- 學位論文
基於深度學習的二階段口腔全景圖像分類方法
A Two-Phase Dental Panoramic Radiograph Classification Method based on Deep Learning
摘要
隨著資訊科技的進步，越來越多的資料科學家開始利用機器學習演算法對電腦影像進行更多不同類型的資料處理與分析，例如近年來深度學習(Deep Learning)方法以多層次類神經網路的方式進行處理，找出潛在於大量資料中的特徵。在過去臨床醫療過程中，口腔全景圖(Dental Panoramic Radiograph, DPR)常被醫療人員用來作為了解病患牙齒狀況的重要參考診斷依據，藉此協助牙科醫師提供病人最即時的醫療服務。但在現行環境下，對於口腔全景圖的標記與識別，都必須仰賴專業的醫療人員，因此間接加重了工作負荷。在進行大量口腔全景圖片資料內容的專業判定流程中，若能透過深度學習的圖片辨識方法，自動分析病患之口腔全景圖，將有助於加快標記與識別口腔全景圖內容的專業知識，可大量節省寶貴的醫師人力成本與時間，提高牙科服務的醫療品質。本論文提出一個「基於深度學習的二階段口腔全景圖像分類方法」，此方法是一個基於卷積神經網路(Convolutional Neural Networks, CNN)模型的牙齒定位辨識系統與牙齒病灶類型分類系統，使用深度學習方法中的AlexNet模型，由五個卷基層和三個完全連接層所組成。系統分為兩個階段，第一階段進行口腔全景圖之牙齒定位與診斷辨識的工作，第二階段進一步為不同的牙齒病灶類型進行分類，以提供資訊協助牙科醫師作為醫療診斷決策支援的工具，並為病患進行後續相關的牙齒治療。經過適當的資料前處理步驟，第一階段模型準確率達到89.36%，第二階段模型準確率達到90.01%。
並列摘要
With the advanced of information technology, many data scientists began to apply machine learning algorithms to solve problems in computer vision. For example, in recent years, the deep learning method which works in multi-layer neural networks is a popular approach to find the potential features in the numerous data. The Dental Panoramic Radiograph (DPR) has been widely used by medical professionals to understand the detailed dental information of patient's tooth condition. However, the labeling and identification tasks of DPR must be relied on professional medical staffs, thus the tasks aggravate the workload of medical personnel. If the numerous of patients’ DPR image could be analyzed and recognized through deep learning method automatically, it will be helpful to accelerate the efficiency of labeling and identification tasks. In this thesis, we will propose a novel method which named "A Two-Phase Dental Panoramic Radiograph Classification Method based on Deep Learning". The goal of this thesis is to design and construct a DPR classification system based on deep learning (AlexNet) and Convolutional Neural Networks (CNN) model for tooth position localization and condition recognition. Moreover, this approach could classify the dental diagnosis condition, thus can provide useful information to assist the professional dentists and medical staffs. This medical diagnostic decision support system can be used by dentists to do the follow-up dental treatments for patients. The experiment was divided into two phase: teeth positioning and condition classification. The accuracy of the first stage is 89.36%, and second stage is 90.01%.