Skip to main content
. Author manuscript; available in PMC: 2021 Mar 4.
Published in final edited form as: Res Dev Disabil. 2021 Feb 8;110:103854. doi: 10.1016/j.ridd.2021.103854

Table 2.

Video-based technological approaches for studying GMs.

ID Reference* Camera Type;
Technique;
Predictions 		Features (F): Quantity / Description;
Body Parts (BP)		 Classification Methods Age;
Sample Size 		Accuracy (A);
Precision (P);
Recall (R);
Specificity (S)
Conventional Machine Learning Classification
1 Baccinelli et al., 2020 2D Video;
Movement Detection;
Extract movement features 		F: 7/trajectory, motion, image;
BP: hands and feet 		Extraction of quantitative measures 39–41 weeks (GA);
300 videos (90 infants high risk ASD) 		NR (ICC: 87–98)
2 Caruso et al., 2020 103 videos (53 low risk, 50 high risk ASD)
3 Doroniewicz et al., 2020 2D Video;
Pose Estimation;
Classify WMs and PR 		F: 16/scope, nature, and location of each limb’s movement;
BP: limbs 		SVM-RBF, RF, LDA 38–42 weeks (GA);
31 videos 		(SVM): A:80;P:64;R:71;S:83 (RF): A:81;P:53;R:44;S:93 (DA): A:80;P:50;R:40;S:94
4** Tsuji et al., 2020 2D Video;
Movement Detection;
Classify GMs 		F: 25/movement magnitude, balance, rhythm, body centre;
BP: limbs 		LLGMN 25–40 weeks (GA), 0–15 weeks (PTA), NR for half of the infants;
47 videos (21 infants) 		A:91;P:NR;R:NR;S:NR
5 Schroeder et al., 2020 RGB-D Video;
Shape and Pose Estimation;
Classify GMs 		F: 6890/SMIL;
BP: 23 joints 		RGB-D, 3D SMIL (Auto-Generated) 2–4 months (PTA);
29 videos (high risk CP) 		A:80;P:NR;R:NR;S:92
6 Hesse et al., 2019 Custom Model 2–4 months (PTA);
12 videos 		NR (PCkh 2.0, P:90)
7 Hesse, Boden-steiner, et al., 2019
8 Hesse, et al., 2018 2–4 months (PTA);
136 videos (37 infants)
9 Hesse, Schroeder, et al., 2018
10 Hesse et al., 2017 F: NA/Random Ferns;
11 Hesse et al., 2015 NR;
1 infant (3D model) 		NA
12 Ihlen et al., 2020 2D Video;
Movement Detection CIMA (MEMD);
Predict CP 		F: 990/Optical Flow, BP: head, trunk, limbs LDA 9–15 weeks (PTA);
377 videos (high-risk CP) 		A:93;P:NR;R:NR;S:82
13 Adde et al., 2018 2D Video;
Movement Detection;
Quantify FMs vs WMs, Classify GMs 		F: NR/spatial (no temporal), CSD;
BP: head, trunk, limbs 		LR, Variability of CSD 3–5,10–15 weeks (PTA);
54 videos (27 infants preterm) 		NR (CSD is 7.5% lower during FMs in comparison to the WMs period)
14 Støen et al., 2017 2D Video;
Movement Detection;
Detect FMs 		F: NR/spatial and temporal, CSD;
BP: neck, trunk, limbs 		Variability of CSD 10–15 weeks (PTA);
241 videos (150 infants: 48 abnormal) 		NR (CSD varies between R:80;
S;80–90)
15 Rahmati et al., 2016 2D Video;
Movement Detection;
Predict CP 		F: NR/Optical Flow, FFT;
BP: hands, feet, head, trunk, arms 		SVM, MRF, Particle Matching 2–4 month (PTA);
78 videos (78 infants: 14 CP) 		(SVM) A:91;P:NR;R:86;S:92
16 Rahmati et al., 2015 2D Video;
Movement Detection;
Predict CP 		F: NR/Optical Flow;
BP: hands, feet, head, trunk, arms 		(SVM) A:87;P:NR;R:NR;S:NR
17 Rahmati, Amo, et al., 2014 20 Video;
Movement Detection;
Predict CP 		F: NR/LDOF, graph-cut;
BP: hands, feet, head, trunk 		SVM, MRF A:87;P:NR;R:50;S:95
18 Rahmati, Dragon, et al., 2014 A:NR;P:96;R:NR;S:NR
19 Adde et al., 2013 2D Video;
Movement Detection;
Detect FMs, Predict CP 		F: NR/motion, Cs, Qmean, Qsd CPP;
BP: neck, trunk, limbs 		CPP 9–17 weeks (PTA);
104 videos (52 infants: 24M, 28F) 		(FMs) A:NR;P:NR;R:89;S:79 (CPP) A:NR;P:NR;R:89;S:74
20 Stahl et al., 2012 2D Video;
CIMA;
Detect FMs, Predict CP 		F: 3/Optical Flow (GPU), wavelet, spatio-temporal;
BP: head, limbs 		SVM 10–15 weeks (PTA);
136 videos (82 infants: 15 atypical, 67 typical) 		A:96;P:NR;R:88;S:98
21 Adde et al., 2010 2D Video;
Movement Detection;
Predict CP 		F: NR/CPP, CSD, VSD, ASD, Qmean, Qmedian, QSD;
BP: neck, trunk, limbs 		CPP 10–15 weeks (PTA);
30 videos (high-risk: 13M, 17F) 		(CPP) A:NR;P:NR;R:85;S:88
22 Marchi et al., 2020 SMART-D Video (10 cameras + markers);
Movement Detection;
Correlate FMs age with other measures 		F: NR/coordination, distance, global movement quality;
BP: hands and feet 		Custom Model 9–20 weeks (PTA);
8 videos 		NR (Regression, R2:97)
23 Marchi et al., 2019 2D Video;
Pose Estimation;
Classify GMs 		F:NR/OpenPose;
BP: 25 joints 		Extraction of quantitative measures 8–17 weeks (PTA);
21 videos (14 typical, 7 atypical)
24** Chambers et al., 2019 2D Video;
Pose Estimation;
Estimate risk 		F: 38/OpenPose and kinematics, NGBS;
BP: 25 joints 		Naive Bayes, Kinematics Data 4–11 months;
104 videos: 85 Youtube, 19 clinical 		A:NR;P:92;R:94;S:NR
25 Dai et al., 2019 2D Video;
Movement Detection;
typical vs atypical 		F: NR/Wavelet, PCA;
BP: neck, trunk, limbs 		SVM, XGBoost 10–12 weeks (PTA);
120 videos (60 typical, 60 atypical) 		A:93;P:NR;R:95;S:92
26 Gajniyarov et al., 2019 2D Video Movement Detection;
Analyse GMs 		F: NR/segmentation, wavelet, limb speed;
BP: hands and feet 		Data Pre-processing 10 weeks (PTA);
18 videos 		NR (study on data preprocessing)
27 Raghuram et al., 2019 2D Video;
Movement Detection;
Detect atypical 		F: 289/skin model, LDOF;
BP: neck, trunk, limbs 		Logistic Regression 3–5 months (PTA);
152 videos 		A:66;P:NR;R:79;S:63
28 Orlandi et al., 2018 F: 643/skin model, LDOF;
BP: neck, trunk, limbs 		AdaBoost, Random Forest 3–5 months (PTA);
127 videos (98 typical, 29 atypical) 		A:92;P:NR;R:44;S:88
29** Das et al., 2018 2D Video;
Movement Detection;
Detect kicks 		F: 5/KAZE, legs in same y-direction;
BP: lower limbs 		SVM 4–7 months (PTA);
16 videos 		A:91;P:88;R:85;S:NR
30 Cenci et al., 2017 RGB-D Video;
Movement Detection;
Probability of change 		F: 10/velocity, acceleration amplitude, volume;
BP: limbs 		K-means, Markov Chains 37–38 weeks (GA);
35 videos (1 infant) 		NR (initial test-phase)
31 Machireddy et al., 2017 2D Video;
Movement Detection;
Detect FMs 		F: NR/sensor fusion, EKF;
BP: limbs 		SVM 2–4 months;
20 videos 		A:84;P:NR;R:NR;S:NR
32 Marschik et al., 2017 2D Video;
Multimodal Detection;
NA 		F: NR/multimodal fusion;
BP: the whole body 		Heuristic 0–4 months;
NA 		NA
33** Shivakumar et al., 2017 RGB-D Video;
Movement Detection;
Track Body Attributes 		F: NR/Optical Flow;
BP: limbs 		Adaptive Window, K-means 3–11 months (PTA);
3 videos (typical) 		A:NR;P:NR;R:NR;S:NR
34** Serrano et al., 2016 RGB-D Video;
Pose Estimation;
Kicking Patterns Analysis 		F: NR/lower limb pose, RPSR;
BP: lower limbs 		Kicking Patterns of Robot NR;
1 robotic infant 		NR (qualitative analysis)
35** Olsen, 2015 RGB-D Video;
Pose Estimation;
Detect Kickings 		F: NR/Optical Flow;
BP: stomach, head, limbs, feet 		K-NN, Classification Tree, SVM 1–6 months;
11 videos 		A:90;P:NR;R:NR;S:NR
Deep Learning Classification
36 McCay et al., 2020 2D Video;
Pose Estimation;
Classify GMs 		F: NR/OpenPose, HOJO2D, HOJD2D;
BP: 14 joints 		FCNet model 2–4 months (PTA);
12 videos 		A:NR;P:NR;R:NR;S:NR
37 McCay et al., 2019
38 Moccia et al., 2020 RGB-D Video;
Pose Estimation;
Detect Joints 		F: NR/spatio-temporal;
BP: shoulders, elbows, wrists, hips, knees, ankles 		Dual CNNs 31–36 weeks (GA);
16 videos 		A:NR;P:NR;R:NR;S:NR
39 Moccia et al., 2019
40 Schmidt et al., 2019 2D Video;
Movement Detection;
Classify GMs 		F: NR/OpticalFlow, FFT, Keras VGG19;
BP: limbs 		LSTM 2–4 month (PTA);
78 videos (78 infants: 14 CP) 		A:65;P:NR;R:51;S:27
*

Articles are first arranged in descending order of the publication year, followed by ascending order of the last name of the first author. Studies with an inherent connection, i.e., leading authors are identical or worked jointly, are stacked together and shaded with the same background colour, also ordered first by the publication year and then by the last name of the first author.

**

Studies in which the ages of the participants fell (partly) beyond the appropriate range according to the standard GMA (Einspieler et al., 2014), or the age range was (partly) missing.

Key of Terms.

Generic: ASD – Autism Spectrum Disorder; CP – Cerebral Palsy; CS – Cramped Synchronised; FM – Fidgety Movements; GA – Gestational Age; GMS – General Movements; GMA – General Movement Assessment; NA – Not Applicable; NR – Not Reported; PTA – Postterm age. PR – Poor Repertoire; WM – Writhing Movements.

Techniques and Models: ASD – Acceleration Standard Deviation; CIMA – Computer-based Infant Movement Assessment; CPP – Cerebral Palsy Predictor; CSD – Standard Deviation of the Center of Motion; FFT – Fast Fourier Transformation; HOJD2D – Histograms of Joint Displacement 2D; HOJO2D – Histograms of Joint Orientation 2D; ICC – Intraclass Correlation Coefficient; LDA – Linear Discriminant Analysis; LDOF – Large Displacement Optical Flow; LLGMN – Log-linearised Gaussian Mixture; LR – Logistic Regression; MEMD – Multivariate Empirical Mode Decomposition; MRF – Multi-label Markov Random Field; NGBS – Naive Gaussian Bayesian Surprise; PCKh 2.0 – Percentage of Correct Keypoints in Relation to Head Segment Length (two times the head segment length); QMEAN – Quantity of Motion Mean; Qmedian – Quantity of Motion Median; QSD – Quantity of Motion Standard Deviation; RBF – Radial Basis Function Kernel; RF – Random Forests; RPSR – Robust Point Set Registration; SMIL – 3D Skinned Multi-Infant Linear (Based on SMPL Model for Adults); SMPL – Skinned Multi-Person Linear Model; SVM – Support Vector Machine; VSD – Standard Velocity Deviation.