Hybrid CNN Classifier for Accurate Detection of Oral Cysts in Panoramic Dental Radiograph

In dentistry, machine learning approaches play a significant role in the early detection and diagnosis of oral cysts. When planning accurate medical treatment, oral and maxillofacial surgeons can benefit from preoperative detection of these cysts. Surgical planning is directly affected by classifying oral cysts as dentigerous, radicular, ameloblastoma, odontogenic carcinoma, and odontogenic keratocyte. In dentistry, the outcomes of differential diagnosis based solely on imaging are unacceptable. This research proposes a three-step framework for autonomous cyst classification, segmentation, and detection based on a Convolutional Neural Network (CNN). To begin, DENT-CNN, a modified CNN architecture is proposed to classify the oral cyst as normal or abnormal. Second, UNet is a CNN variant used to segment the categorized cyst regions on panoramic dental images. Third, the Gray Level Co-occurrence Matrix (GLCM) is combined with CNN to identify segmented cysts as benign or malignant. The proposed work is extensively evaluated using 1500 real-time dental panoramic radiographic images of the oral cyst collected from a dental hospital. To alleviate overfitting, the data augmentation approach is used. The performance of the proposed work is analyzed in terms of a cyst classification rate of 96.8 %, a Dice Score of 96.3 % for cyst segmentation, and an accuracy for the Identification phase of 96.5%.