A Fused Deep Learning Architecture for the Detection of the Relationship between the Mandibular Third Molar and the Mandibular Canal

Abstract

The study aimed to generate a fused deep learning algorithm that detects and classifies the relationship between the mandibular third molar and mandibular canal on orthopantomographs. Radiographs (n = 1880) were randomly selected from the hospital archive. Two dentomaxillofacial radiologists annotated the data via MATLAB and classified them into four groups according to the overlap of the root of the mandibular third molar and mandibular canal. Each radiograph was segmented using a U-Net-like architecture. The segmented images were classified by AlexNet. Accuracy, the weighted intersection over union score, the dice coefficient, specificity, sensitivity, and area under curve metrics were used to quantify the performance of the models. Also, three dental practitioners were asked to classify the same test data, their success rate was assessed using the Intraclass Correlation Coefficient. The segmentation network achieved a global accuracy of 0.99 and a weighted intersection over union score of 0.98, average dice score overall images was 0.91. The classification network achieved an accuracy of 0.80, per class sensitivity of 0.74, 0.83, 0.86, 0.67, per class specificity of 0.92, 0.95, 0.88, 0.96 and AUC score of 0.85. The most successful dental practitioner achieved a success rate of 0.79. The fused segmentation and classification networks produced encouraging results. The final model achieved almost the same classification performance as dental practitioners. Better diagnostic accuracy of the combined artificial intelligence tools may help to improve the prediction of the risk factors, especially for recognizing such anatomical variations.

The diagnostic performance of impacted third molars in the mandible: A review of deep learning on panoramic radiographs

“… Yoo et al (2021) determined the sensitivity and specificity using positive predictive value (PPV) calculations. Furthermore, the studies reviewed in this work did not use a consistent mathematical formula ( Ariji et al, 2022 , Buyuk et al, 2022 , Celik, 2022 , Choi et al, 2022 , Maruta et al, 2023 , Takebe et al, 2022 , Vinayahalingam et al, 2019 , Yoo et al, 2021 , Zhu et al, 2021 ). Celik, 2022 , Choi et al, 2022 , Takebe et al, 2022 , and Zhu et al (2021) , used PPV calculations to determine precision in their research.…”

“…Meanwhile, Celik (2022) used the true positive formula divided by all the data as a recall. Buyuk et al, 2022 , Vinayahalingam et al, 2021a provided calculations in accordance with the guidelines for medical diagnostic tests in equations (1), (2), (3), (4), and (5) in Table 2 ( Leeflang, 2014 ).…”

“…The NPV denotes the ratio of actual negative data to the data that deep learning algorithms interpret as negative. Accuracy is used to calculate the overall deep learning performance ( Buyuk et al, 2022 , Vinayahalingam et al, 2021a ). Referring to Table 2 , the calculation method is the same from a computational and medical perspective but only for the calculation of accuracy.…”

“…Technology advancements have enabled the integration of AI into CAD systems, which can be used to assess the relationship between the third molars on either side (bucco-lingual) of the mandible and the mandibular canal. GoogLeNet, VGG-16, AlexNet ( Buyuk et al, 2022 , Fukuda et al, 2020 ), YOLOv3 ( Takebe et al, 2022 ), YOLOv4 ( Zhu et al, 2021 ), ResNet-50 ( Choi et al, 2022 , Sukegawa et al, 2022b ) are some deep learning architectures that have been investigated for radiographic evaluation of impacted third molar related to the mandibular canal. In addition, U-Net is also used for image segmentation of the mandibular canal ( Ariji et al, 2022 , Buyuk et al, 2022 , Vinayahalingam et al, 2019 ).…”

“…The position of impacted mandibular third molars that are buccal or lingual to the mandibular canal but not in contact with it can be revealed by the superimposed radiographic appearance ( Choi et al, 2022 ). Buyuk et al (2022) addressed the needs of dentists because deep learning can assess the association between ITM and MC using standards that are nearly identical to those used in dentistry by classifying panoramic radiographs of ITM into four categories ( Table 3 ). The fused deep learning method was utilized, segmentation was performed using the U-Net architecture, and the AlexNet architecture was used to assess the relationship between the ITM and MC ( Buyuk et al, 2022 ).…”