Exploring Transfer Learning, Fine-tuning of Thyroid Ultrasound Images

Deep Learning for medical imaging has been on the forefront of its numerous 1 applications, thanks to its versatility and robustness in deployment. In this paper 2 we explore various classification methodologies that are employed for datasets 3 of relatively small in size to actually train a deep learning algorithm from scratch. 4 Thyroid ultrasound images are classified using a small CNN from scratch, transfer 5 learning and fine-tuning of Inception-v3, VGG-16. We present a comparison of 6 the aforementioned methods through accuracy, sensitivity and specificity. 7