Automatic Multimodal Assessment of Neonatal Pain

McCafery [1] describes pain as, ”whatever the experiencing person says it is, existing whenever the experiencing person says it does.” Unfortunately, neonates do not have the ability to communicate this experience verbally (self-evaluation) or non-verbally by writing or pointing (Visual Analog Scale). The limited ability of neonates to communicate pain and the earlier misconception about the absence of neurological substrate for the perception of pain in neonates have led pediatricians to believe, for several decades, that neonates do not feel or remember pain. Several scientific studies [2; 3; 4; 5] disproved this earlier belief and reported a strong association between repeated pain exposure (under-treatment) and alterations in the brain structure and function. This association has led to the increased use of anesthetic medications. However, recent studies [6; 7] found that the excessive use of analgesic medications such as Morphine and Fentanyl may cause several side effects (e.g., hypotension and feeding intolerance). The current standard for assessing neonatal pain involves observing, by bedside caregivers, multiple behavioral (e.g., facial expression) and physiological (e.g., vital signs) responses of pain. At least 29 response-based pain scales [8] have been developed to evaluate procedural and postoperative pain in neonates. This practice has three main shortcomings. First, it relies on the caregiver’s direct observation and interpretation of multiple responses. It is highly affected by several idiosyncratic factors, such as the observer’s cognitive bias, identity, culture, and gender [9]. The interand intraobserver variations can lead to inconsistent assessment and treatment of pain. Second, caregivers assess pain at different time intervals. The discontinuity of assessment can lead to missing pain while the neonate is left unattended; therefore, it may result in delayed intervention. Third, this practice requires a substantial time commitment and a large number of well-trained caregivers to ensure the proper utilization of the pain scale. The substantial cost of this practice makes it infeasible in underdeveloped countries where medical professionals and resources are scarce. This dissertation introduces an automatic, comprehensive, and multimodal neonatal pain assessment system. The proposed system addresses the shortcomings of the current practice and provide continuous, consistent, and inexpensive pain assessment to guide treatment. The main contributions of this dissertation can be summarized as follows: