Assessing AI‐Enhanced Learning in Bone Loss Detection among Dental Students

Abstract

A key component of dental education is to equip students with clinical decision-making skills. With the increasing integration of artificial intelligence (AI) supported clinical decision support (CDS) tools, dental education has focused on preparing students to become competent clinicians who can use these technologies responsibly and ethically [1]. One key AI CDS tool is AI-enhanced radiographic imaging, which provides visual guidance for pathologic dental findings, such as caries, periodontal bone loss, and periapical radiolucency [2]. AI-enhanced radiographic imaging employs deep convolutional neural network models to detect periodontal bone loss (PBL). These models achieve PBL sensitivity, specificity, and accuracy similar to human oral maxillofacial radiologists, endodontists, and periodontists [3]. This suggests that these models possess diagnostic accuracy similar to the dental educators who teach radiographic interpretation to dental students. There is a paucity of literature and studies on new technology and their impact on dental education. In this pilot study, our aim is to determine if training with AI-enhanced radiographic imaging can help students learn to identify periodontal bone loss with an accuracy similar to training with the conventional radiographic imaging software. The Periodontist faculty at Western University of Health Sciences, College of Dental Medicine, (Pomona, CA) conducted a study involving first-year and fourth-year predoctoral dental students in October 2024. A total of 25 dental students volunteered to participate in the study (13 first-year, 12 fourth year). First-year dental (D1) students were recruited in the study to assess the impact of using a conventional digital imaging program (MiPACS, LEAD Technologies Inc, Charlotte, NC) versus an AI-enhanced CDS (Overjet AI) as a learning tool without prior knowledge of basic periodontal diagnosis and bone level assessment (Figure 1A). Fourth-year dental (D4) students were recruited to assess whether prior clinical and radiographic interpretation experience and a developed understanding of periodontal diagnosis would lead to different performance outcomes compared to first-year students. Block randomization was employed to assign students to either the MiPACS group or the Overjet group. All participants completed a pretest that consisted of 30 two-dimensional (2D) intraoral radiographs. This was followed by a 10-min lecture covering periodontal diagnosis and radiographic interpretation. Participants then engaged in 15 min of independent practice using their assigned imaging tool before completing a posttest (Figure 1B). Students in the MiPACS group were instructed to manually measure the distance from the alveolar crest (AC) to the cementoenamel junction (CEJ) using the ruler tool. In contrast, students in the Overjet group visualizes this distance automatically. Both first year and fourth year groups showed significant accuracy improvement from pretest to posttest (Figure 2A, p < 0.05, two-tailed test; 13 D1 students: 0.72 ± 0.14 to 0.92 ± 0.05; 12 D4 students: 0.83 ± 0.08 to 0.89 ± 0.04). There was no statistically significant difference in learning outcomes between two groups (Figure 2B, p = 0.67, two-tailed t-test; Gains: D1-MiPACS: 0.21 ± 0.16; D1-AI:0.17 ± 0.11; D4-MiPACS: 0.07 ± 0.10; D4-AI: 0.06 ± 0.06), although the MiPACS group trended toward slightly higher improvement. These findings suggest that continued training in radiographic interpretation improves diagnostic accuracy regardless of the tool used or clinical experience. Overjet aids students in learning radiographic interpretation and diagnosis similar to conventional digital image tools. A larger sample size is needed in a future study to determine whether the observed trend might reach statistical significance. This would allow further exploration of whether the higher level of interaction and active learning by conventional imaging tool, as indicated by student feedback, has a measurable impact on learning outcomes. The protocol for this study was reviewed by the Western University of Health Sciences Institutional Review Board and approved (X24IRB083). The authors declare no conflicts of interest.

Ähnliche Arbeiten