Classification of Ultrasound Images for Thyroid Detection using Fuzzy C-Means and Artificial Neural Network (ANN) Classifier Technique

Abstract

Nowadays the health issue is being worried a lot and so the workload of a doctor becomes comparatively huge.
In current scenario, unidentified thyroid has lots of serious medical issues. So, to reduce the workload, the image
segmentation and classification of thyroid ultrasound images is most necessary. In our proposed project, the ultrasound
images first undergo a pre-processing stage, where it involves processes like grayscale conversion, intensity calculation
and histogram equalization. Most commonly, K means clustering algorithm is being used for the segmentation of
ultrasound images. It is effective when the clusters are not overlapped. In our work, thyroid tumor region is segmented and
by using Fuzzy C Means (FCM) clustering it is extracted, which is advantageous compared to K means algorithm. After
segmentation, Scale Invariant Feature Transform(SIFT) algorithm is used to detect local features in images. Since the
thyroid ultrasound images are of different rotations and scales, SIFT is more efficient. The feature output that is obtained is
then applied to the PCA (principal component analysis) and then to GLCM as input. The main work of PCA is for
dimension reduction purpose. After the feature extraction, all features are extracted from each data and combined them
into a single matrix for classification. ANN is a family of artificial intelligence. For classification purposes, ANN is chosen
since it is self-learning, has high speed and parallelism, error tolerance and associative memory. Feedforward ANN that is
trained with backpropagation algorithm is a model of ANN that is used here. The most commonly adopted classifier is
SVM. In our proposal, classification is done by ANN because SVM is a binary classifier whereas ANN classifier is more
efficient compared to SVM.