ABSTRACT Background and objective: To establish a ultrasound radiomics machine learning model based on endobronchial ultrasound (EBUS)to assistdoctors in distinguishing between benign and malignant diagnoses ofmediastinal and hilar lymph nodes.

Methods: The clinical and ultrasonic image data of 197 patients wereretrospectively analyzed. The radiomics features were extracted by EBUS.based radiomics and dimensionality reduction was performed on thesefeatures by the least absolute shrinkage and selection operator (LASSO)EBUS-based radiomics model was established by support vector machine(SVM).205 lesions were randomly divided into a training group (n=143)and a validation group (n=62). The diagnostic efficiency was evaluated byreceiver operating characteristic (ROC).Results: A total of 13 stable features with non-zero coefficients wereselected. The support vector machine (SV) model exhibited promisingperformance in both the training and verification groups. In the traininggroup, the SVM model achieved an area under the curve (AUC) of 0.892(95% CI: 0.885-0.899), with an accuracy of 85.3%, sensitivity of 93.2%and specificity of 79.8%.In the verification group, the SVM modeldemonstrated an AUC of 0.906 (95% C: 0.890-0.923),along with anaccuracy of 74.2%,sensitivity of 70.3%, and specificity of 74.1% Conclusion:EBUS-based radiomics model can be used to differentiatemediastinal and hilar benign and malignant lymph nodes. The SVM modeldemonstrates superiority and holds potential as a diagnostic tool in clinical practice

Inclusion Criteria:

• (1)chest CT showing enlarged mediastinal or hilar lymph nodes or positive findings of mediastinal and hilar lymph nodes on PET/CT (SUV≥2.5); (2)patients who underwent EBUS-TBNA examination; (3)no contraindications for surgery.

Exclusion Criteria:

• (1)prior treatment of target lymph nodes before EBUS-TBNA; (2)unclear diagnostic results; (3)loss to follow-up.

Henan Provincial People's Hospital