Retrospective evaluation of the clinical utility of reconstructed computed tomography images using artificial intelligence in the oral and maxillofacial region

Abstract

The aim of this study was to convert medical images stored in 3 mm slices in the picture archiving and communication system (PACS) to 1 mm slices, using artificial intelligence (AI), and to analyze the accuracy of the AI. The original 1.0 mm CT slices of the facial bone were obtained from 30 patients and reformatted to a rough CT slice of 3.0 mm. CT slices of 1.0 mm were subsequently reconstructed from those of 3.0 mm using AI. The AI and rough CT images were superimposed on the original CT images. Fourteen hard-tissue and five soft-tissue landmarks were selected for measuring the discrepancy. The overall average differences in values for the hard-tissue landmarks were 1.31 ± 0.38 mm and 0.81 ± 0.17 mm for the rough and AI CT images, respectively. The values for the soft-tissue landmarks were 1.18 ± 0.35 mm and 0.54 ± 0.17 mm for the rough and AI CT images, respectively. The differences for all the landmarks, excluding point A and pogonion, were statistically significant. Within the limitations of the study it seems that CT images reconstructed using AI might provide more accurate clinical information with a discrepancy of less than 1.0 mm.