CNN and ResNet50 Model Design for Improved Ultrasound Thyroid Nodules Detection
Ghufran Basim Alghanimi, Hadeel K. Aljobouri, Khaleel Akeash Al-shimmari · 2024
A thyroid nodule refers to the abnormal development of cells inside the thyroid gland, which can be caused by factors such as high iodine consumption, thyroid degeneration, excessive inflammation, and other disorders. Ultrasound has been used to detect thyroid nodules due to its high accuracy and efficiency in diagnosing. Nowadays, Deep learning model design is introduced to allow an accurate and efficient classification of thyroid nodules in ultrasound images. In this work, a Convolutional Neural Network (CNN) and transfer learning Residual Network (ResNet50) feature extraction techniques are used to classify thyroid nodule ultrasound imaging as benign and malignant. A free dataset was fed to the proposed model design. This dataset includes 800 images divided into 400 benign and 400 malignant. The suggested system was evaluated based on its accuracy, precision, recall, and F1 score. The classification accuracy for CNN was 91.25%, and for ReNet50 was 89.37%. This study could assist radiologists in overloaded medical centers in efficiently classifying malignant and benign thyroid nodule datasets, especially in the local health care centers.