A Novel Framework Using Graph-Based Clustering for Dental X-Ray Image Search in Medical Diagnosis

Abstract

Abstract-Dental X-Ray image search is an important process in medical diagnosis for diagnosing exactly dental diseases of a patient. This problem is regarded as the matching of a dental X-Ray image with diseases patterns in the database. In this paper, we propose a novel framework using graph-based clustering for dental X-Ray image search. This framework firstly extracts dental features from an X-Ray image to a dental feature database and then uses a vector quantization algorithm to clarify the principal records from the database. Each record is now regarded as a node in a graph which is classified by a graph-based clustering algorithm according to the disease patterns. The dental X-Ray image is classified having disease or non-disease according to other disease patterns in the same group. The new method is experimental validated on a real dataset of 13 dental X-ray images taken from Hanoi Medical University, Vietnam at the period of 2014-2015. Three variants of the framework namely Prim spanning tree (GCP), Kruskal spanning tree (GCK), and Affinity Propagation Clustering (APC) has been implemented. The experimental results suggest the best variant in term of accuracy.Index Terms-Dental X-ray images, graph-based clustering, medical diagnosis, affinity propagation clustering. I. INTRODUCTIONA computerized medical diagnosis system is of great interest to clinicians for accurate decision making of possible diseases and treatments. Because of knowledge re-use and the capability to learn by examples, it has an important role to routine works of clinicians. In recent years, there have been many researches that aim to develop such the system. In 2012, Subgahata Chattopadhyay et al. [1] presented an application of Bayesian network to diagnose toothache. Kavitha et al. [2] used Support Vector Machine (SVM) to predict the osteoporosis from dental images. In 2014, Kantesh and Xu [3] proposed a fuzzy-based method to predict heart risk. Sutton [4] used Fuzzy Neighbor K-Nearest Neighbour (FKNN) method in different dentistry problems.In dentistry, dental X-Ray image search is the core process of medical diagnosis for diagnosing exactly dental diseases of a patient. This problem is regarded as the matching of a dental X-Ray image with diseases patterns in the database. If it is similar to a disease pattern then its status is 'disease'; otherwise is 'non-disease'. There have been some achievements in form of image searching based on graph modeling [5]- [7]. Based on these studies, we model the dental X-Ray image search in a graph and then use a graph-based clustering method to determine the cluster containing the new X-Ray image. From that, diagnosing information is determined via membership degrees of the cluster.In this paper, we propose a novel framework using graph-based clustering for dental X-Ray image search. The novel parts of this framework in comparison with the relevant researches are: 1) generate a dental feature database of dental X-ray images; 2) use a vector quantization algorithm to clarify the principal reco...