การจำแนกภาพเอกซเรย์ในช่องปากเพื่อแยกโรคอะมีโลบลาสโตมา โรคโอดอนโตเจนิก เคอราโตซีสต์และโพรงปากที่ไม่มีรอยโรคด้วยวิธีการเรียนรู้ของเครื่อง

Abstract

Ameloblastoma (AME) and Odontogenic keratocyst (OKC) are common jaw lesions in oral surgery, often presenting similar features in panoramic X-ray images. However, their treatment strategies differ significantly, so emphasizing the critical importance of accurate diagnosis. This research developed machine learning models for the classification of these lesions utilizing panoramic x-ray images, comprising 100 normal and 124 diseased cases, categorized into AME (73) and OKC (51). Initially, a model was trained to classified Normal dental and diseased cases, followed by a model to distinguish between AME and OKC. In the preprocessing phase, images augmentation techniques and feature extraction methods were used, aimed at enhancing model performance. Moreover, various model architectures were explored to identify the optimal configuration. The achieved accuracy for classifying AME and OKC images demonstrated an overall accuracy of 74.1% and a weighted F1 score of 0.75. The highest performing model indicated that focusing on the lesion area, in feature extraction, markedly improved accuracy compared to others. Moreover, the model's performance rivaled or surpassed manual classification by dentist in distinguishing between AME and OKC.