Interpretable Drawing Psychoanalysis via House-Tree-Person Test

As the number of people with psychological disorders continues to increase in recent years, it is particularly important to identify patients at an early stage. As one of the widely recognized methods of drawing psychoanalysis methods, the House-Tree-Person (HTP) test is commonly used for psychological assessment. Among the standards for HTP drawing analysis, the analysis of holistic features such as size and position is the most important evaluation standard. Traditional manual analysis for test results is time-consuming and highly subjective. Although image classification models provide fast and accurate predictions, they lack reliable decision-making interpretation. Existing methods for interpreting classification models can offer a certain degree of interpretation, but they are unable to provide specific interpretation related to psychology. In this paper, we propose a method to quantitatively analyze the interpretable results of decisions made by image classification models in relation to expert knowledge. Specifically, we initially utilize the interpretable method to identify the important features that influence the model's decision-making process. Then, we use self-annotated files containing important symbols relevant to psychology to segment important features, obtaining psychological features associated with the model's decision-making process. Finally, we quantitatively analyze the size and position through the psychological features respectively, comparing them with the widely used expert knowledge and rules. The experimental results reveal that the decisions of image classification model on the HTP dataset are consistent with the relevant universal rules of psychological features. The proposed method improves the reliability of image classification algorithm in psychotherapy and psychodiagnosis.