Deep Learning based Classification of Thyroid Cancer using Different Medical Imaging Modalities : A Systematic Review

Abstract

Deep learning algorithms have achieved a tremendous triumph in task-specific feature classification. Deep learning methods are very much effective when a large amount of training data is scarce. It has been significantly applied for disease classification from medical imaging. The paper aims to identify and summarize the scenario of current research on thyroid cancer using deep learning methods through different medical imaging modalities which are found at present so that reseachers become capable to select a useful and the most relevant approach which might be fruitful in dealing with thyroid cancer. This may also raise a need for more work out while dealing with future challenges. This Systematic literature review (SLR) has been presented by reviewing research articles published in well-reputed venues between 2017 to 2021. A comprehensive review was performed to appraise the deep learning approaches that have been applied in classifying a thyroid nodule disorder from different medical imaging modalities. The analysis is performed based on different parameters reported in selected research studies which include classification accuracy, true-positive (TP), false-positive (FP), true-negative (TN), false-negative (FN) sensitivity, specificity, and the area under the receiver operating characteristic curve (AUROC). A total of 2,149 research studies have been obtained by applying search queries in different journals’ databases, out of them 40 papers have been selected for this SLR. Among them 22 studies have contributed sufficiently to the construction of the evaluation table which enabled the test process of methods of deep learning, having sensitivity varies between 75% to 100% (mean 89.50%) and specificity ranged from 64% to 100% (mean 84.4 %). The outputs revealed that the Convolutional Neural Network (CNN) has produced significant accuracy and has been extensively applied in the diagnosis of thyroid cancer by medical professionals. Furthermore, it is concluded that if the thyroid cancer exposure is inappropriate then it may restrict the deep learning mechanism and make its reliability challenge able.