RESEARCH ARTICLE Thyroid nodules classification and diagnosis in ultrasound images using fine-tuning deep convolutional neural network Olfa Moussa 1 | Hajer Khachnaoui 1 | Ramzi Guetari 2 | Nawres Khlifa 1 1 Laboratoire de Biophysique et Technologies Médicales, Université de Tunis El Manar, Tunis, Tunisia 2 Institut Superieur d'Informatique, Université de Tunis El Manar, Tunis, Tunisia Correspondence Olfa Moussa, Laboratoire de Biophysique et Technologies Médicales, Université de Tunis El Manar, 1006 Tunis, Tunisia. Email: olfa.moussa@gmail.com Abstract Ultrasonography AKA diagnostic sonography is a noninvasive imaging technique that allows the analysis of an organic structure, thanks to the ultrasonic waves. It is a valu- able diagnosis method and is also seen as the evidence-based diagnostic method for thyroid nodules. The diagnosis, however, is visually made by the practitioner. The automatic discrimination of benign and malignant nodules would be very useful to report Thyroid Imaging Reporting. In this paper, we propose a fine-tuning approach based on deep learning using a Convolutional Neural Network model named resNet- 50. This approach allows improving the effectiveness of the classification of thyroid nodules in ultrasound images. Experiments have been conducted on 814 ultrasound images and the results show that our proposed approach dramatically improves the accuracy of the classification of thyroid nodules and outperforms The VGG-19 model. KEYWORDS computer-aided diagnosis (CAD) system, deep convolutional neural network, deep learning, fine- tuning, thyroid nodule, ultrasound image 1 | INTRODUCTION The thyroid is an endocrine gland that secretes thyroid hor- mones. These hormones mainly influence the metabolic rate and protein synthesis. They also have many other effects espe- cially those having an importance regarding the development of the human body. Various disorders can affect the good functioning of the thyroid. Among the problems that can affect the thyroid there is the appearance of a thyroid nodule. This latter is a small mass formed at the level of the thyroid gland and requires a medical assessment in order to make an accurate diagnosis of the cause of the nodule and determine the possible treatment. There are several types of thyroid nodules. The colloidal nodule is the most common one and consists of normal cells. The cysts, mostly benign, are formations, up to several centi- meters in diameter, filled with fluid. An inflammatory nod- ule can occur most often to people with thyroiditis. The nodule can also be linked to an adenoma, benign tumor, which evolves slowly. In about 5% to 10% of the cases, it can be a cancer of the thyroid which concerns the women in 75% of the cases and which is usually perfectly treated. The thyroid assessment is systematically requested by doctors when they note the presence of nodules. It consists in a set of biological analyzes performed from a blood test. It usually does not prejudge the malignant or benign nature of the nodule, but it allows to know if the operation of the thyroid is normal or not. The examination of the neck region by ultrasound is systematically performed also. This exami- nation allows doctors to measure the size of the thyroid, the number, the size as well as the shape of the nodules and to detect the possible presence of other abnormalities. A thyroid ultrasound examination provides information about the structure of the nodules and their characteristics. This helps the diagnosis of various thyroid disorders, includ- ing echogenicity, texture, composition (cystic, solid, or mixed), existence or absence of microcalcification, halo, and intranodular blood flow. Received: 15 December 2018 Revised: 23 July 2019 Accepted: 1 August 2019 DOI: 10.1002/ima.22363 Int J Imaging Syst Technol. 2019;1–11. wileyonlinelibrary.com/journal/ima © 2019 Wiley Periodicals, Inc. 1