Comparison of fractal analysis values obtained from panoramic radiographs with chatgpt-based measurements

INTRODUCTION: The aim of this study was to evaluate the extent to which fractal dimension values obtained from panoramic radiographs correspond to those obtained manually by an oral and maxillofacial radiologist and to Generative Pre-trained Transformer (GPT) -based fractal analysis (FA) values, and to quantitatively examine the agreement between the two methods.
METHODS: Panoramic radiographs of 50 patients were examined in the study. Three regions of interest (ROIs) of 15x15 pixels were selected from each image. The oral and maxillofacial radiologist measurements were calculated in ImageJ 1.49x, and artificial intelligence measurements were calculated in the ChatGPT-4o environment. Intraclass correlation coefficient (ICC) was used for agreement, Pearson correlation was used for correlation, and Bland–Altman analysis was used for method differences.
RESULTS: The ICC, calculated using a two-way mixed model and absolute agreement definition, was found to be negative for both single measurements (ICC(A,1) = –0.06; p = 0.739) and average measurements (ICC(A,2) = –0.134; p = 0.708). The relationship between the measurements was assessed using Pearson correlation analysis. No significant linear relationship was found between the two methods (r = –0.099; p = 0.492).
DISCUSSION AND CONCLUSION: In its current form, ChatGPT-4o regarding FA does not provide acceptable agreement with radiologist measurements. It is anticipated that artificial intelligence (AI) assisted FA could become a helpful tool in the future with the stabilization of the algorithms/parameters, expansion of the training data, and software validation studies.

Keywords: artificial intelligence, fractal analysis, panoramic radiography