An Algorithm for the Detection of General Movements of Preterm Infants Based on the Instantaneous Heart Rate

Abstract

Video recording and editing of general movements (GMs) takes time. We devised an algorithm to automatically extract the period of GMs emergence to assist in the assessment of GMs. The algorithm consisted of δHR: subtracting the moving average heart rate (HR) for the past 60 s from the average instantaneous HR; and %δHR: the percentage of the instantaneous HR to the moving average HR. Ten-second sections in which δHR was positive for three consecutive sections and contained at least one section with %δHR > 105% were extracted. Extracted periods are called automated extraction sections (AESs). We evaluated the concordance rate between AESs and GMs in three periods (gestational age 24–32, 33–34, and 35–36 weeks). The records of 84 very low birth weight infants were evaluated. Approximately 90% of AESs were accompanied by GMs at any period in both the supine and prone positions. The proportion of full-course (beginning to end) GMs among GMs in the AES was 80–85% in the supine position and 90% in the prone position in all periods. We could extract a sufficient number of assessable GMs with this algorithm, which is expected to be widely used for assisting in the assessment of GMs.

Keywords

• general movements
• heart rate
• accelerations
• automated extraction
• supine position
• prone position