Text

Introduction: With the increasing awareness of mental health, the number of tools designed to measure mental health metrics, such as well-being or depression, is also increasing. In most cases, these tools rely on verbal and textual assessment scales [1], which can be a barrier, especially for children. This is because children may not be able to express their feelings appropriately or without inhibition due to a limited verbal repertoire [2]. Projective tests, such as the House-Person-Tree Drawing Test (HPT), could offer an easy-to-understand and engaging alternative. Given the current lack of digital solutions aimed at data collection and analysis of HPT drawings [3], this investigation presents a semantic scene segmentation approach for detecting age and gender differences in children's HPT drawings.

Methods: From a cross-sectional study, 303 HPT drawings from children (175 girls, 127 boys, and one of unknown gender) aged between 6 and 12 years (M = 8.5, SD = 1.14) were available. All drawings, except for the one where the child's gender was missing, were processed by a Scene Sketch Segmentation Model (SSSM) to calculate similarity matrices for the house, person, and tree objects. Using these matrices and a threshold of 0.71, at which the chosen SSSM performed best in its evaluation of performance with the test data [4], the number of pixels exceeding this threshold was determined. Subsequently, these pixels counts were analyzed using a two-way Multivariate Analysis of Variance (MANOVA), with age and gender serving as predictors.

Results: According to the results of the two-way MANOVA, there are significant multivariate effects. First, age exhibits a strong influence on the number of pixels (Pillai's Trace = .044, F = 4.525, p = .004). Second, with similar influence, gender is also a statistically significant predictor (Pillai's Trace = .061, F = 3.125, p = .005). An interaction effect between age and gender, however, could not be found (Pillai's Trace = .008, F = .817, p = .485).

Discussion: The analysis indicates that differences in pixel counts based on age and gender can be detected using a SSSM. The age effect, in particular, is in accordance with findings in children’s expressive happy and sad drawings [5], suggesting that children noticeably develop their drawing skills with age. But further research is needed as these age and gender effects might be explained by the frequency with which the three objects had been drawn, rather than solely by limited drawing skills. Furthermore, the results are highly dependent on the SSSM’s quality and may contain a bias, even if this SSSM can be considered state-of-the-art.

Conclusion: Interpreting children's HPT drawings with an SSSM seems feasible. Consequently, this approach could potentially supplement conventional HPT assessment methods for identifying developmental disorders in children. However, further research is needed to strengthen these results.

The authors declare that they have no competing interests.

The authors declare that an ethics committee vote is not required.

Literatur

[1] Sajatovic M, Ramirez LF. Rating scales in mental health. JHU Press; 2012.
[2] Aldridge M, Wood J. Talking about feelings: Young children"s ability to express emotions. Child Abuse & Neglect. 1997 Jan 1;21(12):1221-33. DOI: 10.1016/S0145-2134(97)00097-5
[3] Unger S, Robens S, Anderle L, Ostermann T. Digital Drawing Tools for Assessing Mental Health Conditions–A Scoping Review. In: German Medical Data Sciences 2024. 2024. p. 251-9. DOI: 10.3233/SHTI240864
[4] Bourouis A, Fan JE, Gryaditskaya Y. Open vocabulary semantic scene sketch understanding. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 2024. p. 4176-86.
[5] Jolley RP, Fenn K, Jones L. The development of children’s expressive drawing. British Journal of Developmental Psychology. 2004 Nov;22(4):545-67. DOI: 10.1348/0261510042378236