| Thyroid nodules are one or more abnormal structure masses occurs within thyroid. Thyroid nodules are a common endocrine tumor; The nodules can be benign or malignant. About 7% of all the nodules are malignant. Thyroid cancer is the most common malignant tumor of the endocrine system, and is one of the ten major cancers in women. With the increasing use of neck imaging technology, the incidence of thyroid nodules is increasing. These nodules can be either benign or malignant, although most nodules are benign. Malignant nodules represent only approximately 7% of all nodules, but the incidence of thyroid cancer is increasing. Studies have reported that the prevalence of thyroid cancer has increased at a rate of 3% per year, one of the most rapidly increased cancers. Especially in the coastal city, the incidence of thyroid cancer has increaseed by an average of 4%, which is one of the fastest growing malignant tumor. Therefore, doctors must identify whether thyroid nodules are benign or malignant to make appropriate clinical decisions (e.g., clinical therapy and surgical methods). Malignant thyroid lesions should take the comprehensive treatment based on operation as soon as possible, while benign lesions can be treated conservatively,which means operations are not necessary, and some lesions may only need regular follow-up. Optimal and prompt treatment plans can ease patient pain and anxiety, reduce medical costs and improve patient prognoses. There are several methods that can be used to diagnose the nature of thyroid nodules, such as clinical examination (especially palpation), ultrasound, Ct, MRI, PET Ct and FNA, etc. The gold standard for the clinical diagnosis of thyroid nodules is fine-needle aspiration biopsy (FNA). However, FNA is invasive, resulting in physical and psychological harm to the patients. FNA may have the risk of injuring important arteries and nerves. Furthermore, FNA is too labor intensive and time-consuming to be used for large-scale screenings. Additionally, FNA may produce false negative results due to small nodules, cystic nodules,too small Puncture needle and the operator’s lack of experience. Therefore, FNA is only recommended when nodules are suspected of being malignant.Several imaging modalities can be used to identify the nature of thyroid nodules in the clinic, such as US, computed tomography (CT), magnetic resonance imaging (MRI) and positron emission tomography (PET). Of these modalities, US is most commonly used because it is expedient, efficient, inexpensive, non-invasive and non-radiative. Therefore, US is suitable for large-scale screenings. Many studies have demonstrated that US can differentiate between malignant and benign nodules by describing the size, shape and position of the nodules. Nevertheless, the following disadvantages affect the accuracy of US.1) Because US can contain significant speckles and noises, the normal anatomical background can be camouflaged.2) Due to the lack of unified standards, visual interpretation of the US images by ultrasonography and expert clinicians is typically subjective.3) Only doctors with sufficient knowledge and experience can provide extensive real-time evaluations. Other medical imaging methods,such as CT,MRI,PET,are not wildly used in clinical because of theirs Radiation damage, long-time-consuming or high expense.In conclusion, for a long time, due to the lack of effective methods for differentiating the nature of thyroid nodules before surgery, many patients accepted unnecessary surgical operation or FNA to know the nature of nodules. How to dig more useful clinical information the existing image and use them in the early diagnosis of thyroid nodules is a urgent clinical problems that need to be solved in practical work. Naturally, US images contain significant objective information, with clinical significance that the naked eye cannot identify. Therefore, various data mining techniques have been widely used to extract features from the raw images to characterize the patterns quantitatively. Among these techniques, texture features have been applied in several studies, particularly gray-level co-occurrence matrix (GLCM) texture features. It has been suggested that GLCM features can help to distinguish between benign and malignant lesions. This technique is a classic and commonly used method in extracting image features, including the components of angular second moment (energy), contrast, correlation, sum of squares, inverse difference moment, sum average, sum variance, sum entropy, entropy, difference variance, difference entropy, information measures of correlation, and maximal correlation coefficient. These extracted features typically exhibit clinical significance and can discriminate between malignant and benign nodules using characteristics that are unseen by the naked eye. Various pattern recognition models, particularly trees, boost, neural networks, SVM, and random forest, can be used to build discriminative functions to differentiate between malignant and benign thyroid nodules using receiver operating characteristic (ROC) curves.In the present study, a nonlinear median filter was first used to denoise the raw images. GLCM was further used to extract the above-mentioned features. To obtain the optimal discriminate functions to differentiate between malignant and benign thyroid nodules, the six abovementioned pattern recognition models were used to fit the feature dataset, and 10-fold cross-validation was used to select the optimal discriminate function with multiple indices, including accuracy, true positive rate, false positive rate, sensitivity, specificity, precision, recall, F-measure and area under the curve (AUC).Analysis results:1.All 16 variables (Energy, Entropy, Homogeneity, Inverse Difference Momentl, Inverse Difference Moment2, Correlation 1, Correlation2, Contrast, Variance, Maximum probability, Cluster Tendency, Mean, Difference-mean, Sum-Entropy, Difference-Entropy) extracted from thyroid nodules ultrasound images differed significantly between the benign and malignant nodules and also exhibited the potential power to discriminate between the nodules.2, The logistic model exhibited the best performance among all the six models (trees, boost, neural networks, SVM, logistic, and random forest), with optimal values for all 7 assessment indices, followed by the ANN model, random forest, boost, SVM and random tree models. The overall cross-matrix of the logistic model, with an AUC approaching 0.84 (asymptotic 95% confidence interval,0.781-0.906, p<0.05), predictive accuracy is 78.5%, True positive rate is 0.785, false positive rate is 0.215, sensitivity is 0.789, specificity 0.784。 Major innovative conclusions:An invasion and objective diagnostic opinion is needed in daily clinic to help the Differentiation of thyroid nodules. This study feed the GLCM texture features extracted from ultrasound images to six kind of pattern recognition and found that Sixteen GLCM features coupled with logistic regression provided the best prediction performance among the six models considered; thus, these features can be used as an adjunct diagnostic tool in daily clinical practice. Major innovation:1) The highlight innovation of the research is the construction of statistic pattern recognition, basing on the texture features extracted from ultrasound images, used to differentiate the thyroid nodules early, Objectively, quantitatively and standardization.2) Use the gray level co-occurrence matrix to extract the data features from Ultrasound image and provides a new idea for the development and utilization of B Ultrasound images.3) Use exhaustive modeling strategy to filter an efficient, sensitive, specific early differential diagnostic models and provide a new strategy to the construction of differentiating models of other diseases basing on medical imaging.Defects:1) The number of samples is small2) the features extracted are simple... |
PDF Full Text Request