Deep U-Net Architecture for Semantic Segmentation of Dental Carries

Abstract

Dental caries is one of the serious health problems across the world which leads to severe pain, dental pulp, or even abscess. The major reason for exponential increase in dental caries is the increased consumption of sugar containing beverages like juices and sodas. These sugars over the time undergoes various bacterial metabolisms to form acids which leads to loss of tooth enamel and dentine. To reduce the pain due to dental caries time to time diagnosis and effective treatment is obvious. Naked-eye detection and radiographic analysis are the conventional diagnosis techniques. These techniques would require a highly qualified doctors and it is a time consuming and may sometimes leads to misdiagnosis. We employ semantic segmentation with Dental Caries U-Net (DCUN) architecture for identifying the dental caries in an efficient and timely manner. We have collected a panoramic radiographic image containing dataset which represented different caries types like shallow, moderate, and deep caries. These dental panoramic images are fed into the u-net architecture which would contain series of convolution layers, max pooling, up sampling, and down sampling. We have compared our results with state-of the-art techniques and demonstrated the superiority of our method.