Certain Investigations on Tissue Segmentation and Classification of Thyroid Ultrasound Images

Abstract

ABSTRACT newline newline newlineThyroid is a small butterfly shaped gland located in front of the neck just below the laryngeal prominence, commonly referred to as Adam s apple. Thyroid is one of the endocrine glands, which control metabolism by producing hormones. Diseases that are affecting the thyroid include hyperthyroidism, hypothyroidism, goiter, and thyroid nodules. Abnormalities of the thyroid gland are normally detected and classified using ultrasound imaging. Image processing techniques are becoming popular in medical applications. Medical image processing uses a variety of imaging modalities such as ultrasound (US), computed tomography (CT) and magnetic resonance image (MRI). Medical image processing plays an important role in diagnosing diseases. Ultrasound imaging is noninvasive, inexpensive and the human body is not subjected to any ionizing radiations. newline newlineThe main problem is regarding the selection of the segmentation method. All the segmentation methods have a problem related to the initial position. If the selection of initial position is not correct then the desired segment is not obtained. Modified region growing (MRGW) is used for segmentation of thyroid, which takes more computation time to segment the thyroid. In this method, depending on the selection of seed pixels the obtained region is also different. So here a boundary detection method based on edge following method is used for the segmentation of the thyroid. Next problem is to select the classifier. The classifiers like feed-forward back propagation network (FFBNN) and Adaptive Neuro Fuzzy Inference System (ANFIS) are used to test the performance of the classification. But the result shows less accuracy and high computation time. Training algorithm selection is another important problem. Hybrid learning algorithm takes more computation time to train the network. To increase the convergence rate and to improve the accuracy, Artificial Bee Colony (ABC) algorithm is used to tune the network. In this work an Adaptive Artificial Bee Colony (AABC) algorithm is used to optimize the parameters of Adaptive Neuro Fuzzy Inference System. newline newlineIn the proposed methodology, the images are fetched from the thyroid database and pre-processed using Adaptive Median Filter (AMF) in order to remove the speckle noise. Later, eight statistical features such as mean, variance, contrast, correlation, histogram, homogeneity, energy and Block Difference Inverse Possibility (BDIP) are extracted from the pre-processed images and given to the FFBNN to classify whether the given image is normal or abnormal. newline newline The image obtained in the classification process is subjected to segmentation process. Segmentation plays an important role in medical imaging to obtain the location of the object of interest as well as to detect the area. Here, a boundary detection method based on edge following is used. This provides efficient computation time than MRGW algorithm and other contour models. Then Local Gabor XOR Pattern (LGXP) and Gray Level Co-Occurrence Matrix (GLCM) features are extracted from the segmented image and given to FFBNN, ANFIS, ANFIS-ABC and ANFIS-AABC to classify the tissues as benign or malignant. newline newline The proposed technique, ANFIS-AABC is evaluated by giving more number of images to the well trained ANFIS. The accuracy of the proposed technique is 95.2%. The ANFIS-ABC, ANFIS and FFBNN have 85.7%, 76.1% and 71.4% of accuracy respectively. The experimental results show that the combination of ANFIS algorithm with AABC optimization algorithm produces better accuracy and less computation time as compared to other classification techniques. newline newline