THYROID MALIGNANCY DETECTION USING ARTIFICIAL INTELLIGENCE

The rise in thyroid cancer cases and the issue of unreliable false positive diagnostic rates in expert-reviewed ultrasound images underscore the need for precise tumor diagnosis. Convolutional Neural Networks (CNNs), a cutting-edge deep learning technique, offer remarkable capabilities in tackling computer vision challenges. This paper introduces a fine-tuned VGG-19 CNN model tailored to multi-classify thyroid nodules in pre-processed ultrasound (US) images. The experimental results demonstrate that the proposed model achieves accuracies of 0.6521, 0.7572, and 0.9201 for 50, 100, and 150 epochs, respectively. Notably, the optimal testing loss manifests at the 100-epoch juncture, signifying an equilibrium between effective model training and generalization. Moreover, an intriguing observation is the gradual enhancement in Grad-CAM visualizations with increasing epochs, indicating the model's evolving proficiency in discerning pertinent features associated with thyroid malignancy.