Predicting BRAFV600E mutations in papillary thyroid carcinoma using six machine learning algorithms based on ultrasound elastography

Oct 13, 2023

Scientific Reports volume 13, Article number: 12604 (2023) Cite this article

Metrics details

The most common BRAF mutation is thymine (T) to adenine (A) missense mutation in nucleotide 1796 (T1796A, V600E). The BRAFV600E gene encodes a protein-dependent kinase (PDK), which is a key component of the mitogen-activated protein kinase pathway and essential for controlling cell proliferation, differentiation, and death. The BRAFV600E mutation causes PDK to be activated improperly and continuously, resulting in abnormal proliferation and differentiation in PTC. Based on elastography ultrasound (US) radiomic features, this study seeks to create and validate six distinct machine learning algorithms to predict BRAFV6OOE mutation in PTC patients prior to surgery. This study employed routine US strain elastography image data from 138 PTC patients. The patients were separated into two groups: those who did not have the BRAFV600E mutation (n = 75) and those who did have the mutation (n = 63). The patients were randomly assigned to one of two data sets: training (70%), or validation (30%). From strain elastography US images, a total of 479 radiomic features were retrieved. Pearson's Correlation Coefficient (PCC) and Recursive Feature Elimination (RFE) with stratified tenfold cross-validation were used to decrease the features. Based on selected radiomic features, six machine learning algorithms including support vector machine with the linear kernel (SVM_L), support vector machine with radial basis function kernel (SVM_RBF), logistic regression (LR), Naïve Bayes (NB), K-nearest neighbors (KNN), and linear discriminant analysis (LDA) were compared to predict the possibility of BRAFV600E. The accuracy (ACC), the area under the curve (AUC), sensitivity (SEN), specificity (SPEC), positive predictive value (PPV), negative predictive value (NPV), decision curve analysis (DCA), and calibration curves of the machine learning algorithms were used to evaluate their performance. ① The machine learning algorithms' diagnostic performance depended on 27 radiomic features. ② AUCs for NB, KNN, LDA, LR, SVM_L, and SVM_RBF were 0.80 (95% confidence interval [CI]: 0.65–0.91), 0.87 (95% CI 0.73–0.95), 0.91(95% CI 0.79–0.98), 0.92 (95% CI 0.80–0.98), 0.93 (95% CI 0.80–0.98), and 0.98 (95% CI 0.88–1.00), respectively. ③ There was a significant difference in echogenicity,vertical and horizontal diameter ratios, and elasticity between PTC patients with BRAFV600E and PTC patients without BRAFV600E. Machine learning algorithms based on US elastography radiomic features are capable of predicting the likelihood of BRAFV600E in PTC patients, which can assist physicians in identifying the risk of BRAFV600E in PTC patients. Among the six machine learning algorithms, the support vector machine with radial basis function (SVM_RBF) achieved the best ACC (0.93), AUC (0.98), SEN (0.95), SPEC (0.90), PPV (0.91), and NPV (0.95).

The BRAFV600E mutation is a significant contributor to the papillary thyroid carcinoma (PTC) phenotype, which aids in the diagnosis and differential diagnoses of PTC before surgery1,2. BRAFV600E diagnosis requires genetic testing of cell eluate by ultrasound-guided fine-needle aspiration (FNA), which is invasive. Ultrasound-guided FNA cytological examination of thyroid nodules can diagnose PTC before surgery, but 15% to 30% of the cytological results belong to the Bethesda system definition with uncertain detection results (including Bethesda Type III: atypical lesions or follicular lesions of unknown significance (AUS/FLUS), Type IV: follicular tumors/suspected follicular tumors, and Type V: suspected malignant tumors (SUSP)). Therefore, "TBSRTC Classification Malignant Risk and Management Recommendation" recommends FNA cytology combined with BRAFV600E mutation detection, but are all invasive. As a result, it is critical in clinical practice to adopt non-invasive approaches to forecast the status of BRAFV600E mutations, so as to reduce the FNA and molecular detection rate.

More importantly, According to earlier research, BRAFV600E mutation in thyroid tumors is thought to be a sign of severe illness and PTC-related mortality3. It is interesting to note that the presence of the BRAFV600E mutation has become a more reliable molecular marker for PTC recurrence3. Therefore, finding BRAFV600E mutations in thyroid tumors has implications for prognosis and serves as a marker for tumor recurrence. The BRAFV600E mutation is also strongly linked to the existence of extrathyroidal extension (ETE) and cervical lymph node metastasis (CLNM) in PTC patients, suggesting an invasion4,5. Jin et al.6 observed a substantial relationship between BRAFV600E mutation with CLNM and ETE in a Mayo Clinic research. Xing et al.7 revealed a close link between BRAFV600E mutation and ETE, CLNM, and advanced illness stages in a large comprehensive international multicenter investigation. Even if either gene mutation is capable of identifying aggressive cancer types, the genetic analysis still requires specimen tissue for examination, which is often collected by invasive surgical procedures. Furthermore, identifying tumors using FNA biopsy samples from type 4a nodules is difficult because tumor cells vary in quantity, quality, and purity8. A sensitive and precise BRAFV600E mutation detection method will thus aid in the early detection of PTC9. Because BRAFV600E gene mutations produce 99.8% of malignant nodules, it is a significant tumor marker for PTC.

With the increasing popularity of tumor thermal ablation technology (microwave, radio frequency, and laser) in China, an increasing number of patients with PTC, particularly those with minimal papillary thyroid cancer (MPTC), are willing to accept thermal ablation as a minimally invasive procedure to maximize the preservation of thyroid function. As a result, proper preoperative detection of BRAFV600E status in PTC patients is critical for patients to choose therapy methods. Surgery is currently chosen over ablation for highly invasive PTC in China.

The main imaging method for evaluating thyroid nodules is ultrasound (US)10. Grayscale US has recently been found to be able to predict the mutation of BRAFV600E in PTC. However, the findings are still up for debate11,12, which may be related to the drawbacks of the conventional US image, such as its dependence on the radiologist's experience and interobserver variation13. The ability of a tissue to resist deformation whenever a force is exerted on it or to restore its initial form when that force is withdrawn is called elasticity, which is what elastography often examines in tissues. Depending on the Doppler US technology, strain images can be presented in grayscale or in colors that reflect the stiffness and elasticity of the tissues14,15,16,17. Elastography is also based on the grayscale US which is also subjective and operator dependent. There have been several significant developments to increase the performance of US elastography such as elastography estimation from Doppler imaging using central difference and least-squares algorithms17.

Radiomics analysis using US images has been used to forecast the molecular characteristics of several malignancies, which include PTC18,19. Artificial intelligence (AI) has considerably expanded in recent years as a cutting-edge data analysis tool in the medical field20. As a result of its extensive digital data sets, radiology in particular is well suited for AI21. Recently, there has been a lot of interest in the medical area in the use of radiomics in conjunction with machine learning, which is an important subset of AI and plays a tremendous supporting role in increasing diagnostic and prognostic accuracy22,23. However, there are currently few studies that used machine learning models primarily on elastography US radiomics data to detect the existence of BRAFV6OOE mutation in PTC. There has also been no research that employed different machine learning classifiers in assessing BRAFV6OOE in PTC patients using elastography US radiomic features. Based on elastography US radiomic features, this study seeks to create and validate six distinct machine-learning algorithms to predict BRAFV6OOE mutation in PTC patients prior to surgery.

A retrospective analysis was performed on PTC patients who had undergone preoperative thyroid US elastography, BRAFV600E mutation diagnosis, and surgery at Jiangsu University Affiliated People's Hospital and traditional Chinese medicine hospital of Nanjing Lishui District between January 2014 and 2021. The enrolling process is displayed in Fig. 1. 138 PTCs of 138 patients (mean age, 41.63 ± 11.36 [range, 25–65] years) were analyzed in this study. The patients were divided into BRAFV600E mutation-free group (n = 75) and BRAFV600E mutation group (n = 63). Using a stratified sample technique at a 7:3 ratio, all patients were randomly assigned to either the training group (n = 96) or the validation group (n = 42). The following criteria were required for inclusion: postoperative pathology indicated PTC; preoperative thyroid US elastography evaluation; related US images and diagnostic outcomes; maximum nodule diameter > 5 mm, and < 5 cm; and unilateral and single focal lesion. The exclusion criteria included a maximum nodule diameter of > 5 cm and indistinct US imaging of nodules caused by artifacts. The clinical details of the enrolled patients were documented, including age, sex, nodule diameter, nodule location, nodular echo, nodule boundary, nodule internal and peripheral blood flow, nodule elastic grading, calcification, CLNM, and BRAFV600E mutation results. The Jiangsu University Affiliated People's Hospital and the traditional Chinese medicine hospital of Nanjing Lishui District Ethics Committee approved this study. Because it was retrospective in nature, it did not require written informed consent.

Schematic diagram of the patient selection. PTC, papillary thyroid carcinoma.

There were two ultrasonic devices used: the Philips Q5 (both Healthcare, Eindhoven, Netherlands) and the GE LOGIC E20 (GE Medical Systems, American General) (L12-5 linear array probe, frequency: 10–14 MHz).

To acquire longitudinal and transverse images of the thyroid nodules, continuous longitudinal and transverse scanning was done while the patients were supine. Blood flow in and around the nodule, strain elastic grading of the nodule, calcification, and CLNM were all visible on the coexisting diagram, which also included the nodule diameter, location, echo, and boundary.

The cross-sectional image's position and size of the sampling frame were adjusted, and the strain elastic imaging mode was activated. With an ROI that was larger than the nodules (generally more than two times), the nodules were placed in the middle of the elastic imaging zone. Pressure was applied steadily (range 1–2 mm, 1–2 times/s) while the probe was perpendicular to the nodule. When the linear strain hint graph (green spring) suggested stability, the freeze key was pressed to get an elastic image; the ROI's color changed (green indicated soft; red indicated hard), and the nodule's hardness was determined based on elasticity. The elastic image was graded according to the following criteria: one point equals a nodular area that alternates between red, green, and blue; two points equal nodules that are partially red and partially green (mostly green, area > 90%); three points equal a nodule area that is primarily green, with surrounding tissues visible in red; four points equal a nodule area that is primarily red, with the red area > 90%; and five points equal a nodule area that is completely covered in red.

One week prior to surgery, thyroid US exams were conducted. US image segmentation was done manually. Using the ITK-SNAP program (http://www.itksnap.org), the ROIs were manually drawn on each image (Fig. 2). The grayscale images were used to create a sketch outline of the tumor regions in the elastography US images.

(A) Ultrasound conventional B-mode image of papillary thyroid carcinoma. (B) corresponding ultrasound elastography image, with the circle, labeled A indicating a lesion region and the circle labeled B indicating a reference area. (C) Corresponding image after region of interest (ROIs) segmentation step.

Radiomic features were extracted using PyRadiomics (https://github.com/Radiomics/pyradiomics). A total of 479 radiomic features were recovered from each ROI's elastography US images. Among those included were first-order Gray Level Co-occurrence Matrix (GLCM), Gray Level Run Length Matrix (GLRLM), Gray Level Size Zone Matrix (GLSZM), Gray Level Dependence Matrix (GLDM), and Neighbouring Gray Tone Difference Matrix (NGTDM) features, as well as features deduced from wavelet filter images containing first-order GLCM, GLRLM, GLSZM, GLDM, and NGTDM features.

The retrieved features were normalized using a standard scalar to reduce bias and overfitting in the study. The dataset was divided into training and validation cohorts. To make each characteristic substantially independent, the row spatial dimension of the feature matrix was reduced using the Pearson correlation coefficient (PCC). Every pair of features with a PCC of more than 0.80 was deemed redundant.

After PCC, recursive feature elimination (RFE) for feature selection was applied to the whole dataset using the Scikit-learn python module24 to choose representative features for the training cohort. During the RFE procedure, the following parameters were taken into consideration (cross-validation was set to stratifiedkfold with the number of splits being 10, the random state was set to 101, minimum features to select was set to 3, and accuracy was employed for the scoring.

The Support Vector Machine with the linear kernel (SVM_L), Support Vector Machine with radial basis function kernel (SVM_RBF), LogisticRegression (LR), Naïve Bayes (NB), K-nearest Neighbors (KNN), and Linear Discriminant Analysis (LDA) classifiers were used to build the prediction models using the RFE’s key features. All six algorithms were implemented using the Scikit-learn machine learning library24

The same feature sets were chosen and fed into the model during the validation process. Standard clinical statistics like the area under the curve (AUC), sensitivity, specificity, negative predictive value (NPV), positive predictive value (PPV), and accuracy (ACC) were used to evaluate the model's performance on the training and validation datasets.

Python (version 3.7, https://www.python.org/ Accessed 8 July 2021) and IBM SPSS Statistics (Monk Ar, New York, New York State, USA.) for Windows version 26.0 were used for statistical analyses. Pearson's chi-square and Fisher's exact tests were used to compare the differences in categorical characteristics. The independent sample t-test was used for continuous factors with normal distribution, whereas the Mann–Whitney U test was used for continuous factors without normal distribution.

A two‐sided P < 0.05 indicated statistically significant differences. PyRadiomics (version 2.2.0, https://github.com/Radiomics/pyradiomics Accessed 10 August 2021) and scikit‐learn version 1.224 were used to extract radiomic features and build the prediction models. Each prediction model's AUC, sensitivity, specificity, ACC, NPV, and PPV were calculated.

Medcalc Statistical Software was used to calculate the six models’ AUCs and evaluate the predictions. The DeLong method was used to compare the AUCs of the six machine learning classifiers. To create calibration curves, the sci-kit-learn version 1.224 was used. R software (version 3.6.1, https://www.r-project.org) was used to perform the decision curve analysis.

The study was conducted in accordance with the Declaration of Helsinki and approved by the Jiangsu University-Affiliated People’s Hospital and traditional Chinese medicine hospital of Nanjing Lishui District Ethics Committee.

Patient consent was waived by the Jiangsu University-Affiliated People’s Hospital and traditional Chinese medicine hospital of Nanjing Lishui District ethics committee due to the retrospective nature of the study.

There were 138 PTC patients in all, 87 of whom were women (mean age, 41.81 ± 11.23 [range, 25–57] years), and 51 of them were men (mean age, 43.82 ± 12.18 [range, 28–65] years). In Table 1, 138 patients' clinical information and imaging comparisons between the training and validation groups are displayed.

The relationship between the BRAFV600E mutation and ultrasonic imaging characteristics and the predictive performance of the machine learning algorithms is shown in Tables 2, 3.

After PCC and RFE with stratified tenfold cross-validation, 27 radiomic features were chosen in the training cohort (Figs. 3 and 4).

Boxplot of Selected features after RFE. Features were reduced to twenty-seven features during training of the machine learning models.

Recursive feature elimination (RFE) with tenfold cross‐validation; number of features selected vs. cross‐validation score.

The following features were chosen to develop the predictive models for BRAFV600E based on six machine learning algorithms (Table 4).

In the training cohort, AUCs for KNN, LDA, LR, NB, SVM_L, and SVM_ RBF were 0.96 (95% confidence interval [CI]: 0.89–0.99), 1.00 (95% CI 0.96–1.00), 1.00 (95% CI 0.96–1.00), 0.96 (95% CI 0.89–0.99), 1.00 (95% CI 0.96–0.1.00), and 1.00 (95% CI 0.96–1.00), respectively (Table 3 and Fig. 5). The SVM_RBF, SVM_L, LDA, and LR models performed best. KNN and NB followed. All machine learning models performed well.

The mixed ROC curves of the six machine learning models in the training cohort. ROC: receiver operating characteristic; KNN: K-nearest neighbor; LDA: linear discriminant analysis; LR: logistic regression; NB: Naïve Bayes;SVM_L: support vector classifier with the linear kernel; SVM_RBF: support vector classifier with the radial basis function.

In the validation cohort, AUCs for NB, KNN, LDA, LR, SVM_L, and SVM_RBF were 0.80 (95% confidence interval [CI]: 0.65–0.91), 0.87 (95% CI 0.73–0.95), 0.91(95% CI 0.79–0.98), 0.92 (95% CI 0.80–0.98), 0.93 (95% CI 0.80–0.98), and 0.98 (95% CI 0.88–1.00), respectively. The SVM_RBF model performed the best in the validation cohort, followed by the LR, SVM_L, LDA, KNN, and NB models, in that order (Fig. 6). All machine learning-based models performed well. The SVM_RBF model's sensitivity, specificity, PPV, and NPV were 0.95, 0.90, 0.91, and 0.95, respectively (Table 3).

The mixed ROC curves of the six machine learning models in the validation cohort. ROC: receiver operating characteristic. KNN: K-nearest neighbor; LDA: linear discriminant analysis; LR: logistic regression; NB: Naïve Bayes; SVM_L: support vector classifier with the linear kernel; SVM_RBF: support vector classifier with the radial basis function.

Furthermore, the DCA was used to evaluate the clinical utility of these models (Fig. 7). The calibration curve was used for evaluating the probability accuracy of the machine learning models in predicting an individual outcome event in the future (Fig. 8).

Decision curve for predictive models based on machine learning models in the training cohort (A) and the validation cohort (B). KNN K-nearest neighbor; LDA Linear discriminant analysis;LR Logistic regression; NB Naïve Bayes;SVM_L Support vector classifier with the linear kernel; SVM_RBF Support vector classifier with the radial basis function.

Calibration curves of the machine learning (A) K-nearest neighbor (B) Linear discriminant analysis (C) LR Logistic regression (D) Naïve Bayes (E) Support vector classifier with the linear kernel; (F) SVM_RBF Support vector classifier with the radial basis function.

The BRAF V600E mutation has become a distinctive and important molecular marker in the management of PTC due to its substantial relationship with aggressive clinical pathological outcome, serious molecular derangements, and selection of treatment methods in PTC.

Real-time tissue strain elastography helps assess the anatomical structure and biological traits of PTC because it represents the relative hardness of the lesion and its surrounding tissues, whereas the hardness of PTC tissues is closely related to its internal pathological structure. Bojunga et al.25 showed that elastography US could effectively distinguish between benign and malignant thyroid nodules. Recent research has demonstrated that varied cell arrangements and cell compositions result in varying tissue elastography indices26,27; the harder elastography US imaging is for PTC, the higher its malignancy. This gives the US diagnosis of BRAFV600E mutation-positive and negative PTC a new viewpoint. Using US images, it was determined that patients with and without BRAFV600E mutations had significantly different elastic moduli or levels of hardness. Elastography might more accurately describe the pathological features of tissues and directly and quantitatively indicate the absolute hardness of BRAFV600E mutation-positive and -negative PTC. This study used machine learning methods to extract imaging omics features behind PTC strain elastic US images, and then used these features to build models to predict BRAF gene status, which is more objective than the semi-quantitative method of dividing PTC strain elastic US images into 1–5 points based on naked eye observation.

In a previous study28, we developed three radiomic models to predict BRAFV600E mutation in PTC patients: a radiomic model based on grayscale US radiomic features, a radiomic model based on elastography radiomic features, and a third model based on combined gray US radiomic and elastography radiomic features using logistic regression classifier and the least absolute shrinkage selection operator (LASSO) feature selection algorithm. The gray scale US radiomic model did not perform well in the validation cohort (ACC: 0.67, AUC: 0.73), but the elastography model (ACC: 0.88, AUC: 0.93) and combined model (ACC: 0.91, AUC: 0.94), did perform well. Considering that elastic US images superimpose information on the hardness of the mass on the basis of grayscale US images, image feature extraction not only extracts the hardness characteristics of the mass, but also extracts rich features of the internal structure of the mass. So we decided to focus on elastography radiomics features in this current study to explore the performance of multiple machine learning classifiers or algorithms as well as other feature selection methods.

Machine learning algorithms benefit from learning from input data and automatically recognizing patterns and trends in that data automatically. There have been numerous studies29,30,31 on the use of machine learning to differentiate between benign and malignant thyroid nodules. However, there have been only a few studies using machine learning models to predict BRAFV600E mutation in PTC patients. Additionally, there hasn't been a comparison of the effectiveness of various machine learning algorithms for the prediction of BRAFV600E mutation in PTC patients based on US elastography radiomic features. In the current study, we constructed six machine-learning models to distinguish PTC patients with BRAFV600E gene mutations from PTC patients without BRAFV600E gene mutations using preoperative US elastography radiomic data. There were three noteworthy discoveries. First, based on preoperative US elastography radiomic features, the six machine learning models could differentiate PTC patients with BRAFV600E from non-BRAFV600E PTC patients. Second, when the six machine learning models were compared, SVM_RBF had the best prediction performance. Third the machine learning algorithms' diagnostic performance was based on 27 radiomic features.

In order to determine whether US-based radiomics could perhaps assess the occurrence of BRAFV600E mutations among patients with PTC, Yoon et al.10 established a radiomic score using a dataset of 527 patients who had undergone surgical treatment for PTC and who had all undergone BRAFV600E mutation analysis on surgical specimen.They reported that radiomics features extracted from US have limited value as a non-invasive biomarker for predicting the presence of BRAFV600E mutation status of PTC, with a c-statistic value (equivalent to AUC) of 0.63 in the validation cohort and 0.72 in the training cohort. In comparison to this study, our study reported AUCs range of (0.96–1.00) and (0.87–0.98) in the training and validation cohorts, respectively, and ACCs range of (0.89–0.98) and (0.81–0.93) in the training and validation cohorts, respectively for the six machine learning algorithms employed in this study.

The higher AUCs and ACCs in the current study could be attributed to the US elastography radiomic features used, which provided more information than the grayscale US. Also in this study, machine learning algorithms were used in conjunction with radiomics, and several machine learning classifiers were used to build the models. To the best of our knowledge, this is the first study to develop models based on different machine-learning algorithms to predict BRAFV600E mutations in PTC patients.

Comparing the current study results to our previous study28, the ACCs of SVM_L (0.88), LDA (0.88), and LR (0.88) were the same as the ACC of the logistic regression classifier (0.88) utilized in our previous study. However, the KNN (0.81) and NB (0.81) ACCs values were lower than the LR (0.88) method utilized in our previous study.

The AUCs of the KNN (0.87), LDA (0.91), LR (0.92), and NB (0.80) classifiers were all lower than the logistic regression classifier (0.93) utilized in our previous study. The SVM_L AUC (0.93) score, on the other hand, was the same as the logistic regression algorithm (0.93) utilized in our previous work. When compared to the logistic regression classifier (ACC: 0.88 and AUC: 0.93) employed in our previous investigation, the SVM_RBF had higher ACC (0.93) and AUC (0.98) values. The disparity in performance could be attributed to the various feature preprocessing techniques and feature selection strategies used in the current investigation.

On US examination, Kabaker et al.32 discovered that vertical and horizontal diameter ratios greater than one, as well as low echo, were all associated with the BRAFV600E mutation. Similarly, Hahn et al.33 discovered that vertical and horizontal diameter ratios greater than one were linked to BRAFV600E gene mutations. Consistent with these studies, we discovered in this current study that there was a significant difference in echogenicity, vertical and horizontal diameter ratios, and elasticity between PTC patients with BRAFV600E and PTC patients without BRAFV600E.

The calibration curve of the prediction model is an essential metric for evaluating the probability accuracy of a disease risk model in predicting an individual outcome event in the future. A high degree of calibration shows that the prediction model is accurate, whereas a low degree of calibration indicates that the model may exaggerate or underestimate the risk of illness. The blue line reflects the performance of the machine learning algorithms, while the diagonal dotted line represents an ideal prediction (Fig. 8). A closer match to the diagonal dotted line suggests a better prediction. When the calibration curves were near the diagonal line, SVM_RBF, SVM_L, and NB algorithms demonstrated good agreement between the real status of BRAFV600E gene mutation and the predicted probability.

Furthermore, the DCA was used to evaluate the clinical utility of these models (Fig. 7). Assuming that all patients do not have BRAFV600E gene mutation, the solid black line (negative line) indicates that when no patient accepts intervention or treatment, the net benefit is zero. On the contrary, the solid grey line (positive line) indicates the net benefits when all patients have BRAFV600E and receive treatments or interventions. According to the incidence of BRAFV600E among patients with PTC, the reasonable range of thresholds was set from 0.3 to 0.99. In the entire range, all machine learning-based algorithms showed higher net benefits than the two extreme lines (negative line and positive line). In almost the entire threshold probability range, the SVM_RBF algorithm had the highest net benefit in both the training and validation cohorts (Fig. 7).

There are some limitations of the study, in the construction of the models to predict BRAFV600E in PTC patients, the gene in healthy individuals was not analyzed, and the focus was on evaluating the BRAFV600E mutation in PTC patients, which could have resulted in a selection bias. Also, this was a small sample retrospective study conducted at two institutions; thus, a selection bias may exist. In the future, we aim to conduct a multicenter study with a larger sample size.

Finally, our study found that machine learning-based US elastography radiomic models performed well in predicting the potential of BRAFV600E in PTC patients, which can assist physicians in identifying the risk of BRAFV600E in PTC patients. SVM_RBF achieved the greatest prediction performance of the six machine learning models tested.

The original contributions presented in the study are included in the article. Further inquiries can be directed to the corresponding authors.

Accuracy

Artificial intelligence

Area under the curve

Atypical lesions or follicular lesions of unknown significance

Cervical lymph node metastasis

Chinese Society of Clinical Oncology

Decision curve analysis

Extrathyroidal extension

Food and drugs authority

Fine needle aspiration

Gray Level Co-occurrence Matrix

Gray Level Dependence Matrix

Gray Level Run Length Matrix

Gray Level Size Zone Matrix

Knearest neigbor

Least absolute shrinkage selection operator

Linear discriminant analysis

Logistic regression

National Comprehensive Cancer Network

Neighbouring Gray Tone Difference Matrix

Naïve beyes

Negative predictive value

Pearson correlation coefficient

Positive predictive value

Papillary thyroid carcinoma

Recursive feature elimination

Receiver operating characteristics

Sensitivity

Specificity

Suspected malignant tumors

Support vector machine with radial basis function

Support vector machine with linear kernel

Ultrasound

Czarniecka, A., Oczko-Wojciechowska, M. & Barczyński, M. BRAF V600E mutation in prognostication of papillary thyroid cancer (PTC) recurrence. Gland. Surg. 5, 495–505. https://doi.org/10.21037/gs.2016.09.09 (2016).

Article PubMed PubMed Central Google Scholar

Yan, C., Huang, M., Li, X., Wang, T. & Ling, R. Relationship between BRAFV600E and clinical features in papillary thyroid carcinoma. Endocr. Connect. 8, 988–996. https://doi.org/10.1530/EC-19-0246 (2019).

Article CAS PubMed PubMed Central Google Scholar

Yarchoan, M., LiVolsi, V. A. & Brose, M. S. BRAF mutation and thyroid cancer recurrence. J. Clin. Oncol. 33(1), 7–8. https://doi.org/10.1200/JCO.2014.59.3657 (2015).

Article PubMed Google Scholar

Rashid, F. A., Munkhdelger, J., Fukuoka, J. & Bychkov, A. Prevalence of BRAFV600E mutation in asian series of papillary thyroid carcinoma—A contemporary systematic review. Gland. Surg. 9, 1878–1900. https://doi.org/10.21037/gs-20-430 (2020).

Article PubMed PubMed Central Google Scholar

Sun, J. et al. BRAF V600E, and TERT promoter mutations in papillary thyroid carcinoma in Chinese patients. PLoS ONE 11, e153319. https://doi.org/10.1371/journal.pone.0153319 (2016).

Article Google Scholar

Jin, L. et al. BRAF mutation analysis in fine needle aspiration (FNA) cytology of the thyroid. Diagn. Mol. Pathol. 15(3), 136–143. https://doi.org/10.1097/01.pdm.0000213461.53021.84 (2006).

Article CAS PubMed Google Scholar

Xing, M. et al. BRAF mutation predicts a poorer clinical prognosis for papillary thyroid cancer. J. Clin. Endocrinol. Metab 90(12), 6373–6379. https://doi.org/10.1210/jc.2005-0987 (2005).

Article CAS PubMed Google Scholar

Jarząb, B. et al. Diagnostics and treatment of thyroid carcinoma. Endokrynol. Polska 67(1), 74–107. https://doi.org/10.5603/EP.2016.0011 (2016).

Article Google Scholar

Bentz, B. G., Miller, B. T., Holden, J. A., Rowe, L. R. & Bentz, J. S. B-RAF V600E mutational analysis of fine needle aspirates correlates with diagnosis of thyroid nodules. Otolaryngology–Head Neck Surg. Off. J. Am. Acad. Otolaryngol.-Head Neck Surg. 140, 709–714. https://doi.org/10.1016/j.otohns.2009.01.007 (2009).

Article Google Scholar

Lee, J. Y. et al. 2020 imaging guidelines for thyroid nodules and differentiated thyroid cancer: Korean society of thyroid radiology. Korean J. Radiol. 22(5), 840–860. https://doi.org/10.3348/kjr.2020.0578 (2021).

Article PubMed PubMed Central Google Scholar

Yoon, J. et al. Artificial intelligence to predict the BRAFV600E mutation in patients with thyroid cancer. PLoS ONE 15, e24280–e24286. https://doi.org/10.1371/journal.pone.0242806 (2020).

Article CAS Google Scholar

Yoon, J. H. et al. Radiomics in predicting mutation status for thyroid cancer: A preliminary study using radiomics features for predicting BRAFV600E mutations in papillary thyroid carcinoma. PLoS ONE 15, e228968. https://doi.org/10.1371/journal.pone.0228968 (2020).

Article CAS Google Scholar

Kim, T. H. et al. The association of the BRAF(V600E) mutation with prognostic factors and poor clinical outcome in papillary thyroid cancer: A meta-analysis. Cancer 118(7), 1764–1773. https://doi.org/10.1002/cncr.26500 (2012).

Article CAS PubMed Google Scholar

Itoh, A. et al. Breast disease: clinical application of US elastography for diagnosis. Radiology 239(2), 341–350. https://doi.org/10.1148/radiol.2391041676 (2006).

Article PubMed Google Scholar

Lee, S. H. et al. Practice guideline for the performance of breast ultrasound elastography. Ultrasonography 33(1), 3–10. https://doi.org/10.14366/usg.13012 (2014).

Article PubMed Google Scholar

Omar, K. et al. Breast cancer in Egypt: a review of disease presentation and detection strategies. EMHJ-Eastern Mediterr. Heal. J. 2003 (2003).

Athamnah, S. I., Oglat, A. A. & Firas, F. F. Diagnostice breast elastography estimation from doppler imaging using central difference (CD) and least-squares (LS) algorithms. Biomed. Signal Process. Control 68, 102667 (2021).

Article Google Scholar

Yang, L. et al. Can CT-based radiomics signature predict KRAS/NRAS/BRAF mutations in colorectal cancer?. Eur. Radiol. 28(5), 2058–2067. https://doi.org/10.1007/s00330-017-5146-8 (2018).

Article PubMed Google Scholar

Gevaert, O. et al. Predictive radiogenomics modeling of EGFR mutation status in lung cancer. Sci. Rep. 7, 41674. https://doi.org/10.1038/srep41674 (2017).

Article ADS CAS PubMed PubMed Central Google Scholar

West, E., Mutasa, S., Zhu, Z. & Ha, R. Global trend in artificial intelligence-basedpublications in radiology from 2000 to 2018. AJR Am. J. Roentgenol. 213(6), 1204–1206. https://doi.org/10.2214/AJR.19.21346 (2019).

Article PubMed Google Scholar

Pesapane, F., Codari, M. & Sardanelli, F. Artificial intelligence in medical imaging: Threat or opportunity? Radiologists again at the forefront of innovation inmedicine. Eur. Radiol. Exp. 2(1), 35. https://doi.org/10.1186/s41747-018-0061-6 (2018).

Article PubMed PubMed Central Google Scholar

Zhao, C. K. et al. A comparative analysis of two machine learning-based diagnostic patterns with ACR TiRADS for thyroid nodules: Diagnostic performance and unnecessary biopsy rate. Thyroid 31(3), 470–481. https://doi.org/10.1089/thy.2020.0305 (2020).

Article CAS PubMed Google Scholar

Gillies, R. J., Kinahan, P. E. & Hricak, H. Radiomics: Images are more than pictures, they are data. Radiology 278(2), 563–577. https://doi.org/10.1148/radiol.2015151169 (2016).

Article PubMed Google Scholar

Pedregosa, F. et al. Scikit-learn: Machine learning in python. J. Mach. Learn. Res 12, 2825–2830 (2011).

MathSciNet MATH Google Scholar

Bojunga, J. & Mondorf, A. Schilddrüsen-Elastografie [Thyroid Elastography]. Laryngorhinootolog 98, 150–156. https://doi.org/10.1055/a-0790-0885 (2019).

Article Google Scholar

Yi, L., Qiong, W., Yan, W., Youben, F. & Bing, H. Correlation between ultrasound elastography and histologic characteristics of papillary thyroid carcinoma. Sci. Rep. 7, 45042. https://doi.org/10.1038/srep45042 (2017).

Article ADS CAS PubMed PubMed Central Google Scholar

Li, N. et al. Nomogram including elastography for prediction of contralateral central lymph node metastasis in solitary papillary thyroid carcinoma preoperatively. Cancer Manag. Res. 12, 10789–10797. https://doi.org/10.2147/CMAR.S278382 (2020).

Article PubMed PubMed Central Google Scholar

Wang, Y. G. et al. Radiomic model for determining the value of elasticity and grayscale ultrasound diagnoses for predicting BRAFV600E mutations in papillary thyroid carcinoma. Front. Endocrinol. 13, 872153. https://doi.org/10.3389/fendo.2022.872153 (2022).

Article Google Scholar

Choi, Y. J. et al. A computer-aided diagnosis system using artificial intelligence for the diagnosis and characterization of thyroid nodules on ultrasound: Initial clinical assessment. Thyroid Res. 27(4), 546–552. https://doi.org/10.1089/thy.2016.0372 (2017).

Article Google Scholar

Sollini, M., Cozzi, L., Chiti, A. & Kirienko, M. Texture analysis and machine learning to characterize suspected thyroid nodules and differentiated thyroid cancer: Where do we stand?. Eur. J. Radiol. 99, 1–8. https://doi.org/10.1016/j.ejrad.2017.12.004 (2018).

Article PubMed Google Scholar

Zhang, B. et al. Machine learning-assisted system for thyroid nodule diagnosis. Thyroid Res. 29(6), 858–867. https://doi.org/10.1089/thy.2018.0380 (2019).

Article Google Scholar

Kabaker, A. S. et al. Suspicious ultrasound characteristics predict BRAFV600E-positive papillary thyroid carcinoma. Thyroid 22, 585–589. https://doi.org/10.1089/thy.2011.0274 (2012).

Article PubMed PubMed Central Google Scholar

Hahn, S. Y. et al. Ultrasound and clinicopathological features of pa-pillary thyroid carcinomas with BRAF and TERT promoter mutations. Oncotarget 8, 108946–108957. https://doi.org/10.18632/oncotarget.22430 (2017).

Article PubMed PubMed Central Google Scholar

Download references

This study was financially supported by the National Natural Science Foundation of China (Project No.: 81971629). Jiangsu Provincial Health Commission research project (Z2021071), 2023 Medical Education Collaborative Innovation Fund of Jiangsu University (JDYY2023015).

Ultrasound Medical Laboratory, Department of Ultrasound, Affiliated People’s Hospital of Jiangsu University, Zhenjiang, 212002, China

Enock Adjei Agyekum, Fei-Ju Xu, Yong-zhen Ren, Xian Wang, Jenny Olalia Taupa & Xiao-qin Qian

School of Medicine, Jiangsu University, Zhenjiang, 212013, China

Enock Adjei Agyekum, Yong-zhen Ren, Kevoyne Hakeem Chambers & Jenny Olalia Taupa

Department of Ultrasound, Traditional Chinese Medicine Hospital of Nanjing Lishui District, Nanjing, China

Yu-guo Wang

School of Public Health, Kwame Nkrumah University of Science and Technology, Kumasi, Ghana

Debora Akortia

You can also search for this author in PubMed Google Scholar

You can also search for this author in PubMed Google Scholar

You can also search for this author in PubMed Google Scholar

You can also search for this author in PubMed Google Scholar

You can also search for this author in PubMed Google Scholar

You can also search for this author in PubMed Google Scholar

You can also search for this author in PubMed Google Scholar

You can also search for this author in PubMed Google Scholar

You can also search for this author in PubMed Google Scholar

Conceptualization, E.A.A. and X.Q.; methodology, E.A.A.; software, E.A.A. and D.A.; validation, Y.-G.W., J.O.T., and Y.-Z.R.; formal analysis, E.A.A.; investigation, E.A.A. and F.-J.X.; resources, E.A.A., and Y.-G.W.; data curation, E.A.A., X.W., and X.W.; writing—original draft preparation, E.A.A.; writing—review and editing, E.A.A., X.Q., and K.H.C.; visualization, E.A.A. and K.H.C.; supervision, X.Q., project administration, E.A.A., and X.Q; funding acquisition, X.Q. All authors have read and agreed to the published version of the manuscript.

Correspondence to Xiao-qin Qian.

The authors declare no competing interests.

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.

Reprints and Permissions

Agyekum, E.A., Wang, Yg., Xu, FJ. et al. Predicting BRAFV600E mutations in papillary thyroid carcinoma using six machine learning algorithms based on ultrasound elastography. Sci Rep 13, 12604 (2023). https://doi.org/10.1038/s41598-023-39747-6

Download citation

Received: 31 May 2023

Accepted: 30 July 2023

Published: 03 August 2023

DOI: https://doi.org/10.1038/s41598-023-39747-6

Anyone you share the following link with will be able to read this content:

Sorry, a shareable link is not currently available for this article.

Provided by the Springer Nature SharedIt content-sharing initiative

By submitting a comment you agree to abide by our Terms and Community Guidelines. If you find something abusive or that does not comply with our terms or guidelines please flag it as inappropriate.