| 摘要: | 背景與目的：在牙科診斷中，使用根尖X光影像來評估植體系統及其邊緣骨組織的喪失情況極為關鍵。隨著人工智慧技術在放射影像分析領域展現出越來越多的應用價值，本研究旨在開發一種深度學習模型，使用根尖X光影像來辨識和分類牙植體系統，並測量植體周圍邊緣骨骼流失的程度。
材料與方法：本研究使用了800張根尖X光影像組成的數據集，將其分為訓練、驗證和測試集，並應用了最新的物件偵測深度學習演算法(YOLO v7)進行植體系統及其邊緣骨喪失程度的辨識。模型的性能通過平均精確度 (mAP50; Mean Average Precision)、精確度、召回度和F1分數等關鍵指標進行評估。
結果與討論：研究結果顯示，該模型在整體性能評估中精確度為0.94、召回率為0.92、F1分數為0.93和mAP為0.94。在植體邊緣骨喪失程度的辨識方面，精確度為0.90、召回率為0.84、F1分數為0.87、mAP為0.86；在非植體邊緣骨喪失程度的辨識方面，精確度為0.97、召回率為0.95、F1分數為0.96、mAP為0.97；而在植體系統辨識方面，精確度在0.90~0.99之間、召回率在0.86~0.98之間、F1分數在0.92~0.96之間以及mAP在0.95~0.98之間。本研究結果證明，YOLO v7模型能夠準確地從根尖X光影像中辨識植體及其邊緣骨組織的喪失，早期預測植體周圍骨質流失可以幫助牙醫及早辨識和治療植體周圍炎，此技術為牙科專業人士及及面臨植體問題的患者提供了有力的支持和應用。
Background and Purpose: In dental diagnostics, the use of periapical X-ray images to assess the loss of implant systems and their surrounding bone tissue is crucial. With the increasing application value of artificial intelligence technology in the field of radiographic image analysis, this study aims to develop a deep learning model that uses periapical radiographs to identify and classify dental implant systems, as well as measure the extent of marginal bone loss around the implants.
Materials and Methods: This study utilized a dataset composed of 800 periapical X-ray images, which were divided into training, validation, and testing sets. The latest object detection deep learning algorithm (You Only Look Once vision 7; YOLO v7) was applied to identify implants and the degree of surrounding bone loss. The model's performance was evaluated using key indicators such as a mean Average Precision (mAP), precision, recall, and F1 score.
Results and Discussion: The results showed that the model achieved an overall performance precision of 0.94, recall of 0.92, F1 score of 0.93, and an mAP of 0.94. In identifying the degree of marginal bone loss around implants, the model demonstrated a precision of 0.90, a recall of 0.84, an F1 score of 0.87, and mAP of 0.86. For non-implant marginal bone loss, t the model exhibited a precision of 0.97, a recall of 0.95, an F1 score of 0.96, and mAP of 0.97. In implant system identification, the model's precision ranged from 0.90 to 0.99, recall ranged from 0.86 to 0.98, F1 score ranged from 0.92 to 0.96, and mAP ranged from 0.95 to 0.98. The YOLO v7 model accurately identifies implants and their surrounding bone tissue loss from periapical radiographs. Early prediction of peri-implant bone loss helps dentists to promptly identify and treat peri-implantitis. This technology provides valuable support and practical applications for dental professionals and patients facing implant-related issues. |
