Eur Radiol: Is there any metastasis of the lymph nodes in front of patients with non-small cell lung cancer?

Objective: The purpose of this study is to establish the best machine learning (ML) model using 18F-fluorodeoxycycline glucose (FDG) positive electron emission fault scanning/computer fault scanning (PET/CT) to evaluate the metastatic vertical lymph nodes (MedLNs) of non-small cell lung cancer and compare the diagnostic results with 1stdicated results of nuclear medicine physicians.
: A total of 1329 MedLNs were included in this study.
, logic regression, support vector machines, neural networks, and decision forest models were compared.
compared the diagnostic efficacy of the best ML model with that of physicians.
ML methods are divided into ML (MLq) with only quantitative variables and ML (MLc) that adds clinical information.
we analyzed the intake of 18F-FDG based on MedLNs.
results: The elevated decision tree model obtained higher sensitivity and negative predictions than doctors, but lower specificity and positive predictions.
differences between the accuracy of physicians and MLq (79.8% vs. 76.8%, p s 0.067).
accuracy of MLc is significantly higher than that of doctors (81.0% to 76.8%, p s 0.009).
in MedLNs with low intake of 18F-FDGs, ML accuracy was significantly higher than physicians (70.0% vs. 63.3%, p s 0.018).
: Although there was no significant difference in accuracy between MLq and physicians, MLc's diagnostic effectiveness was better than that of MLq and physicians.
ML method is helpful for evaluating low metabolic MedLN.
, adding clinical information to quantitative variables of 18F-FDG PET/CT can improve ML diagnostic results.
。