Classification of Thyroid Nodules in Ultrasound Images Using Direction-Independent Features Extracted by Two-Threshold Binary Decomposition

Abstract

In recent years, several computer-aided diagnosis systems emerged for the diagnosis of thyroid gland disorders using ultrasound imaging. These systems based on machine learning algorithms may offer a second opinion to radiologists by evaluating a malignancy risk of thyroid tissue, thus increasing the overall diagnostic accuracy of ultrasound imaging. Although current computer-aided diagnosis systems exhibit promising results, their use in clinical practice is limited. One of the main limitations is that the majority of them use direction-dependent features. Our intention has been to design a computer-aided diagnosis system, which will use only direction-independent features, that is, it will not be dependent on the orientation and the inclination angle of the ultrasound probe when acquiring the image. We have, therefore, applied histogram analysis and segmentation-based fractal texture analysis algorithm, which calculates direction-independent features only. In our study, 40 thyroid nodules (20 malignant and 20 benign) were used to extract several features, such as histogram parameters, fractal dimension, and mean brightness value in different grayscale bands (obtained by 2-threshold binary decomposition). The features were then used in support vector machine and random forests classifiers to differentiate nodules into malignant and benign classes. Using leave-one-out cross-validation method, the overall accuracy was 92.42% for random forests and 94.64% for support vector machine. Results show that both methods are useful in practice; however, support vector machine provides better results for this application. Proposed computer-aided diagnosis system can provide support to radiologists in their current diagnosis of thyroid nodules, whereby it can optimize the overall accuracy of ultrasound imaging.

The Diagnostic Efficiency of Ultrasound Computer–Aided Diagnosis in Differentiating Thyroid Nodules: A Systematic Review and Narrative Synthesis

“…Of these, seven analyzed the performance of various statistical textural features [33,34,35,36,37,38,39], while six studies analyzed performance of a combination of textural features and other features, namely texture and wavelet transform features [40,41,42], texture and morphological features [43], and texture and radiological features [44], as well as texture analysis, elastography and grey scale ultrasound [45]. Two studies evaluated the performance of the combination of histogram and fractal texture analysis for support vector machine (SVM) and random forest classifiers [46,47] and one study assessed the accuracy of wavelet texture analysis for different classifiers [48]. Three studies focused on artificial intelligence texture analysis; two evaluated the diagnostic performance of the combination of artificial neural network (ANN) textural analysis with SVM [49], and ANN with binary logistic regression analysis [50], while another evaluated deep learning convolutional neural network feature classification performance using a random forest classifier [51].…”

Diagnostic Value of Machine Learning-Based Quantitative Texture Analysis in Differentiating Benign and Malignant Thyroid Nodules

“…In the work by Chang et al [22], the support vector machines classifier showed a diagnostic accuracy reaching up to 98.3% for 59 nodules. A recent work by Prochazka et al [38] employed a random forest and support vector machine classifier for evaluating segmentation-based fractal texture analysis, and the authors achieved a diagnostic accuracy of 94.3% with their model. Although all studies mentioned above have yielded promising results, the use of a small sample size in ML-based diagnostic models will undoubtedly introduce bias and variance [39].…”

Digital Medicine in Thyroidology: A New Era of Managing Thyroid Disease

“…In their review, Acharya et al [9] claimed that ML classifiers using non-clinical features showed higher classification accuracies than were obtained by radiologists using clinical features, and also presented the following additional advantages of using non-clinical features: (1) such features can be automatically extracted from US images and quantified as numerical values that can be used for automated classification; (2) there is no need for visual inspection and interpretation, and therefore, the results are more objective; and (3) such CAD systems can be written as software applications at a low cost and installed on existing computers in clinics at no extra cost. The accuracy of ML classifiers for thyroid nodule classification in recently published studies is shown in Table 1 [121314151617].…”

“…Reproduced from Prochazka et al [17]. DWT, discrete wavelet transform; CEUS, contrast-enhanced ultrasonography; kNN, k-nearest neighbor; SVM, support vector machine; HRUS, high-resolution ultrasound; SFTA, segmentation-based fractal texture analysis; RF, random forest.…”