Enhanced level set segmentation method for dental caries detection

Abstract

Caries detection system is important for dental disease diagnosis and treatment. It can be identified using X-ray imaging. The X-ray image contains interest point of dental to get the teeth information according to specific diagnostic intention. The Region of Interest (ROI) includes the caries area on tooth surface. The imaging challenges like noise, intensity inhomogeneities and low contrast causes the difficulty for identifying correctly the ROI in dental images. According to the recent studies, among all medical image segmentation methods, level set has the best segmentation accuracy. However, there are several components in the level set that need to be enhanced to determine the exact boundary to separate the ROI. The signed force function to control the direction of level set evaluation process, speed function to control the speed of movement and Initial Contour (IC) generation to obtain a more accurate ROI require an enhancement for the better accuracy. In this research, a new enhancement of segmentation method has been proposed based on finding an accurate outcome. The method includes two phases: IC generation and intelligent level set segmentation. In addition, caries detection process is performed with new detection method. To generate the IC for dental X- ray images, a new local IC selection for level set method is proposed. Statistical and morphological information of image is extracted to establish a technique that is able to find a suitable IC. In the second phase, statistical information of the pixels inside and outside the generated contour and linear motion filtering is used to construct the region-based signed force function to provide more stabilisation to proposed method. Furthermore, 31 features of image are extracted to train the neural network and to generate proper speed function parameter. The results of proposed method provide the high accuracy and efficiency in the process of getting teeth boarder. The next process is to detect from the segmented images. The research also proposed a new method using integral projection and feature map for every single tooth to obtain the information of caries area. The achieved overall performance of proposed segmentation method is evaluated at 120 periapical dental radiograph (Xray), with 90% accuracy rate. In addition, the caries detection accuracy rate on 155 segmented images is 98%.