A content-based image retrieval system for echo images using SQL-based clustering approach

Abstract

Content-based image retrieval (CBIR) consists of retrieving the most visually similar images to a given query image from a database of images. CBIR from medical image databases does not aim to replace the physician by predicting the disease of a particular case but to assist him/her in diagnosis. The visual characteristics of a disease carry diagnostic information and oftentimes visually similar images correspond to the same disease category. By consulting the output of a CBIR system, the physician can gain more confidence in his/her decision or even consider other possibilities. In this paper, we aim at building an efficient content-based echo image retrieval (CBEIR) system. Echocardiography provides important morphological and functional details of the heart which can be used for the diagnosis of various cardiac diseases. Normally two-dimensional (2D) echo and colour Doppler image modalities are used for analysis and clinical decisions. From 2D echo images, features such as dimensions of cardiac chambers (area, volume, ejection fraction, etc.) are extracted, whereas texture properties, kurtosis, skewness, edge gradient, colour histogram, etc., are extracted from colour Doppler images. Hence, this forms a multi-feature descriptor which then is used to retrieve similar images from the database. A novel clustering approach merged with the traditional CBIR model is used for development in order to speed up the retrieval and enhance the accuracy of retrieval. The main focus of our work is the following: efficient segmentation algorithm, accurate detection of cardiac chambers, new and fast method to obtain colour portion of the Doppler image, and finally is able to categorise the type of disease and the severity level. These domain-specific low-level features are very important to build a reliable and scalable CBIR model. The similarity values are obtained by Euclidean distance metric. The feature database is basically a set of quantitative and qualitative features of the images. Our image database is populated with diverse set of approximately 623 images extracted from 60 normal and abnormal patients acquired from a local cardiology Hospital. Exhaustive experimentation has been conducted with various input query images and combinations of features to compute the retrieval efficiency which are validated by domain experts. It has been shown through recall-precision graphs that the proposed method outperforms compared to others reported in the past.