A Benign and Malignant Pattern Identification in Cytopathological Images of Thyroid Nodules using Gabor Filter and Neural Networks
Balasubramanian Gopinath · Asian Journal For Convergence In Technology (AJCT) ISSN -2350-1146 · 2018
This research work presents an automated pattern recognition system to discriminate benign and malignant thyroid nodules using Gabor features based Neural Network classifiers. In the preprocessing step, the required regions of multi-stained Fine Needle Aspiration Cytology images of thyroid nodules are automatically cropped. Then, the segmentation of foreground information is performed using mathematical morphology technique. In the post processing step, the significant statistical features are extracted from the segmented images with the help of Gabor filter under various frequencies and orientations. Finally, the benign and malignant image patterns are discriminated using Elman Neural Network and Auto-associative Neural Network. Based on the performance analysis, the discrimination accuracy of 93.33% is obtained by the Elman Neural Network classifier for the statistical features extracted by Gabor filter bank. However, the auto-associative Neural Network classifier reports a highest discrimination accuracy of 96.66% for the same configuration. This automated discrimination system can be used as a second opinion tool for thyroid nodule analysis by the pathologists.