Using Artificial Intelligence to Identify the Associations of Children’s Performance of Coloring,Origami,and Copying Activities With Visual

Abstract

Importance: Performance of coloring, origami, and copying activities reflects children’s visual–motor integration (VMI), but the levels of association remain unclear.

Objective: To use artificial intelligence (AI) to investigate associations of performance of coloring, origami, and copying activities with VMI.

Design: Cross-sectional study.

Setting: Kindergartens.

Participants: A sample of 370 children (182 boys and 188 girls) in the second and third years of kindergarten.

Outcomes and Measures: Beery–Buktenica Developmental Test of Visual–Motor Integration, 6th Edition (VMI–6).

Results: Data for preschool children from an ongoing project were retrieved. AI models were trained to use photographs of activity products to predict total score on the VMI–6. R ² values were used to identify the variance in VMI–6 standardized scores explained by predicted scores from the activities. That is, R ² values reflected associations between activity performance and VMI. The R ² values for the combination of origami and copying were the largest (.390–.577). These R ² values were larger than those for each individual activity (.340–.473) and similar to those for the combination of all three activities (.400–.550).

Conclusions and Relevance: Because moderate R ² values were found between performance of the three activity products and VMI, the three activities have high potential for use in identifying children’s level of VMI or as teaching materials to facilitate the development of children’s VMI. Furthermore, combining origami and copying activities is recommended for teachers and clinicians who need to address VMI.

What This Article Adds: A combination of origami and copying activities had the strongest associations with children’s VMI. Teachers and clinicians can use these two activities when addressing VMI development among preschool children.

This study found that a combination of origami and copying activities had the strongest associations with children’s visual–motor integration (VMI). Teachers and clinicians can use these two activities when addressing VMI development among preschool children.

Visual–motor integration (VMI) refers to an individual’s coordination of visual information with limb movements (Gabbard et al., 2001). This ability is critical to children’s handwriting skills and influences children’s academic achievement (Daly et al., 2003). VMI develops in early childhood and can be refined and improved by practice and exposure to appropriate activities, such as scribbling, drawing, origami, and cutting and copying shapes (Coutinho et al., 2017; Dankert et al., 2003; Lee, 2022). Therefore, it is essential to identify activities relevant to VMI for teachers, clinicians, and caregivers to use as evaluation tools to assess VMI development or as teaching materials to facilitate VMI development.

Coloring, origami, and copying activities are popular with preschool children. Moreover, they are viewed as activities that are related to VMI because they are all guided by the eyes and performed by the hands. Children with learning disabilities and poor VMI have difficulty executing fine motor activities, such as drawing geometric forms, cutting with scissors, copying designs, and coloring (Sanghavi & Kelkar, 2005). Imaroonrak et al. (2018) investigated the effects of origami training on VMI among 15 preschool children age 5 yr. They found that after the intervention, the VMI posttest scores significantly differed between the experimental and control groups. Mattison et al. (1986) found that, compared with their peers, children with learning disabilities and visual–motor problems had significantly more trouble with design-copying tasks that involved visual–motor components. Lee (2022) compared the effectiveness of a computer-based intervention and a traditional activity-based intervention (including drawing, copying shapes, and coloring pictures of shapes) on the visual–motor abilities of children with autism. A significant improvement in visual–motor abilities was found in the therapeutic practice intervention group. Coutinho et al. (2017) recruited a control group who received traditional occupational therapy (including drawing, tracing, copying activities, etc.) that targeted visual–motor skills. Improvement in visual–motor skills after the intervention was found in both groups. Dankert et al. (2003) provided children with developmental delays with activities requiring fine motor and visual–perceptual skills: drawing and copying shapes, coloring shape pictures, and cutting shapes with scissors. Significant improvements were detected in children’s visual–motor and visual–perceptual skills. Accordingly, performance of these three activities reflects children’s VMI, and children’s development of VMI may be facilitated by performing the three activities.

However, associations between VMI and activity performance level remain unclear. The reason is that no objective criteria for identifying the quality of the products have been developed. As a result, the extent to which VMI is involved in these three activities remains unknown. A lack of knowledge of these associations may limit users’ confidence in the ability of these activities to reflect development of VMI. Therefore, a better understanding of the levels of association between VMI and performance on these activities could allow teachers and clinicians to use them to assess a child’s VMI development, and such knowledge could serve as a reference for choosing proper activities to facilitate VMI.

Artificial intelligence (AI) can be applied to investigate the relationship of activity performance to VMI. Two AI techniques—namely, computer vision and data prediction and classification—have been widely used in the educational and medical fields to diagnose diseases and simplify measures (Audibert & Maschio, 2021; Duran et al., 2022; Illavarason et al., 2019; Lin et al., 2023). Computer vision captures important features of images and transforms those features into numerical data. Data prediction and classification use numerical data to predict an outcome of interest. In practice, activity products (e.g., children’s colored or copied pictures) could be photographed and then transformed via computer vision into analyzable data. The technique of data prediction can be applied to analyzable data to predict an outcome of interest, for example, scores on a standardized assessment of VMI. Specifically, the process of data prediction is similar to that of traditional statistical analysis (i.e., regression analysis), so it can be used to examine the relationship of activity performance to VMI.

The purpose of this study was to use AI computer vision to first extract features from photographs of the products and then to use those features to identify relationships between performance on three activities (origami, coloring, and copying activities) and VMI.

Method

Participants

This was a cross-sectional study. In total, we retrieved the data of 370 children in the second and third years of kindergarten, collected from 2020 to 2021 as part of an ongoing AI project. Part of the data were reported in a previous study (Yu et al., 2021). The large sample size allowed the machine learning (ML) algorithms to have a higher predictive accuracy and stability to predict behavioral and cognitive scores for clinical patients (Cui & Gong, 2018). When a sample size larger than 200 was used in Cui and Gong’s (2018) study, there appeared to be no significant difference in predictive accuracy and stability. Therefore, the sample size in our study was adequate for the ML algorithms to make precise and stable predictions (Cui & Gong, 2018). The children whose data were retrieved were in the second and third years of kindergartens and lived in Kaohsiung or Tainan City. No children attended a special school. The children were able to participate in group activities and follow group instructions. This study was approved by National Taiwan University Hospital, a medical center in Taipei, and I-Da Hospital, a teaching hospital in Kaohsiung, Taiwan.

Measures

Origami Activity

In the origami activity, the children made a dog (Figure 1a) by following four steps. After the dog was completed, it was photographed eight times (four times from the front and four from the back, all at different angles). Details regarding the photographs are described elsewhere (Yu et al., 2021).

Figure 1.

Samples of products of the (a) origami, (b) coloring, and (c) copying activities.

Coloring Activity

A black-and-white picture of a train (Figure 1b) was provided to the children. The children were asked to color it while staying within the lines.

Copying Activity

As depicted in Figure 1c, a drawing of a person appeared in the upper half of a sheet of paper, and the children were asked to copy the drawing in a black box, using the same shapes and scale.

Visual–Motor Integration

The Beery–Buktenica Developmental Test of Visual–Motor Integration, Sixth Edition (VMI–6) was used to assess the children’s VMI ability (Beery, 2010). The VMI–6 contains 30 items. The first 3 items require children to scribble on a piece of paper, and the remaining 27 items require children to copy a shape. Each item is rated on a dichotomous scale (0 = incorrect, 1 = correct), and the total score can range from 0 to 30. A higher total score indicates greater ability. The VMI–6 has been reported to have good psychometric properties. Its internal consistency is acceptable to excellent, and it has good test–retest reliability (r = .88) and excellent intrarater reliability (r = .93; McCrimmon et al., 2012). The VMI–6 has good convergent validity with the Copying subtest of the Developmental Test of Visual Perception (Hammill et al., 1993) and the Drawing subtest of the Wide Range Assessment of Visual Motor Abilities (Adams & Sheslow, 1995).

Procedures

Kindergarten principals were contacted by research assistants for assistance in distributing research information to caregivers. Caregivers who agreed to participate in the study returned a signed informed consent form to the researchers. The children participated in group activities led by the researchers, with each group consisting of three to five children. The researchers led the children in performing origami, coloring, and copying activities and also administered the VMI–6. For each activity, the researchers first demonstrated the activity and then helped the children complete the activity with verbal guidance. No physical assistance was provided to the children during the entire process. The children spent 5 to 10 min on each activity and on the VMI–6. The total time to complete all procedures was about 30 to 45 min.

Statistical Analyses

The sample was split into a training sample and a test sample for further analyses. The training sample was used to train the models to connect activity performance (photographs of activity products) with VMI (scores on the VMI–6), and the test sample was used to identify the relationship of activity performance with VMI.

We further developed models of all combinations of the three activities to comprehensively identify their relationships to VMI. The models were conducted in two phases: (1) model training and (2) model testing. In these two phases, the same steps (Figure 2) were executed for both the training and the test samples: First, photographs of the activity products were transformed into features using the ResNet50 model. Second, features were used to predict total scores on the VMI–6 using the XGBoost model. The models were run via Google Colab. Third, the children’s ages and predicted scores on the VMI–6 were transformed into standardized ages and scores. Fourth, the standardized ages and standardized predicted scores were again used to predict standardized scores on the VMI–6. R ², a common index in the AI field, was used to identify associations between activity performances and the VMI–6. Values of R ² reflect the variance in the standardized scores for the VMI–6 explained by the predicted standardized scores for the VMI–6. A higher R ² indicates that VMI–6 scores were better explained by the predicted scores, and, therefore, higher R ² values represent greater associations between activity performances and the VMI–6 in our study.

Figure 2.

An artificial intelligence model example of investigating the relationship of activity performance to VMI.

On the basis of the results of our pilot study, two of the eight photographs for the origami activity provided better model performance for prediction. Therefore, for the origami activity, two of the eight photographs were transformed into numerical features. Because which photographic angle captured more information about the dog origami was unknown, we entered all possible combinations of any two of the eight photographs (28 combinations) into the models.

For the copying activity, we preprocessed the photographs (i.e., the colored photographs were converted into black-and-white images) to make the copied shapes clearer and then used the postprocessed images for further analyses.

Results

Participant Characteristics

In total, 370 children participated in the study. Nearly half were boys (n = 182; 49.2%), and the M age was 5.2 (SD = 0.63) years. Ages ranged from 40 to 77 mo. Most of the children (75.1%) were from Tainan City in southern Taiwan. Demographic characteristics of the participants are presented in Table 1.

Table 1.

Participant Characteristics (N = 370)

Variable	n (%)
Age, mo
40–47	7 (1.9)
48–53	35 (9.5)
54–59	79 (21.4)
60–65	104 (28.1)
66–71	90 (24.3)
72–77	55 (14.9)
Gender
Male	182 (49.2)
Female	188 (50.8)
Region
Kaohsiung	92 (24.9)
Tainan	278 (75.1)
Total VMI–6 score, M (SD)	12.85 (3.07)

Note. VMI–6 = Beery–Buktenica Developmental Test of Visual–Motor Integration, Sixth Edition.

Relationships Between the Children’s Performance of All Activity Combinations and VMI

Table 2 presents associations of the activity models in the training and test samples. For the training sample, R ² values reached .980 for all training models. For the test sample, the R ² value was highest for the origami activity (maximum R ² = .473), followed by the copying activity (.421), and that of the coloring activity was the lowest (.340). When combinations of two of the three activities were used, R ² values of the models increased, ranging from .300 to .577. For all three activities together, R ² values of the models ranged from .400 to .550. R ² values for the models of three activities together were close to those of the models combining the origami and copying activities.

Table 2.

R ² Values for Models of Origami, Coloring, and Copying Activities

Model	R ²
Model	Training Sample	Test Sample
Coloring activity	.980	.340
Origami activity	.980	.269–.473
Copying activity	.980	.421
Coloring and origami activities	.980	.300–.510
Coloring and copying activities	.980	.412
Origami and copying activities	.980	.390–.577
All three activities	.980	.400–.550

Discussion

The purpose of this study was to use AI to investigate the association of performance of coloring, origami, and copying activities with VMI. Our study produced three main findings. First, for individual activities, the origami activity had the strongest associations with VMI (i.e., higher R ² values), followed by the copying activity, and those of the coloring activity were the lowest. Second, when two of the three activities were combined, the associations of the combination of origami and copying with VMI were stronger than those of the other combinations and were also stronger than those of individual activities. Third, the associations of the combination of all three activities with VMI were close to those of the combination of origami and copying.

Associations were higher among the training sample than among the test sample because the model parameters were continually modified to develop the best model prediction in the model training phase. Associations between activity performance and VMI were overestimated for the training sample. Therefore, in this study, associations between activity performance and VMI refer to R ² values between each activity and VMI in the test sample.

Among the three activities, the children’s performance of origami and copying had stronger associations with VMI than did coloring alone. This result seems reasonable. On the one hand, origami involves a large amount of spatial ability and manual dexterity (Bae, 2013; Cakmak et al., 2014). The copying activity and the VMI–6 both required children to copy geometric shapes. On the other hand, coloring may require less visual perception (children only needed to identify shape boundaries), but it requires higher dexterity (control to color within the shapes). Moreover, because the prediction model revealed moderate associations between performances on the three activities and VMI, the results may support the use of all three activities to reflect VMI development.

Moreover, compared with individual activities, the combination of origami and copying had stronger associations with VMI. The reason could be that origami and copying activities involve different aspects of visual perception. For example, the visual perception skill required for origami is the ability to discern two-dimensional and three-dimensional spatial relationships (Cakmak et al., 2014), whereas the skills for copying may involve position in space, two-dimensional spatial relationships, and shape discrimination (Kulp & Sortor, 2003). Therefore, the features of the two activities may provide different information about children’s VMI. These results can serve as a reference for clinicians to comprehensively identify children’s VMI level. That is, clinicians may identify VMI level from both origami and copying activities.

We noted that associations of the combination of all three activities with VMI were not much higher than those of the combination of two activities. These results may be explained in two ways. First, the visual perception ability (e.g., identification of position in space) needed for the coloring activity was also needed for the origami and copying activities. As a result, adding features of the coloring activity to the origami and copying activities might not have provided more information about VMI. Conversely, features of the coloring activity may have provided noise, which might have interfered with training of the AI model. Second, the sample size may have been too small for the AI model because the features of the three activities were large. Therefore, more studies using larger sample sizes are suggested to investigate relationships of the combination of all three activities to VMI.

To the best of our knowledge, this is the first study to use AI to analyze children’s performance of activities and identify their relationship to VMI. This study makes three contributions. First, we developed an innovative approach to identifying children’s activity performance. Future studies can apply this approach to various performance areas. For example, children’s writing assignments can also be analyzed to identify their relationships to VMI or other outcomes of interest. Second, we found that the combination of origami and copying may be most appropriate for addressing children’s VMI. These two activities have high potential to be indicators of VMI. Third, the results of our study can also provide a foundation for future studies. For example, randomized controlled trials are recommended to investigate the effectiveness of using origami and copying activities as VMI interventions for preschool children.

Four limitations of our study must be noted. First, a group activity was used to collect the children’s activity products. Children’s performance in a group activity may be different from their performance in an individual activity, because children may be affected by their peers. Future studies should compare children’s performance of individual activities with VMI. Second, only data on children in the second and third years of kindergarten were retrieved for this study because, in our ongoing project, children in the first year of kindergarten did not participate in the copying activity. This may have influenced the generalizability of our results. Third, we used a convenience sample, which might have created sampling bias. For example, children in our study were from southern Taiwan and might differ from children living in northern Taiwan. Future studies are suggested to include children from other areas to increase the generalizability of the results. Fourth, we did not collect data related to children’s play experiences, which may have affected their performance of these activities. Future studies are suggested to include children’s play experiences as a covariate to clarify the associations between activity performance and VMI.

Implications for Occupational Therapy Practice

Teachers and clinicians can use the origami and copying activities together when assessing VMI development of preschool children.

Conclusion

We found that a combination of origami and copying activities had the strongest associations with children’s VMI. Moreover, the results of our study can provide a foundation for future studies using randomized controlled trial designs to investigate the effectiveness of applying origami and copying activities as VMI interventions with preschool children.

Footnotes

Acknowledgments

We appreciate all of the children and caregivers who participated in our study, and we gratefully thank the principals and teachers of the kindergartens. This study was supported by the Ministry of Science Technology, Taiwan (109-2636-B-214-001, 109-2314-B-038-147, 110-2636-B-002-023, 110-2628-B-038 -013, 111-2628-B-038 -021) and the Taipei Medical University (TMU109-AE1-B06).

References

Adams

, & Sheslow

(1995). Wide Range Assessment of Visual Motor Abilities. Wide Range.

Audibert

R. B.

, & Maschio

V. M.

(2021). FINNger—Applying artificial intelligence to ease math learning for children. arXiv. https://doi.org/10.48550.2105.12281

Bae

J. H.

(2013). The effects of origami on the improvement of hand dexterity. Journal of International Academy of Physical Therapy Research , 4 , 588–594. https://doi.org/10.5854/JIAPTR.2013.10.25.588

Beery

K. E.

(2010). Beery VMI: The Beery–Buktenica Developmental Test of Visual–Motor Integration. Pearson.

Cakmak

Isiksal

, & Koc

(2014). Investigating effect of origami-based instruction on elementary students’ spatial skills and perceptions. Journal of Educational Research , 107 , 59–68. https://doi.org/10.1080/00220671.2012.753861

Coutinho

Bosisio

M. E.

Brown

Rishikof

Skaf

Zhang

, . . . Dahan-Oliel

(2017). Effectiveness of iPad apps on visual-motor skills among children with special needs between 4y0m–7y11m. Disability and Rehabilitation. Assistive Technology , 12 , 402–410. https://doi.org/10.1080/17483107.2016.1185648

Cui

, & Gong

(2018). The effect of machine learning regression algorithms and sample size on individualized behavioral prediction with functional connectivity features. NeuroImage , 178 , 622–637. https://doi.org/10.1016/j.neuroimage.2018.06.001

Daly

C. J.

Kelley

G. T.

, & Krauss

(2003). Relationship between visual–motor integration and handwriting skills of children in kindergarten: A modified replication study. American Journal of Occupational Therapy , 57 , 459–462. https://doi.org/10.5014/ajot.57.4.459

Dankert

H. L.

Davies

P. L.

, & Gavin

W. J.

(2003). Occupational therapy effects on visual–motor skills in preschool children. American Journal of Occupational Therapy , 57 , 542–549. https://doi.org/10.5014/ajot.57.5.542

10.

Duran

Stark

Saglam

Semmelweis

Lioba Wunram

Spiess

, & Schoenau

(2022). Artificial intelligence to improve efficiency of administration of gross motor function assessment in children with cerebral palsy. Developmental Medicine and Child Neurology , 64 , 228–234. https://doi.org/10.1111/dmcn.15010

11.

Gabbard

Goncalves

V. M. G.

, & Santos

D. C. C.

(2001). Visual-motor integration problems in low birth weight infants. Journal of Clinical Psychology in Medical Settings , 8 , 199–204. https://doi.org/10.1023/A:1011373721162

12.

Hammill

D. D.

Pearson

N. A.

, & Voress

J. K.

(1993). Developmental Test of Visual Perception (2nd ed.). Pro-Ed.

13.

Illavarason

Arokia Renjit

, & Mohan Kumar

(2019). Medical diagnosis of cerebral palsy rehabilitation using eye images in machine learning techniques. Journal of Medical Systems , 43 , 278. https://doi.org/10.1007/s10916-019-1410-6

14.

Imaroonrak

Phunwutikorn

Phattharayuttawat

Ngamthipwatthana

Sumalrot

, & Auampradit

(2018). The effects of origami training on creativity and visual-motor integration in preschool children. Journal of the Medical Association of Thailand , 101 , S85–S89.

15.

Kulp

M. T.

, & Sortor

J. M.

(2003). Clinical value of the Beery Visual-Motor Integration Supplemental Tests of Visual Perception And Motor Coordination. Optometry and Vision Science , 80 , 312–315. https://doi.org/10.1097/00006324-200304000-00009

16.

Lee

S. C.

(2022). Comparison of computer-based and therapeutic activity interventions for improving visual motor abilities of preschoolers with autism: A randomized control study. Open Journal of Occupational Therapy , 10 , 1–11. https://doi.org/10.15453/2168-6408.1856

17.

Lin

G. H.

Lee

S. C.

Y. T.

, & Huang

C. Y.

(2023). Machine learning-based brief version of the Caregiver-Teacher Report Form for preschoolers. Research in Developmental Disabilities , 134 , 104437. https://doi.org/10.1016/j.ridd.2023.104437

18.

Mattison

R. E.

McIntyre

C. W.

Brown

A. S.

, & Murray

M. E.

(1986). An analysis of visual-motor problems in learning disabled children. Bulletin of the Psychonomic Society , 24 , 51–54. https://doi.org/10.3758/BF03330501

19.

McCrimmon

A. W.

Altomare

A. A.

Matchullis

R. L.

, & Jitlina

(2012). Test review: the Beery Developmental Test of Visual-Motor Integration. Sage.

20.

Sanghavi

, & Kelkar

(2005). Visual-motor integration and learning disabled children. Indian Journal of Occupational Therapy , 37 , 33–38.

21.