Scoring model of convex probe endobronchial ultrasound multimodal imaging in differentiating benign and malignant lung lesions

Diagnosis of benign and malignant peripheral lung lesions based on a feature model constructed by the random forest algorithm for grayscale and contrast-enhanced ultrasound

Rationale and objectives

To construct a predictive model for benign and malignant peripheral pulmonary lesions (PPLs) using a random forest algorithm based on grayscale ultrasound and ultrasound contrast, and to evaluate its diagnostic value.

Materials and methods

We selected 254 patients with PPLs detected using chest lung computed tomography between October 2021 and July 2023, including 161 malignant and 93 benign lesions. Relevant variables for judging benign and malignant PPLs were screened using logistic regression analysis. A model was constructed using the random forest algorithm, and the test set was verified. Correlations between these relevant variables and the diagnosis of benign and malignant PPLs were evaluated.

Results

Age, lesion shape, size, angle between the lesion border and chest wall, boundary clarity, edge regularity, air bronchogram, vascular signs, enhancement patterns, enhancement intensity, homogeneity of enhancement, number of non-enhancing regions, non-enhancing region type, arrival time (AT) of the lesion, lesion-lung AT difference, AT difference ratio, and time to peak were the relevant variables for judging benign and malignant PPLs. Consequently, a model and receiver operating characteristic curve were constructed with an AUC of 0.92 and an accuracy of 88.2%. The test set results showed that the model had good predictive ability. The index with the highest correlation for judging benign and malignant PPLs was the AT difference ratio. Other important factors were lesion size, patient age, and lesion morphology.

Conclusion

The random forest algorithm model constructed based on clinical data and ultrasound imaging features has clinical application value for predicting benign and malignant PPLs.

Prediction of lymph node metastasis in operable cervical cancer using clinical parameters and deep learning with MRI data: a multicentre study

OBJECTIVES

To develop and validate a magnetic resonance imaging-based (MRI) deep multiple instance learning (D-MIL) model and combine it with clinical parameters for preoperative prediction of lymph node metastasis (LNM) in operable cervical cancer.

METHODS

A total of 392 patients with cervical cancer were retrospectively enrolled. Clinical parameters were analysed by logistical regression to construct a clinical model (M1). A ResNet50 structure is applied to extract features at the instance level without using manual annotations about the tumour region and then construct a D-MIL model (M2). A hybrid model (M3) was constructed by M1 and M2 scores. The diagnostic performance of each model was evaluated by the area under the receiver operating characteristic curve (AUC) and compared using the Delong method. Disease-free survival (DFS) was evaluated by the Kaplan‒Meier method.

RESULTS

SCC-Ag, maximum lymph node short diameter (LNmax), and tumour volume were found to be independent predictors of M1 model. For the diagnosis of LNM, the AUC of the training/internal/external cohort of M1 was 0.736/0.690/0.732, the AUC of the training/internal/external cohort of M2 was 0.757/0.714/0.765, and the AUC of the training/internal/external cohort of M3 was 0.838/0.764/0.835. M3 showed better performance than M1 and M2. Through the survival analysis, patients with higher hybrid model scores had a shorter time to reach DFS.

CONCLUSION

The proposed hybrid model could be used as a personalised non-invasive tool, which is helpful for predicting LNM in operable cervical cancer. The score of the hybrid model could also reflect the DFS of operable cervical cancer.

CRITICAL RELEVANCE STATEMENT

Lymph node metastasis is an important factor affecting the prognosis of cervical cancer. Preoperative prediction of lymph node status is helpful to make treatment decisions, improve prognosis, and prolong survival time.

KEY POINTS

• The MRI-based deep-learning model can predict the LNM in operable cervical cancer. • The hybrid model has the highest diagnostic efficiency for the LNM prediction. • The score of the hybrid model can reflect the DFS of operable cervical cancer.

The synergized diagnostic value of VTQ with chemokine CXCL13 in lung tumors

Virtual Touch Tissue Quantification (VTQ) offers several advantages in the diagnosis of various lung diseases. Chemokine expression levels, such as CXCL13, play a vital role in the occurrence and development of tumors and aid in the diagnosis process. The purpose of this study was to evaluate the combined value of VTQ and changes in CXCL13 expression levels for the diagnosis of lung tumors. A total of 60 patients with thoracic nodules and pleural effusion were included, with 30 of them having malignant pleural effusion (based on pathology) and the remaining 30 having benign thoracic nodules and pleural effusion. The relative expression level of CXCL13 was measured in the collected pleural effusions using Enzyme-Linked Immunosorbent Assay (ELISA). The relationship between CXCL13 expression levels and various clinical features was analyzed. A Receiver Operating Characteristic (ROC) curve analysis was conducted on the VTQ results and relative expression levels of CXCL13, and the areas under the curve, critical values, sensitivity, and specificity were calculated. Multivariate analysis incorporating multiple indicators was performed to determine the accuracy of lung tumor diagnosis. The results showed that the expression levels of CXCL13 and VTQ were significantly higher in the lung cancer group compared to the control group (P < 0.05). In the Non-Small Cell Lung Cancer (NSCLC) group, CXCL13 expression levels increased with later TNM staging and poorer tumor differentiation. The expression level of CXCL13 in adenocarcinoma was higher than that in squamous cell carcinoma. The ROC curve analysis revealed that CXCL13 had an area under the curve (AUC) of 0.74 (0.61, 0.86) with an optimal cut-off value of 777.82 pg/ml for diagnosing lung tumors. The ROC curve analysis of VTQ showed an AUC of 0.67 (0.53, 0.82) with a sensitivity of 60.0% and a specificity of 83.3%, and an optimal diagnostic cut-off of 3.33 m/s. The combination of CXCL13 and VTQ for diagnosing thoracic tumors had an AUC of 0.842 (0.74, 0.94), which was significantly higher than either factor alone. The results of the study demonstrate the strong potential of combining VTQ results with chemokine CXCL13 expression levels for lung tumor diagnosis. Additionally, the findings suggest that elevated relative expression of CXCL13 in cases of malignant pleural effusion caused by non-small cell lung cancer may indicate a poor prognosis. This provides promising potential for using CXCL13 as a screening tool and prognostic indicator for patients with advanced lung cancer complicated by malignant pleural effusion.

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.