Computer vision application in diagnosis of the need for apicoectomy surgery

Background and aim: Diagnosing the need for apicoectomy surgery with manual methods is oftencostly and time-consuming. When the infection progresses and reaches the end of the tooth rootand root-end opening, endodontists are forced to rift the gum and remove the infection from theend of the tooth root. This procedure is called apicoectomy surgery. If the infection progresses areignored there will be consequences which include possible tooth loss, jaw, brain and other organsinfection. Therefore, early diagnosis of the need for this surgery is essential. This article aimsto present and report the performance of a machine learning model that diagnoses the need forapicoectomy surgery in panoramic images by using computer vision techniques and deep neuralnetworks.Method: This model is a deep neural network that learns to fulfill several related tasks simultaneouslywith the help of self-supervised learning techniques, this leads to extraction of meaningfulfeatures. Therefore, the proposed model can diagnose the need for surgery with high accuracy.The data used to train the model consists of orthopantomogram (OPG) images taken from patients’jaws and collected from clinics and labeled by skilled endodontists. The dataset contains۷۹۹ samples in total with positive or negative labels. The samples have widths and heights of ۷۰۰pixels. The data were fed to the model after augmentation, segmenting regions of interest, andpreprocessing. In this study we used F-۱ score as an evaluation metric of methods. Moreover,K-fold cross-validation technique was used as an assessing method.Results: The proposed model obtained a mean F-۱ score of ۸۶% in cross-validation. This resultsuggests that the model has reached desirable accuracy on new samples.Conclusion: Misdiagnosis can endanger the patient’s health in addition to incurring extra costsand irreversible damages. The study results demonstrate that the deep learning system showshigh accuracy in the diagnosis of the need for apicoectomy surgery. Additionally, unlike the previousmethods, which endodontists required more information regarding the patient besides theradiographic image to make their decision, this research shows that proposed model has made itpossible to handle this task with only one image of the patient’s jaw. In the light of these improvements,this model can operate as a doctor’s assistant in clinics, leading to increased accuracy andfaster diagnosis of the need for this surgery. Among the works to improve the proposed method,we mention training the model on a larger volume of data and assigning more parallel tasks toextract more meaningful features.

کلیدواژه ها:

Medical image processing ، Computer vision ، Dentistry ، Deep learning ، Self-supervised learning ، Apicoectomy