Influence of the measurement method of features in ultrasound images of the thyroid in the diagnosis of Hashimoto's disease

A systematic review of machine learning based thyroid tumor characterisation using ultrasonographic images

Ultrasonography is widely used to screen thyroid tumors because it is safe, easy to use, and low-cost. However, it is simultaneously affected by speckle noise and other artifacts, so early detection of thyroid abnormalities becomes difficult for the radiologist. Therefore, various researchers continuously address the limitations of sonography and improve the diagnosis potential of US images for thyroid tissue from the last three decays. Accordingly, the present study extensively reviewed various CAD systems used to classify thyroid tumor US (TTUS) images related to datasets, despeckling algorithms, segmentation algorithms, feature extraction and selection, assessment parameters, and classification algorithms. After the exhaustive review, the achievements and challenges have been reported, and build a road map for the new researchers.

Convolutional Neural Network-Based Computer-Assisted Diagnosis of Hashimoto's Thyroiditis on Ultrasound

PURPOSE

This study investigates the efficiency of deep learning models in the automated diagnosis of Hashimoto's thyroiditis (HT) using real-world ultrasound data from ultrasound examinations by computer-assisted diagnosis (CAD) with artificial intelligence.

METHODS

We retrospectively collected ultrasound images from patients with and without HT from 2 hospitals in China between September 2008 and February 2018. Images were divided into a training set (80%) and a validation set (20%). We ensembled 9 convolutional neural networks (CNNs) as the final model (CAD-HT) for HT classification. The model's diagnostic performance was validated and compared to 2 hospital validation sets. We also compared the accuracy of CAD-HT against seniors/junior radiologists. Subgroup analysis of CAD-HT performance for different thyroid hormone levels (hyperthyroidism, hypothyroidism, and euthyroidism) was also evaluated.

RESULTS

39 280 ultrasound images from 21 118 patients were included in this study. The accuracy, sensitivity, and specificity of the HT-CAD model were 0.892, 0.890, and 0.895, respectively. HT-CAD performance between 2 hospitals was not significantly different. The HT-CAD model achieved a higher performance (P < 0.001) when compared to senior radiologists, with a nearly 9% accuracy improvement. HT-CAD had almost similar accuracy (range 0.87-0.894) for the 3 subgroups based on thyroid hormone level.

CONCLUSION

The HT-CAD strategy based on CNN significantly improved the radiologists' diagnostic accuracy of HT. Our model demonstrates good performance and robustness in different hospitals and for different thyroid hormone levels.

The role of histogram analysis of grayscale sonograms to differentiate thyroid nodules identified by 18F-FDG PET-CT

The role of histogram based on ultrasound (US) images for thyroid nodules found in fluorine-18 fluorodeoxyglucose (18F-FDG) Positron Emission Tomography/Computed Tomography (PET-CT) is unknown. We aimed to assess whether histogram analysis using gray scale US could differentiate thyroid nodules detected by PET-CT.In this study, 71 thyroid nodules ≥1 cm were identified in 71 patients by conducting 18F-FDG PET-CT, from January 2010 to June 2013. Subsequently, either grayscale US-guided fine needle aspirations or core needle biopsies were performed on each patient. Each grayscale US feature was categorized according to the Korean Thyroid Imaging Reporting and Data System (K-TIRADS). Histogram parameters (skewness, kurtosis, intensity, uniformity, and entropy) were extracted from the grayscale US images followed by statistical analysis using the Chi-Squared or Mann-Whitney U tests.The 71 nodules comprised 30 (42.3%) benign nodules, 30 (42.3%) primary thyroid malignancies, and 11 (15.4%) metastatic lesions. Tumor size, US findings, and histogram parameters were significantly different between the benign and malignant thyroid nodules (P = .011, P = .000, and P < .02, respectively). A comparison showed that parallel orientation and an absence of calcifications were found more frequently in metastatic thyroid nodules than in primary thyroid malignancies (P = .04, P < .000, respectively). However, histogram parameters and K-TIRADS were not significantly different between primary thyroid malignancies and metastatic lesions.There is a limit to replacing cytopathological confirmation with texture analysis for the differentiation of thyroid nodules detected by PET-CT. Therefore, cytopathological confirmation of nodules appearing malignant on US images cannot be avoided for an ultimate diagnosis of metastasis.

Parametrical modelling for texture characterization-A novel approach applied to ultrasound thyroid segmentation

Texture analysis is an important topic in Ultrasound (US) image analysis for structure segmentation and tissue classification. In this work a novel approach for US image texture feature extraction is presented. It is mainly based on parametrical modelling of a signal version of the US image in order to process it as data resulting from a dynamical process. Because of the predictive characteristics of such a model representation, good estimations of texture features can be obtained with less data than generally used methods require, allowing higher robustness to low Signal-to-Noise ratio and a more localized US image analysis. The usability of the proposed approach was demonstrated by extracting texture features for segmenting the thyroid in US images. The obtained results showed that features corresponding to energy ratios between different modelled texture frequency bands allowed to clearly distinguish between thyroid and non-thyroid texture. A simple k-means clustering algorithm has been used for separating US image patches as belonging to thyroid or not. Segmentation of thyroid was performed in two different datasets obtaining Dice coefficients over 85%.

Histogram analysis of greyscale sonograms to differentiate between the subtypes of follicular variant of papillary thyroid cancer

AIM

To evaluate the diagnostic value of histogram analysis using ultrasound (US) to differentiate between the subtypes of follicular variant of papillary thyroid carcinoma (FVPTC).

MATERIALS AND METHODS

The present study included 151 patients with surgically confirmed FVPTC diagnosed between January 2014 and May 2016. Their preoperative US features were reviewed retrospectively. Histogram parameters (mean, maximum, minimum, range, root mean square, skewness, kurtosis, energy, entropy, and correlation) were obtained for each nodule.

RESULTS

The 152 nodules in 151 patients comprised 48 non-invasive follicular thyroid neoplasm with papillary-like nuclear features (NIFTPs; 31.6%), 60 invasive encapsulated FVPTCs (EFVPTCs; 39.5%), and 44 infiltrative FVPTCs (28.9%). The US features differed significantly between the subtypes of FVPTC. Discrimination was achieved between NIFTPs and infiltrative FVPTC, and between invasive EFVPTC and infiltrative FVPTC using histogram parameters; however, the parameters were not significantly different between NIFTP and invasive EFVPTC.

CONCLUSION

It is feasible to use greyscale histogram analysis to differentiate between NIFTP and infiltrative FVPTC, but not between NIFTP and invasive EFVPTC. Histograms can be used as a supplementary tool to differentiate the subtypes of FVPTC.

Improvement of the classification quality in detection of Hashimoto’s disease with a combined classifier approach

The purpose of the study was to construct an efficient classifier that, along with a given reduced set of discriminant features, could be used as a part of the computer system in automatic identification and classification of ultrasound images of the thyroid gland, which is aimed to detect cases affected by Hashimoto’s thyroiditis. A total of 10 supervised learning techniques and a majority vote for the combined classifier were used. Two models were proposed as a result of the classifier’s construction. The first one is based on the K-nearest neighbours method (for K = 7). It uses three discriminant features and affords sensitivity equal to 88.1%, specificity of 66.7% and classification error at a level of 21.8%. The second model is a combined classifier, which was constructed using three-component classifiers. They are based on the K-nearest neighbours method (for K = 7), linear discriminant analysis and a boosting algorithm. The combined classifier is based on 48 discriminant features. It allows to achieve the classification sensitivity equal to 88.1%, specificity of 69.4% and classification error at a level of 20.5%. The combined classifier allows to improve the classification quality compared to the single model. The models, built as a part of the automatic computer system, may support the physician, especially in first-contact hospitals, in diagnosis of cases that are difficult to recognise based on ultrasound images. The high sensitivity of constructed classification models indicates high detection accuracy of the sick cases, and this is beneficial to the patients from a medical point of view.

Influence of short-term selenium supplementation on the natural course of Hashimoto's thyroiditis: clinical results of a blinded placebo-controlled randomized prospective trial

BACKGROUND

The real efficacy of selenium supplementation in Hashimoto's thyroiditis (HT) is still an unresolved issue.

OBJECTIVES

We studied the short-term effect of L-selenomethionine on the thyroid function in euthyroid patients with HT. Our primary outcome measures were TSH, thyroid hormones, thyroid peroxidase antibody (TPOAb), thyroglobulin antibody (TGAb) levels and thyroid echogenicity after 6 months of L-selenomethionine treatment. The secondary outcome measure was serum CXCL10 levels.

METHODS

In a placebo-controlled randomized prospective study, we have enrolled untreated euthyroid patients with HT. Seventy-six patients were randomly assigned to receive L-selenomethionine 166 µg/die (SE n = 38) or placebo (controls n = 38) for 6 months. TSH, free T₄ (FT₄), free T₃ (FT₃), TPOAb and CXCL10 serum levels were assayed at time 0, after 3 and 6 months. An ultrasound examination of the left and right thyroid lobe in transverse and longitudinal sections was performed. A rectangular region, the region of interest, was selected for analysis.

RESULTS

TSH, FT4, FT3, TPOAb, thyroid echogenicity and CXCL10 were not statistically different between SE and control groups at time 0, after 3 and 6 months. In the SE group, FT₄ levels were significantly decreased (P < 0.03) after 3 months, while FT₃ increased (P < 0.04) after 3 and 6 months versus baseline values. In the control group, the FT₃ decreased after 3 and 6 months (P < 0.02) compared to baseline.

CONCLUSION

The short-term L-selenomethionine supplementation has a limited impact on the natural course in euthyroid HT. Our results tip the balance toward the ineffectiveness of short-term L-selenomethionine supplementation in HT.

Quantitative assessment of the impact of biomedical image acquisition on the results obtained from image analysis and processing

Introduction

Dedicated, automatic algorithms for image analysis and processing are becoming more and more common in medical diagnosis. When creating dedicated algorithms, many factors must be taken into consideration. They are associated with selecting the appropriate algorithm parameters and taking into account the impact of data acquisition on the results obtained. An important feature of algorithms is the possibility of their use in other medical units by other operators. This problem, namely operator’s (acquisition) impact on the results obtained from image analysis and processing, has been shown on a few examples.

Material and method

The analysed images were obtained from a variety of medical devices such as thermal imaging, tomography devices and those working in visible light. The objects of imaging were cellular elements, the anterior segment and fundus of the eye, postural defects and others. In total, almost 200'000 images coming from 8 different medical units were analysed. All image analysis algorithms were implemented in C and Matlab.

Results

For various algorithms and methods of medical imaging, the impact of image acquisition on the results obtained is different. There are different levels of algorithm sensitivity to changes in the parameters, for example: (1) for microscope settings and the brightness assessment of cellular elements there is a difference of 8%; (2) for the thyroid ultrasound images there is a difference in marking the thyroid lobe area which results in a brightness assessment difference of 2%. The method of image acquisition in image analysis and processing also affects: (3) the accuracy of determining the temperature in the characteristic areas on the patient’s back for the thermal method - error of 31%; (4) the accuracy of finding characteristic points in photogrammetric images when evaluating postural defects – error of 11%; (5) the accuracy of performing ablative and non-ablative treatments in cosmetology - error of 18% for the nose, 10% for the cheeks, and 7% for the forehead. Similarly, when: (7) measuring the anterior eye chamber – there is an error of 20%; (8) measuring the tooth enamel thickness - error of 15%; (9) evaluating the mechanical properties of the cornea during pressure measurement - error of 47%.

Conclusions

The paper presents vital, selected issues occurring when assessing the accuracy of designed automatic algorithms for image analysis and processing in bioengineering. The impact of acquisition of images on the problems arising in their analysis has been shown on selected examples. It has also been indicated to which elements of image analysis and processing special attention should be paid in their design.

Influence of the measurement method of features in ultrasound images of the thyroid in the diagnosis of Hashimoto's disease

INTRODUCTION

This paper shows the influence of a measurement method of features in the diagnosis of Hashimoto's disease. Sensitivity of the algorithm to changes in the parameters of the ROI, namely shift, resizing and rotation, has been presented. The obtained results were also compared to the methods known from the literature in which decision trees or average gray level thresholding are used.

MATERIAL

In the study, 288 images obtained from patients with Hashimoto's disease and 236 images from healthy subjects have been analyzed. For each person, an ultrasound examination of the left and right thyroid lobe in transverse and longitudinal sections has been performed.

METHOD

With the use of the developed algorithm, a discriminant analysis has been conducted for the following five options: linear, diaglinear, quadratic, diagquadratic and mahalanobis. The left and right thyroid lobes have been analyzed both together and separately in transverse and longitudinal sections. In addition, the algorithm enabled to analyze specificity and sensitivity as well as the impact of sensitivity of ROI shift, repositioning and rotation on the measured features.

RESULTS AND SUMMARY

The analysis has shown that the highest accuracy was obtained for the longitudinal section (LD) with the method of linear, yielding sensitivity = 76%, specificity = 95% and accuracy ACC = 84%. The conducted sensitivity assessment confirms that changes in the position and size of the ROI have little effect on sensitivity and specificity. The analysis of all cases, that is, images of the left and right thyroid lobes in transverse and longitudinal sections, has shown specificity ranging from 60% to 95% and sensitivity from 62% to 89%. Additionally, it was shown that the value of ACC for the method using decision trees as a classifier is equal to 84% for the analyzed data. Thresholding of average brightness of the ROI gave ACC equal to 76%.