THE UTILITY OF DEEP LEARNING MODEL IN CLINICAL TREATMENT DECISION-MAKING OF MANDIBULAR THIRD MOLAR: A SYSTEMATIC REVIEW AND META-ANALYSIS

Abstract

Objectives

To evaluate the effectiveness of deep learning (DL) in medical imaging for improving clinical procedures related to the mandibular third molar (MM3).

Methods

Our systematic search included databases such as PubMed, EMBASE, Web of Science, and the Cochrane Library, up to April 10, 2024. The Quality Assessment of Diagnostic Accuracy Studies-2 (QUADAS-2) tool was used to assess the potential bias and clinical relevance of the studies. To examine the interaction between MM3 and the inferior alveolar nerve (IAN), we conducted a meta-analysis using a bivariate mixed-effects model to compare the performance of DL methods with that of human experts.

Results

In this comprehensive review, a total of 33 scholarly works involving 45,029 imaging instances were analyzed. Among these, 15 studies specifically focused on evaluating the performance of DL methods in identifying the spatial relationship between MM3 and IAN. The aggregated results showed that DL achieved a diagnostic sensitivity of 0.89 (95% CI, 0.85-0.93) and a specificity of 0.88 (95% CI, 0.80-0.93). In parallel, a comparative meta-analysis that included assessments from 16 clinical experts revealed slightly lower sensitivity (0.76; 95% CI, 0.65-0.85) and specificity (0.86; 95% CI, 0.70-0.94). Additionally, the review explored the utility of DL in predictive modeling of MM3 eruption, assessing the complexity of extraction procedures, and identifying MM3 characteristics. The 18 studies in this area demonstrated diagnostic sensitivity ranging from 0.80 to 0.94 and specificity from 0.75 to 0.97, collectively highlighting the strong prognostic capabilities of DL in these clinical applications.

Conclusions

DL has demonstrated remarkable efficacy, especially in precisely identifying the anatomical connections between MM3 and IAN. Moreover, DL excels in identifying, categorizing, and assessing the complexity associated with MM3 extractions. This study confirms the substantial value of DL, particularly when utilized through image-based methods, in improving clinical management strategies for MM3. The creation of intelligent diagnostic tools using oral panoramic radiographs (OPGs) is strongly encouraged to assist physicians, especially those with limited clinical experience, in enhancing their diagnostic accuracy. These advancements in DL applications are expected to significantly improve the accuracy and efficiency of clinical diagnoses in oral and maxillofacial surgeries.