CLASS BALANCING AND CLASSIFICATION OF THYROID DISORDER IN A PERSON USING MACHINE LEARNING

Abstract [English]

Thyroid nodules are a sign of a number of thyroid disorders, and medical image analysis is essential to their early identification and diagnosis. Within the current framework, machine learning techniques such as Support Vector Machine, Random Forest and Decision tree algorithm are implemented to classify thyroid nodules. In this study we suggest a novel application of transfer learning algorithms to the classification of thyroid nodules. Through the application of transfer learning, neural network models that have already been trained on big datasets can be adjusted for certain tasks that require less data. Our method involves extracting meaningful information from thyroid ultrasound scans using a cutting-edge convolutional neural network (CNN) that has been pre-trained on a variety of medical images. To optimise its performance for accurate classification, the model is trained on a particular dataset of thyroid nodule images. We examine the effectiveness of many transfer learning architectures, such as VGG16 and Xception CNN, and assess their overall accuracy, sensitivity, and specificity. The proposed methodology aims to provide physicians with a reliable thyroid problem diagnosis tool by increasing the categorization efficiency of thyroid nodules. The results pave the way for more precise thyroid image analysis diagnosis by demonstrating how transfer learning can be utilised to maximise model performance even in the presence of sparsely labelled medical data.