Over the last decade, the clinical and scientific value of GMA has been increasingly recognized and it has been revealed to be the most sensitive and specific predictor of later disabilities [
7,
10]. The current international guidelines emphasize that GMA during the FMs period has the best predictive value and accuracy for the early detection of CP and high-risk CP [
44,
45]. In a systematic review, it was reported that the method with the best predictive power for early CP diagnosis was Prechtl’s GMA (sensitivity: 97%, specificity: 89%) performed in the FMs period [
10]. Furthermore, as its associations with later cognitive, speech-language, and motor functions in addition to CP have become more evident, the merits of GMA for early detection of disabilities have been increasingly recognized [
5‐
8].
Given the GMA during the FMS period is a strong predictor for neurodevelopmental disorders, it is extremely important to automate GMA and make it easily accessible. Therefore, with the help of today's technological advancement, researchers have sought automated methods that are easy to use, widely utilized in clinics, independent of the user/observer, and do not interfere with infant movements [
16,
17], and they have suggested several methods. Adde et al. used the general movements toolbox (GMT), a computer-based video analysis method, to distinguish between absent FMs and normal FMs classified according to the observational GMA, and noted that absent FMs were predicted with sensitivity and specificity values of 82% and 70%, respectively [
15]. In another study of the same research group, it was reported that GMT predicted absent FMs with lower sensitivity (80%) and specificity (53%) compared to the previous study [
46]. Machireddy et al. performed a 3D spontaneous movement analysis using a hybrid system that includes an advanced wearable sensor (3D-accelerometer, 3D-gyroscope, and 3D-magnetometer) and computer-based video analysis methods. They stated that absent FMs and normal FMs were classified with an accuracy of 84% using a machine-learning algorithm [
47]. Gao et al. evaluated the spontaneous movements of 1–6 month-old infants with an accelerometer, regardless of FMs and writhing periods, and found that abnormal and normal GMs were discriminated with an accuracy of 80% [
14]. In the previous study, the inclusion of abnormal GMs belonging to two different GMA periods, such as poor repertoire GMs and absent FMs with quite different CP predictive values, in the same group made the interpretation of the results difficult. On the other hand, some limitations of the wearable sensors and GMT used in previous studies raise concerns for clinical use. As previously mentioned, in observational GMA, the infant is in the supine position and untouched, moving free of any external stimulus, and should also be in an appropriate behavioral state. However, it is not known whether these sensors affect infants' spontaneous movements or the infant's wearing procedures, during which the infant has to be touched or manipulated, and often time-consuming could affect the infant's behavioral state in wearable sensor-based approaches [
17]. In addition, although the definition of GMs includes the movements of all body parts, sensor-based approaches usually consider the movements of the arms, legs, and head, not the trunk movements, which indicates that they provide incomplete information about the full-body movement. GMT requires special setup and evaluation conditions, and it is difficult to detect small and fast changes in movements due to 2D video recordings [
23,
48,
49]. In the present study, an automated method using the COP was presented and values of predictive power comparable to previous studies were obtained in the classification of normal FMs and absent FMs (accuracy: 83%, sensitivity: 85%, specificity: 83%). The increase and decrease in the instantaneous velocity R (Std) (velocity variability) were interpreted in favor of normal FMs and absent FMs, respectively. This result supports the reduced variability in absent FMs [
4,
11]. In addition, the increase in the approximate entropy R, which evaluates the movement complexity, increased the risk of absent FMs. It can be inferred that infants exhibiting absent FMs, unlike the complexity in normal FMs, have a chaotic or excessive movement complexity. Moreover, we claim that this presented method has some advantages. First, it is non-intrusive, which means no sensor attachment is necessary. Second, in contrast to method requring sensory attachment, it deals with the global movement of body parts, not separately, in accordance with the linguistic definition of GMs. Lastly, it is easy to use and does not requiring a special setup and lab environment.