Real-time prediction of mediastinal lymph node malignancy by endobronchial ultrasound

Machine learning approaches in the interpretation of endobronchial ultrasound images: a comparative analysis

BACKGROUND

This study explores the application of machine learning (ML) in analyzing endobronchial ultrasound (EBUS) images for the detection of lymph node (LN) malignancy, aiming to augment diagnostic accuracy and efficiency. We investigated whether ML could outperform conventional classification systems in identifying malignant involvement of LNs, based on eight established sonographic features.

METHODS

Retrospective data from two tertiary care hospital bronchoscopy units were utilized, encompassing healthcare reports of patients who had undergone EBUS between January 2017 and March 2023. The ML model was trained and tested using MATLAB, with 80% of the data allocated for training/validation, and 20% for testing. Performance was evaluated based on validation and testing accuracy, and receiver operating characteristic curves with comparing trained models and existing classification rules.

RESULTS

The study analyzed 992 LNs, with 42.3% malignancy prevalence. Malignant LNs showed characteristic features such as larger size and distinct margins. The fine tuned models achieved testing accuracies of 95.9% and 96.4% for fine Gaussian SVM and KNN, respectively. Corresponding AUROC's were 0.955 and 0.963, outperforming other similar studies and conventional analyses.

CONCLUSION

Fine tuned ML applications like SVM and KNN, can significantly enhance the analysis of EBUS images, improving diagnostic accuracy.

Guidelines for endobronchial ultrasound-transbronchial needle aspiration (EBUS-TBNA): Joint Indian Chest Society (ICS)/Indian Association for Bronchology (IAB) recommendations

Over the past decade, endobronchial ultrasound-guided transbronchial needle aspiration (EBUS-TBNA) has become an indispensable tool in the diagnostic armamentarium of the pulmonologist. As the expertise with EBUS-TBNA has evolved and several innovations have occurred, the indications for its use have expanded. However, several aspects of EBUS-TBNA are still not standardized. Hence, evidence-based guidelines are needed to optimize the diagnostic yield and safety of EBUS-TBNA. For this purpose, a working group of experts from India was constituted. A detailed and systematic search was performed to extract relevant literature pertaining to various aspects of EBUS-TBNA. The modified GRADE system was used for evaluating the level of evidence and assigning the strength of recommendations. The final recommendations were framed with the consensus of the working group after several rounds of online discussions and a two-day in-person meeting. These guidelines provide evidence-based recommendations encompassing indications of EBUS-TBNA, pre-procedure evaluation, sedation and anesthesia, technical and procedural aspects, sample processing, EBUS-TBNA in special situations, and training for EBUS-TBNA.

A valuable endobronchial ultrasound scoring system predicting malignant lymph nodes

Background

This study aims to determine the sonographic criteria of lymph nodes to predict malignancy with endobronchial ultrasound.

Methods

A total of 1,987 lymph nodes of 967 patients (666 males, 301 females; mean age: 62.1±11.9 years; range, 21 to 90 years) between May 2016 and July 2020 were retrospectively analyzed. The endobronchial ultrasound images of lymph nodes were evaluated according to the following criteria: size (short axis >1 cm), shape (round or oval), margin (distinct or indistinct), coagulation necrosis sign (present or absent), central hilar structure (present or absent) and echogenicity (homogeneous or heterogeneous). A scoring system was developed for predicting malignancy.

Results

A total of 765 (38.5%) of the lymph nodes were malignant. In the univariate analysis, size >1 cm, round shape, distinct margin, absence of central hilar structure, presence of coagulation necrosis sign, and heterogeneity were significant predictors of malignancy (p<0.001 for all). In the multivariate analysis, the main independent predictors were heterogeneity and presence of coagulation necrosis sign (odds ratio=5.9, 95% confidence interval: 4.2-8.2 vs. odds ratio=3.1 95% confidence interval: 2.2-4.5, respectively). A cut-off value for endobronchial ultrasound score of ≥4 increased the malignancy risk 30 times with a sensitivity of 84.7%, and specificity of 84.5%.

Conclusion

Our study results show that endobronchial ultrasound scoring system with six criteria has a high sensitivity and specificity for predicting malignant lymph nodes.

Prediction of malignancy in mediastinal lymph nodes during endobronchial ultrasound: A comparative validation study

Background

In this study, we aimed to compare the diagnostic performances of three existing prediction tools in visually identifying a malignant lymph node.

Methods

Between April 2016 and January 2021, a total of 827 lymph nodes of 259 patients (211 males, 48 females; mean age: 61.1±7.2 years; range, 41 to 79 years) who underwent endobronchial ultrasound procedure for diagnosis and/or staging of lung cancer and diagnosis of mediastinal lymphadenopathy of unknown origin were retrospectively analyzed. This external validation study was designed to compare the diagnostic yields of the prediction tools developed by Shafiek et al., Alici et al., and Canada Lymph Node Score (CLNS). Endobronchial ultrasoundguided transbronchial needle aspiration results and predictions were compared to gold-standard tool.

Results

Overall, endobronchial ultrasound-guided transbronchial needle aspiration had a sensitivity, specificity, positive and negative predictive value, and accuracy of 95.6%, 100%, 100%, 97.6%, and 98.4%, respectively. Diagnostic performances of proposed tools were quite remarkable. Among them, Alici algorithm had a higher sensitivity and negative predictive value, which were matched by excellent specificity and positive predictive value offered by CLNS ≥3 and Shafiek tool. The area under the curve value of CLNS ≥3 was higher than Shafiek tool and CLNS ≥2.

Conclusion

Conventional prediction tools relying on simple real-time sonographic features were found to be consistent by the means of diagnostic performance in this external validation dataset. Despite being inferior to cytology, their superior performance was proven with defined individual strengths and weaknesses.

Predicting Malignant Lymph Nodes Using a Novel Scoring System Based on Multi-Endobronchial Ultrasound Features

Endobronchial ultrasound (EBUS) features with B-, power/color Doppler, and elastography modes help differentiate between benign and malignant lymph nodes (MLNs) during transbronchial needle aspiration (TBNA); however, only few studies have assessed them simultaneously. We evaluated the diagnostic accuracy of each EBUS feature and aimed to establish a scoring system to predict MLNs. EBUS features of consecutive patients and final diagnosis per lymph node (LN) were examined retrospectively. In total, 594 LNs from 301 patients were analyzed. Univariable analyses revealed that EBUS features, except for round shape, could differentiate MLNs from benign LNs. Multivariable analysis revealed that short axis (>1 cm), heterogeneous echogenicity, absence of central hilar structure, presence of coagulation necrosis sign, and blue-dominant elastographic images were independent predictors of MLNs. At three or more EBUS features predicting MLNs, our scoring system had high sensitivity (77.9%) and specificity (91.8%). The area under the receiver operating curve (AUC) was 0.894 (95% confidence interval (CI): 0.868−0.920), which was higher than that of B-mode features alone (AUC: 0.840 (95% CI: 0.807−0.873)). The novel scoring system could predict MLNs more accurately than B-mode features alone. Multi-EBUS features may increase EBUS-TBNA efficiency for LN evaluation.

Clinical Validation of the Canada Lymph Node Score for Endobronchial Ultrasound

BACKGROUND

The Canada Lymph Node Score (CLNS) uses four sonographic criteria to predict the risk of malignancy in lymph nodes during endobronchial ultrasound (EBUS). CLNS may play a role in identifying targets for biopsy or re-biopsy during invasive mediastinal staging for lung cancer. However, CLNS has not yet been prospectively validated in routine clinical practice.

METHODS

CLNS scores for each lymph node biopsied during EBUS were prospectively captured for one year (2019). CLNS and the presence of malignancy in each node were compared. Univariate binary logistic regression was completed for each ultrasonographic feature, as well as a multivariate logistic regression model.

RESULTS

CLNS and diagnostic pathology results were available for 367 lymph nodes. Incidence of malignancy increased with higher scores. Scores ≥3 were significantly associated with malignancy (specificity 84.4%, positive likelihood ratio 4.0). Area under the curve was 0.76, indicating a good ability of the model to predict presence or absence of malignancy. Nodes scoring <2 and negative on CT and PET were malignant in 10.1%.

CONCLUSIONS

CLNS correlates with the presence or absence of malignancy in thoracic lymph nodes, and may serve as an adjunct to currently available methods of invasive and non-invasive mediastinal staging. CLNS may be most helpful to select which non-diagnostic nodes require re-biopsy. There is a significant risk of a false negative result even with a score of 0, and using a combination of low CLNS and negative conventional radiology to obviate the need for any initial biopsy remains to be studied in prospective trials.

An Artificial Intelligence Algorithm to Predict Nodal Metastasis in Lung Cancer

BACKGROUND

Endobronchial Ultrasound (EBUS) features have high accuracy for predicting lymph node (LN) malignancy. However, their clinical application remains limited due to high operator dependency. We hypothesized that an Artificial Intelligence algorithm (NeuralSeg) is capable of accurately identifying and predicting LN malignancy based on EBUS images.

METHODS

In the derivation phase, EBUS images were segmented twice by an endosonographer and used as controls in 5-fold cross-validation training of NeuralSeg. In the validation phase, the algorithm was tested on new images it had not seen before. Logistic regression and receiver operator characteristic curves were used to determine NeuralSeg's capability of discrimination between benign and malignant LNs, using pathologic specimens as gold standard.

RESULTS

In total, 298 LNs from 140 patients were used for derivation and 108 LNs from 47 patients for validation. In the derivation cohort, NeuralSeg was able to predict malignant LNs with an accuracy of 73.8% (95% CI: 68.4% to 78.7%). In the validation cohort, NeuralSeg had an accuracy of 72.9% (95% CI: 63.5% to 81.0%), a specificity of 90.8% (95% CI: 81.9% to 96.2%) and negative predictive value (NPV) of 75.9% (95% CI: 71.5% to 79.9%). NeuralSeg showed higher diagnostic discrimination during validation compared to derivation (c-statistic= 0.75 [95% CI: 0.65-0.85] vs c-statistic=0.63 [95% CI: 0.54-0.72]).

CONCLUSIONS

NeuralSeg is able to accurately rule out nodal metastasis and can possibly be used as an adjunct to EBUS when nodal biopsy is not possible or inconclusive. Future work to evaluate the algorithm in a clinical trial will be required.

Accuracy and Reproducibility of Endoscopic Ultrasound B-Mode Features for Observer-Based Lymph Nodal Malignancy Prediction

BACKGROUND

Endoscopic ultrasound routinely guides lymph node evaluation for the staging of a known or suspected lung cancer. Characteristics seen on B-mode imaging might help the observer decide on the lymph nodes of risk. The influence of nodal size on the predictivity of these characteristics and the agreement with which operators can combine these for malignancy risk prediction is to be determined.

OBJECTIVES

We evaluated (1) if prospectively scored individual B-mode ultrasound features predict malignancy when further divided by size and (2) assessed if observers were able to reproducibly agree on still lymph node image malignancy risk.

METHODS

Lymph nodes as visualized by EBUS were prospectively scored for B-mode characteristics. Still B-mode images were furthermore collected. After collection, a repeated scoring of a subset of lymph nodes was retrospectively performed (n = 11 observers).

RESULTS

Analysis of 490 lymph nodes revealed the short axis size is an objective measure for stratifying risk of malignancy (ROC area under the curve 0.78). With ≥8-mm size, 210/237 malignant lymph nodes were correctly identified (89% sensitivity, 46% specificity, 61% PPV, and 81% NPV). Secondary addition of B-mode features in <8-mm nodes had limited value. Retrospective analysis of intra- and interobserver scoring furthermore revealed significant disagreement.

CONCLUSIONS

Lymph nodes of ≥8-mm size and preferably even smaller should be aspirated regardless of other B-mode features. Observer disagreement in scoring both small and large lymph nodes suggests it is infeasible to include subjective features for stratification. Future research should focus on (integrating) other (semi)quantitative values for improving prediction.

Endobronchial Ultrasound Multimodal Imaging for the Diagnosis of Intrathoracic Lymph Nodes

BACKGROUND

Endobronchial ultrasound (EBUS) imaging is valuable in diagnosing intrathoracic lymph nodes (LNs), but there has been little analysis of multimodal imaging. This study aimed to comprehensively compare the diagnostic performance of single and multimodal combinations of EBUS imaging in differentiating benign and malignant intrathoracic LNs.

METHODS

Subjects from July 2018 to June 2019 were consecutively enrolled in the model group and July 2019 to August 2019 in the validation group. Sonographic features of three EBUS modes were analysed in the model group for the identification of malignant LNs from benign LNs. The validation group was used to verify the diagnostic efficiency of single and multimodal diagnostic methods built in the model group.

RESULTS

373 LNs (215 malignant and 158 benign) from 335 subjects and 138 LNs (79 malignant and 59 benign) from 116 subjects were analysed in the model and validation groups, respectively. For single mode, elastography had the best diagnostic value, followed by grayscale and Doppler. The corresponding accuracies in the validation group were 83.3%, 76.8%, and 71.0%, respectively. Grayscale with elastography had the best diagnostic efficiency of multimodal methods. When at least two of the three features (absence of central hilar structure, heterogeneity, and qualitative elastography score 4-5) were positive, the sensitivity, specificity, and accuracy in the validation group were 88.6%, 78.0%, and 84.1%, respectively.

CONCLUSIONS

In both model and validation groups, elastography performed the best in single EBUS modes, as well as grayscale combined with elastography in multimodal imaging. Elastography alone or combined with grayscale are feasible to help predict intrathoracic benign and malignant LNs.

Diagnostic value of endobronchial ultrasound image features: A specialized review

Endobronchial ultrasound-guided transbronchial needle aspiration (EBUS-TBNA) technology is important in the diagnosis of intrathoracic benign and malignant lymph nodes (LNs). With the development of EBUS imaging technology, its role in noninvasive diagnosis, as a supplement to pathology diagnosis, has been given increasing attention in recent years. Many studies have explored qualitative and quantitative methods for the three EBUS modes, as well as a variety of multimodal analysis methods, to find the optimal method for the noninvasive diagnosis using EBUS for LNs. Here, we review and comment on the research methods and predictive diagnostic value, discuss the existing problems, and look ahead to the future application of EBUS imaging.

Diagnostic utility of endobronchial ultrasound (EBUS) features in differentiating malignant and benign lymph nodes - A systematic review and meta-analysis

BACKGROUND

EBUS is being widely used today for echolocation of lymph nodes for FNAC. We present a systematic review and meta-analysis to assess the diagnostic accuracy of EBUS characteristics of lymph nodes in diagnosing malignancy.

METHODS

A systematic search of published literature was undertaken using databases like PubMed, Web of Science, Cochrane, Google Scholar and Researchgate. Those studies reporting any endobronchial ultrasonography features of malignant lymph nodes like size, margins, echogenicity, shape, central hilar structure (CHS), coagulation necrosis sign (CNS) or color power doppler index (CPDI) were included for review. Random effects model was used to calculate pooled sensitivity, specificity, positive and negative likelihood ratios (LR), and diagnostic odds ratio (DOR). The review protocol was registered with the International prospective register of systematic reviews (PROSPERO registration no. CRD42019117716).

RESULTS

992 articles were retrieved of which 542 articles were evaluated in detail and finally 29 articles met the inclusion criteria. All EBUS features except CPDI showed a statistically significant area under the SROC curve. CNS showed highest area under the SROC curve [0.81 (SE: 0.09)] with maximum pooled specificity [0.93, 95%CI: 0.92-0.94], maximum pooled LR+ [5.12, 95%CI: 2.56-10.2] and DOR [9.23, 95%CI 3.85-22.15]. Maximum sensitivity was seen for CHS 0.91 [95%CI: 0.90-0.92].

CONCLUSION

EBUS features have the potential to help in more precise location of a malignant lymph node thereby helping in increasing the diagnostic yield. However, high diagnostic accuracy of various EBUS features can currently only be said to supplement tissue diagnosis.

A novel online education module to teach clinicians how to correctly identify ultrasonographic features of mediastinal lymph nodes during endobronchial ultrasound

Background

Ultrasonographic features can be used to predict mediastinal lymph node malignancy during endobronchial ultrasonography. Despite the validity of using these features for this purpose, the features are not being widely used in clinical practice. This may be attributable to the absence of educational programs that teach clinicians how to identify the features. To address this knowledge gap, we developed an online educational module to teach clinicians how to correctly interpret ultrasonographic features.

Methods

The module was designed using corrective feedback and test-enhanced learning theories and distributed to clinicians in relevant specialties. The efficacy of the program was determined by comparing the percentages of correctly identified ultrasonographic features as each clinician progressed through the module. Participants were also asked to self-rate their confidence during the module. Analysis of variance was conducted, and a learning curve and descriptive statistics were generated.

Results

Twenty-two of the 29 participants (76%) completed the module. Analysis of variance indicated that the percentage of correctly identified features increased significantly as clinicians completed the module (p = 0.004); this finding is supported by the positive slope of the learning curve. Even though they initially reported some difficulty with identifying certain features, their confidence increased as they progressed through the module. When asked, 86% of participants reported that they found the educational module helpful and 90% reported that they would recommend it to others.

Conclusion

Participating clinicians were receptive to the interactive educational module. It enhances clinician skill and confidence in interpreting ultrasonographic features. The results of this study provide the foundation needed to test the validity of the educational module in clinical settings and to further explore clinician preferences for educational programs.

The Canada Lymph Node Score for prediction of malignancy in mediastinal lymph nodes during endobronchial ultrasound

OBJECTIVE(S)

During endobronchial ultrasound (EBUS) staging, ultrasonographic features can be used to predict mediastinal lymph node (LN) malignancy. We sought to develop the Canada Lymph Node Score a tool capable of predicting LN metastasis at the time of EBUS.

METHODS

Patients undergoing EBUS staging for lung and esophageal cancer were prospectively enrolled. Features were identified in real time by an endoscopist and video-recorded. Videos were sent to raters. Pathologic specimens from biopsies/surgical resections were used as the gold-standard reference test. Logistic regression, receiver operator characteristic curve, and Gwet's AC1 analyses were used to test the performance, discrimination, and inter-rater reliability, respectively.

RESULTS

In total, 300 LNs from 140 patients were analyzed by 12 endoscopists (raters) across 7 Canadian centers. Beta-coefficients from a multivariate regression model were used to create a 4-point score: short-axis diameter, margins, central hilar structure, and necrosis. The model showed good discriminatory power (c-statistic = 0.72 ± 0.04, 95% confidence interval [CI], 0.64-0.80; bias-corrected c-statistic: 0.66, 95% CI, 0.55-0.76). LNs scoring 3/4 or 4/4 had odds ratios of 15.17 (P < .0001) and 50.56 (P = .001) for predicting malignancy, respectively. Inter-rater reliability for a score ≥3 was 0.81 ± 0.02 (95% CI, 0.77-0.85).

CONCLUSIONS

The Canada Lymph Node Score is a 4-point score demonstrating excellent performance in identifying malignant LNs during EBUS. A cut-off of ≥3 may inform decision-making regarding biopsy, repeat biopsy, or mediastinoscopy if the initial results are inconclusive.

Endoscopic ultrasound: Elastographic lymph node evaluation

Different imaging techniques can bring different information which will contribute to the final diagnosis and further management of the patients. Even from the time of Hippocrates, palpation has been used in order to detect and characterize a body mass. The so-called virtual palpation has now become a reality due to elastography, which is a recently developed technique. Elastography has already been proving its added value as a complementary imaging method, helpful to better characterize and differentiate between benign and malignant masses. The current applications of elastography in lymph nodes (LNs) assessment by endoscopic ultrasonography will be further discussed in this paper, with a review of the literature and future perspectives.