No abstract

Introduction: Ultrasound images are one of the main contributors for evaluating of thyroid nodules. However, reading ultrasound imaging is not easy and strongly depends to doctors’ experiences. Therefore, a CAD system could assist doctors in evaluating thyroid ultrasound images to reduce the impact of subjective experience on the diagnostic results. Objective: with the best of our knowledge there is not any articles that actually provide a systematic review of deep learning application in analyzing ultrasound images of thyroid nodules and Hence, a comprehensive review of studies in this field can be useful, therefore the protocol of this systematic Review will be presented to reach this goal.Method: This protocol includes five stages: research questions definition, search strategy design, study selection, quality assessment and data extraction. We developed search for relevant English language articles using the PubMed, Scopus and Science Direct. Inclusion and exclusion criteria were defined and flow diagram is conducted, from 623 studies retrieved, 27 studies were included, after quality assessment data was extracted based on defined categories.Result: The result of this systematic review can help researchers by comprehensive view and the summary of evidence to present new ideas and further research and represent a state of the art in this field.Conclusion: in this study a protocol was used for doing a systematic review on various deep learning applications in thyroid ultrasound such as feature selection, classification, localization, detection and segmentation. Articles were screened based on the following items: study and patient information, dataset, method, results and comparison method.

Thyroid Computer-Aided Diagnosis (CAD) systems have been developed to assist radiologists in improving efficiency, reliability, and diagnosis performance. Often the performance of these CAD systems is evaluated with different datasets that make it incomparable. A valuable thyroid ultrasound (US) dataset is presented in this work. This dataset consists of 2450 thyroid US images from 2018 to 2020 in Prospective Epidemiological Research Studies in Mashhad, Iran (PERSIAN), a large national cohort study. These US images have the ROI of thyroid nodules and the associated American College of Radiology (ACR) Thyroid Imaging Reporting and Data System (TIRADS) features by expert physicians provided in XML format. Dataset’s images are categorized into five groups based on the ACR-TIRADS (Tirads1-Tirads5). The presented dataset is expected to be a valuable resource to develop and assess thyroid CAD systems to help radiologists better diagnose.