An Artificial Intelligence Algorithm to Predict Nodal Metastasis in Lung Cancer.

Area of Science:

• Pulmonology
• Medical Imaging
• Artificial Intelligence

Background:

• Endobronchial ultrasound (EBUS) offers high accuracy in predicting lymph node (LN) malignancy.
• Clinical application of EBUS is limited by operator dependency.
• An AI algorithm, NeuralSeg, was developed to identify and predict LN malignancy from EBUS images.

Purpose of the Study:

• To evaluate the accuracy of the NeuralSeg AI algorithm in predicting lymph node malignancy using EBUS images.
• To assess the diagnostic performance of NeuralSeg compared to traditional methods.

Main Methods:

• EBUS images were used to train and validate the NeuralSeg algorithm.
• The algorithm underwent 5-fold cross-validation in the derivation phase.
• Logistic regression and ROC curves analyzed NeuralSeg's discrimination between benign and malignant LNs, with pathological specimens as the gold standard.

Main Results:

• NeuralSeg achieved 73.8% accuracy in predicting malignant LNs during derivation.
• In validation, NeuralSeg demonstrated 72.9% accuracy, 90.8% specificity, and 75.9% negative predictive value.
• The AI algorithm showed improved diagnostic discrimination in the validation cohort (c-statistic = 0.75) compared to derivation (c-statistic = 0.63).

Conclusions:

• NeuralSeg accurately rules out nodal metastasis.
• The AI algorithm can serve as an adjunct to EBUS, especially when nodal biopsy is inconclusive or not possible.
• Further clinical trials are necessary to validate the algorithm's performance in real-world settings.