Thyroid nodules, which are characterised as aberrant thyroid cell growth, is due to heavy intake of iodine, or thyroid degeneration or the inflammation, and some diseases. Despite the fact that thyroid nodules are mostly benign, the likelihood that they are malignant rises noticeably every year. The labour of medical practitioners are reduced and optional fine needle aspiration, surgical excision are avoided. Firstly, FNA results are occasionally ambiguous, which means they do not always say for sure whether a nodule is benign or cancerous. This can happen when there aren't enough cells in the FNA sample or when the cellular properties are unclear. Uncertainty and the need for extra testing or repeated biopsies can be brought on by indeterminate results, which can be expensive and stressful. A number of studies have been done to identify thyroid nodules with the use of image recognition analysis based on deep-learning.  A new deep learning architecture is provided in this study. It reliably identifies the benign and malignant thyroid nodules among given dataset. First, we have considered both Vgg19 and Resnet50 models and they were pre trained on the Image Net database and then trained using the thyroid ultrasound image dataset and tested with the testing dataset. In comparison, we have achieved greater accuracy using Resnet50 model rather than Vgg19. The trained model has now able to classify the thyroid nodules into different categories like normal, benign and malignant. A website is also developed which can predict the type of nodule using the accurate model. Overall, the proposed model demonstrated that ultrasound images and deep learning may be used to distinguish between benign and cancerous thyroid nodules with a 75% accuracy and