Application of Machine-learning based on Radiomics Features in Differential Diagnosis of Superficial Lymphadenopathy.

Objective: The accurate diagnosis of superficial lymphadenopathy is challenging. We aim to explore a non-invasive and accurate machine-learning method for distinguishing benign lymph nodes, lymphoma, and metastatic lymph nodes.

Methods: The clinical data and ultrasound images of 160 patients with superficial lymphadenopathy (58 benign lymph nodes, 62 lymphoma, 40 metastatic lymph nodes) admitted to our hospital from January 2020 to November 2022 were retrospectively studied. Patients were randomly divided into a training set and test set according to the ratio of 6:4. Firstly, the radiomics features of each lymph node were extracted, and then a series of statistical methods were used to avoid over-fitting. Then, the gradient boosting machine(GBM) was used to build the model. The area under receiver(AUC) operating characteristic curve, precision, recall rate and F1 value were calculated to evaluate the effectiveness of the model.

Results: Ten robust features were selected to build the model. The AUC values of benign lymph nodes, lymphoma and metastatic lymph nodes in the training set were 1.00, 0.98 and 0.99, and the AUC values of the test set were 0.96, 0.84 and 0.90, respectively.

Conclusion: It was a reliable and non-invasive method for the differential diagnosis of lymphadenopathy based on the model constructed by machine learning.