Artificial Intelligence for Lymph Node Detection and Malignancy Prediction in Endoscopic Ultrasound: A Multicenter...

Area of Science:

• Gastroenterology and Hepatology
• Oncology
• Medical Imaging

Background:

• Endoscopic ultrasound (EUS) is vital for lymph node (LN) characterization in cancer staging.
• Current EUS criteria for malignancy prediction are imprecise, and histology has limitations.
• A novel artificial intelligence (AI) system was developed to enhance LN malignancy prediction from EUS images.

Purpose of the Study:

• To evaluate the effectiveness of a novel AI-based system in predicting lymph node malignancy using EUS images.
• To assess the performance of a deep learning model for lymph node assessment in a multicenter setting.

Main Methods:

• A multicenter study involving 59,992 EUS images from 82 procedures.
• Development of a convolutional neural network (CNN) using the YOLO architecture for detection and classification.
• Definitive diagnoses confirmed by cytology, biopsy, surgical specimens, or a minimum six-month follow-up.

Main Results:

• The AI model achieved high diagnostic performance: 98.8% sensitivity, 99.0% specificity, and 98.3% overall accuracy.
• Negative and positive predictive values for malignancy were 98.8% and 99.0%, respectively.
• The AI system demonstrated excellent detection and classification capabilities for malignant LNs.

Conclusions:

• This study is the first to evaluate deep learning for LN assessment via EUS imaging.
• The AI-powered imaging model shows significant potential as a tool to refine LN evaluation.
• This technology can support more tailored and efficient patient care in oncology.