, and healthcare providers, pain is often poorly assessed, underestimated, and inadequately treated [Hadjistavropoulos et al., 2002]. An objective, reliable, and efficient assay of the occurrence and intensity of pain would advance pain monitoring and intervention, leading to a uniform yet individual management of pain and an efficient utilization of the health care system. Machine analysis for the automatic assessment of pain occurrence and intensity could be of considerable value to achieve these goals. To advance pain assessment, monitoring, and intervention, we propose an automatic facial, head, and body movement analysis for the automatic assessment of the occurrence and intensity of pain. Recent efforts demonstrated the validity of automatic detection of shoulder pain from facial expression in a constrained laboratory context using person-dependent, semi-automatic procedures. The current project extends previous efforts to multimodal measurement of the occurrence and intensity of pain in well-characterized and clinically relevant participants with rotator cuff syndrome (RCS) using person-independent, fully-automatic procedures. To collect pain-related behavioral responses to a full range of pain intensities, we will abduct their shoulders pre- and post- a triangular forearm support (TFS) maneuver (a pain relief procedure, [Fishman et al., 2011]). Participants' face, head, and upper body movement will be recorded using synchronized high-definition digital video and Microsoft Kinect cameras during the abduction and flexion of their shoulders before and after TFS. To investigate possible placebo effect of TFS, comparison behavioral responses to a sham procedure of randomly selected RCS participants will be collected before the TFS procedure. To investigate the specificity of pain response to arm abduction, we will include a non-clinical group of participants without RCS or need for pain relief. To investigate the repeatability of the proposed automatic measures, we will obtain multiple movement trials from each participant during both pre- and post treatment. Face, head, and body movement will be automatically tracked using our newly fully- automatic method for face tracking and the automatic Kinect's body skeleton tracking. We will identify optimal features and classifiers for automatic measurement of occurrence and intensity of pain. To investigate validity, we will compare automated measurement with patient-rated visual analog scale and obtained continuous observer ratings of pain intensity from the video recordings. MANOVA will quantify the relationship between the individual modalities and their combination for pain measurement. HLM will investigate possible influence of participant characteristics (e.g. gender).
Pain typically is measured by patient self-report, but self-reported pain is difficult to interpret and may be impaired or in some circumstances not possible to obtain. We propose to use a clinically well-characterized data to develop an automatic multimodal method for the detection of occurrence and intensity of pain.
| Jonassaint, Charles R; Rao, Nema; Sciuto, Alex et al. (2018) Abstract Animations for the Communication and Assessment of Pain in Adults: Cross-Sectional Feasibility Study. J Med Internet Res 20:e10056 |
| Kacem, Anis; Hammal, Zakia; Daoudi, Mohamed et al. (2018) Detecting Depression Severity by Interpretable Representations of Motion Dynamics. Proc Int Conf Autom Face Gesture Recognit 2018:739-745 |
