Introduction: Image retrieval is the most interesting technique which is being used today in our digital world. CBIR, commonly expanded as Content Based Image Retrieval is an image processing technique which identifies the relevant images and retrieves them based on the patterns that are extracted from the digital images. In this paper, two research works have been presented using CBIR. The first work provides an automated and interactive approach to the analysis of CBIR techniques. CBIR works on the principle of supervised machine learning which involves feature selection followed by training and testing phase applied on a classifier in order to perform prediction. By using feature extraction, the image transforms such as Contourlet, Ridgelet and Shearlet could be utilized to retrieve the texture features from the images. The features extracted are used to train and build a classifier using the classification algorithms such as Naïve Bayes, K-Nearest Neighbour and Multi-class Support Vector Machine. Further the testing phase involves prediction which predicts the new input image using the trained classifier and label them from one of the four classes namely 1- Normal brain, 2- Benign tumour, 3- Malignant tumour and 4- Severe tumour. The second research work includes developing a tool which is used for tumour stage identification using the best feature extraction and classifier identified from the first work. Finally, the tool will be used to predict tumour stage and provide suggestions based on the stage of tumour identified by the system. This paper presents these two approaches which is a contribution to the medical field for giving better retrieval performance and for tumour stages identification.
Methods: The CBIR system proposed for brain tumour image retrieval and stages identification retrieves similar brain tumour images of same stages. It also helps in identifying the stage of the tumour image. This paper produces a comparative analysis based on the architecture of our work given in Fig. 1. As this work is related to medical image processing, there may be noise removal processes which are to be included in the pre-processing step because of the demand in quality required when used in a medical imaging system. The noises which may be present in the images are to be considered and necessary measures are to be taken for removing the noise in the brain tumour images. In this work, such type of noises is removed by using Median filter applied on the dataset which will remove some of the noises that are present in the image before using for classification.
Our image retrieval system consists of three main components. When considering medical images, the quality of the image should be as high as possible without any loss of information or without any errors in image. Hence, the initial
step in the retrieval system is the pre-processing step in which the input images are resized and the redundant noise is removed from the images. The pre-processed images are further transformed into the frequency domain for extracting the texture features from the image as feature vectors. These feature vectors are defined in the form of a one-dimensional array which consists of the features retrieved by the feature extraction techniques. All the feature extraction techniques are given in the following sections. The extracted features are further stored in the form of feature vectors having the class names assigned and concealed as feature values. This is an iterative process until all the features are extracted and the classifier is trained using the three classification techniques.
Results: The dataset for the comparison work is taken from the Brats2015 database which consists of brain tumour images of various stages. Totally there are 200 many images taken and 50 of them are normal, 50 are benign, 50 are malignant and 50 are of severe stages. The classifier is trained by giving the labelled image features. The normal brain tumour images of patients are labelled as 1. The benign brain tumour images are labelled as 2. The malignant brain tumour images are labelled as 3. The severely affected brain tumour images of patients are labelled as 4. Some of the sample MRI brain tumour images of patients are given in Fig. 3.
Conclusion: This paper produces a CBIR architecture consisting of the procedures which are required to be followed while determining the brain tumour stages and prediction of the tumour stage. Texture is a vital visual attribute which determines both the human perception and image analysis systems. This research provides a comparative approach to identify the best feature extraction technique as Shearlet transform and the best classification algorithm as Multi-SVM based on the results which provided maximum accuracy. Although the Shearlet Transform took more time for computation, it performs best when combined with Multi- SVM in CBIR. Based on the experiments conducted, Ridgelet is the fastest and the Shearlet is the slowest for feature extraction. But, shearlet provided more desirable features. An application based on the best combination is constructed for performing tumour stage identification and which is used to give medical suggestions to doctors. The future enhancement for this work can be training upon a larger dataset of brain tumour images. The extracted features can be based on both the shape and texture feature fused together in the future. The research can be further preceded such that the overall cost can be reduced by improving the efficiency of the algorithm.
Keywords: Brain tumour detection, content based image retrieval, classification of tumours, image retrieval.