IMPACT OF PREPROCESSING FILTERS ON THYROID DETECTION: LAPLACIAN FILTER AS AN OPTIMAL CHOICE
Abstract
This study presents a comparative analysis of different image enhancement filters for automated detection and classification of thyroid nodules in ultrasound images. The primary objective was to evaluate how preprocessing techniques influence the diagnostic accuracy of deep learning-based models. Four filters Gaussian, Laplacian of Gaussian, Unsharp, and Laplacian were applied to a balanced dataset containing benign and malignant thyroid images. Each enhanced image set was used to train and test a classifier for nodule detection. The results demonstrated that the Laplacian filter achieved the highest classification accuracy of 99.27%, outperforming Gaussian (97.80%), Unsharp (92.44%), and Laplacian of Gaussian (90.98%) filters. The Laplacian filter’s superior performance can be attributed to its second-order derivative operation, which accentuates rapid intensity transitions and enhances edge details critical for identifying the structural boundaries of thyroid nodules. These enhanced edge features allow the deep learning model to better distinguish between benign and malignant tissues. This study highlights that selecting an optimal image enhancement technique, particularly the Laplacian filter, significantly boosts diagnostic precision in computer-aided thyroid nodule detection systems.
Keywords : Thyroid Nodule; Ultrasound Imaging; Preprocessing; Laplacian Filter; Guassian Filter; Medical Imaging.
