| Abstract: | 在人工智慧高速的發展下，機器不僅能透過大量資料的訓練來學習人類的行 為與判斷能力，還能自動進行資料的歸納、分類與決策，以此完成複雜的任務。 目前已有 AI 技術被應用於臨床醫學進行疾病的診斷或手術前後的規劃及追蹤， 然而大部分的牙科檢查仍維持傳統耗時的檢查模式，鮮少將 AI 技術應用於牙科 臨床診斷領域，如牙位辨識、牙周病或齲齒的診斷，其中牙位辨識必須先利用 FDI 牙位表示法進行編號，再依據每個人的牙齒生長狀況進行紀錄並輸出成文字 檔，造成診斷過程過於耗時的問題。此外，牙周病檢測必須讓患者接受侵入性的 牙周囊袋探測，其過程將會引起患者疼痛或痠痛，導致多數患者拒絕進行此檢測， 無法準確診斷出患者是否患有牙周病。因此本論文利用 Mask R-CNN 深度學習模 型與本論文提出的 PD 檢測演算法開發出牙位標記與牙周病輔助診斷系統，不僅 能透過Mask R-CNN自動標記全口牙齒X光影像中的牙位及牙周病判斷三區域， 還能透過 PD 檢測演算法計算出罹患牙周病的機率，以判斷患者是否罹患牙周病 及其嚴重程度。由實驗結果可知Mask R-CNN進行牙位的辨識的準確率、精準度、 敏感度與特異度分別為 92%、92.3%、92.3%以及 91.7%，在標記牙周病判斷三區 域的準確率、精準度、敏感度與特異度分別為 88%、93.3%、87.5%以及 88.9%， 而利用自行建置的 PD 演算法判斷患者患有牙周病的區域及機率之準確率、精準 度、敏感度與特異度分別為 84%、85.7%、85.7%以及 81.8%，證明此系統除了能 有效輔助醫師診斷，並提高診斷效率及精準度外，也能提升患者針對醫師診斷的 可信度。期望未來能將此研究的應用領域擴大到齲齒診斷、牙結石檢測或口腔癌 前兆檢測，以此提供患者一系列治療流程。
With the rapid development of artificial intelligence, machine can not only learn the behavior and judgement capability of human through the training of inputting the large amount of data but also automatically perform data induction and data decision-making to complete complex tasks. At present, AI technology has been used in clinical medicine for disease diagnosis or preoperative and postoperative planning and tracking. However, most dental inspection still conducts with a traditional mode that is time-consuming, and AI technology is rarely used in the field of dental clinical diagnosis, such as dentition identification, diagnosis of periodontal disease or caries. Among them, dentition identification must be numbered with FDI World Dental Federation notation first, and then record as a text file according to each person’s status of growth of tooth, which cause the problem that the diagnosis process is too time-consuming. In addition, the periodontal disease detection must have the patient to undergo invasive periodontal pocket detection. The process of periodontal pocket detection will cause pain or soreness in patient, causing most patient refuse the detection, and cannot accurately diagnose whether the patient has periodontal disease. Therefore, we use Mask R-CNN deep learning algorithm and the purposed PD detection algorithm in this paper to develop a dentition labeling and periodontal disease auxiliary diagnosis system. Our proposed system can not only use Mask R-CNN to automatically label the dentition and the three areas for periodontal disease judgment in the panoramic x-ray image, but also use PD detection algorithm to calculate the probability of suffering from periodontal disease to identify whether the patient has periodontal disease and its severity. The experimental results show that the accuracy, precision, sensitivit y, and specificity of Mask R-CNN for dentition identification are 92%, 92.3%, 92.3%, and 91.7%, respectively, and the accuracy precision, sensitivity, and specificity of Mask R-CNN for labeling periodontal disease judgment area are 88%, 93.3%, 87.5%, and 88.9%, respectively. And the accuracy, precision, sensitivity, and specificity of proposed PD algorithm which is used to identify periodontal disease judgment area and compute the probability of periodontal disease are 84%, 85.7%, 85.7%, and 81.8%, respectively, which proves that our system can not only effectively assist dentist in diagnosis, improve the diagnostic efficiency and accuracy of diagnosis, but also improve the credibility of dentists in the diagnosis. It is hoped that the application of this research can be expended to caries diagnosis, dental calculus detection, and precursor of oral cancer detection to provide patient with a series of treatment process. |
