Enhanced Thyroid Nodule Classification Adopting Significant Features Selection

Appropriate selection of features play a crucial role for refining precision of classification systems. The classification accuracy and training speed may be significantly intensified by elimination of superfluous features. The present paper addresses the high dimensional data analysis problem throu...

Full description

Saved in:

Bibliographic Details
Published in:	2022 Fourth International Conference on Emerging Research in Electronics, Computer Science and Technology (ICERECT) pp. 1 - 5
Main Authors:	D, Poornima, Karegowda, Asha Gowda, M, Geetha, Hooli, Abhishek, Aparna, R, GK, Prashant
Format:	Conference Proceeding
Language:	English
Published:	IEEE 26-12-2022
Subjects:	Classification Feature extraction Feature selection Image segmentation Kernel Machine learning Multilayer perceptrons Preprocessing Refining Speckle filtering Thyroid nodule Training Ultrasonic imaging
Online Access:	Get full text
Tags:	Add Tag No Tags, Be the first to tag this record!

Description
Summary:	Appropriate selection of features play a crucial role for refining precision of classification systems. The classification accuracy and training speed may be significantly intensified by elimination of superfluous features. The present paper addresses the high dimensional data analysis problem through feature selection approach for refining the classification accuracy of Thyroid Nodules (TNs) as benign and malignant. Thyroid Ultrasound Images (TUS) containing nodules are first de-speckled and further improved using Canny Edge Detection (CED) method. This process is followed by application of segmentation technique Adaptive Regularized Kernel Fuzzy C-means (ARKFCM) where relevant Area of Interest (AOI) is obtained and using AOI, nineteen texture features are mined. Finally, feature subset selection is carried out using five different search methods- Genetic Search (GS), Best First (BF), Linear Forward Selection (LFS), Greedy Step Wise (GSW), and Subset Size Forward Selection (SSFS). Selected features are assessed using ten different classifiers Bayes Net, Naïve Bayes, Logistic, Multilayer Perceptron, Radial Basis Function, Sequential Minimal Optimization, Instance Based K-nearest neighbor, K-star, J-48 and Random Tree. Experimental evaluation revealed, features listed using five search techniques have boosted performance of all considered classifiers in comparison to their performance using original nineteen features.
DOI:	10.1109/ICERECT56837.2022.10060526