Advanced Deep Learning Method for Thyroid Nodule Detection and Evaluation in Ultrasound Images

Thyroid nodule detection in its earliest stages is a difficult endeavor. Thyroid nodule diagnosis requires a lot of time and effort if done manually. The original ultrasound imaging of thyroid nodules is very variable due to differences in equipment and technicians. The issue with nodules is determining whether or not they are cancerous. Thyroid nodule evaluation through ultrasonography is presently advised as the first line of treatment. Ultrasound imaging can be helpful in making a diagnosis; But this relies heavily on the proficiency of the radiologist. There is a great need for faster and more efficient automated techniques of screening ultrasound pictures. This research study presents a deep learning method to detect thyroid nodules quickly and accurately. Some image processing techniques are required for optimal utilization of ultrasound imaging data of thyroid nodules. In this study, authors used deep neural networks InceptionResnet V2 and image augmentation to create a system for autonomous thyroid nodule categorization. For testing, neural networks with four-layer network nodes and the Local Laplacian Filtering technique are used to improve the image quality. The proposed method's reliability is evaluated by using the Thyroid Digital Image Database (TDID), a freely available dataset. The comparative findings demonstrate that the proposed method is superior to the current state of advancement in machine learning. Based on experiments, it is evident that the proposed technique has an Accuracy 0.99, Precision 0.98, Recall 0.96, Fl-Score 0.98, AVC-ROC 0.99.