Automated Pain Assessment In Neonates

The current practice of assessing infants' pain is subjective and intermittent. The misinterpretation or lack of attention to infants' pain experience may lead to misdiagnosis and over-or under-treatment. Studies have found that poor management and treatment of infants' pain can cause permanent alterations to the brain structure and function. To address these shortcomings, the current practice can be augmented with an automated system to monitors various pain indicators continuously and provide a quantitative assessment. In this paper, we present methods to analyze infants' crying sounds, and other pain indicators for the purpose of developing an automated multimodal pain assessment system. The average accuracy of estimating infants' level of cry was around 88%. Combining crying sounds to facial expression, body motion, and vital signs for classifying infants' emotional states as no pain or severe pain yielded an accuracy of 96.6%. The reported results demonstrate the feasibility of developing an automated system that integrates multiple pain modalities for pain assessment in infants.