Enhancing Lymph Node Metastasis Risk Prediction in Early Gastric Cancer Through the Integration of Endoscopic Images...

Area of Science:

• Oncology
• Medical Imaging
• Artificial Intelligence

Background:

• Accurate prediction of lymph node metastasis (LNM) and lymphovascular invasion (LVI) is critical for early gastric cancer (EGC) treatment planning.
• Current methods may lack the precision needed for optimal therapeutic strategies.

Purpose of the Study:

• To develop and validate a deep learning-based clinical decision support system (CDSS) for predicting LNM including LVI in EGC.
• To assess the performance of a transformer-based model against other deep learning approaches using real-world data.

Main Methods:

• A CDSS was developed using endoscopic images, demographic data, biopsy pathology, and CT findings from 2927 EGC patients across five institutions.
• A transformer-based model was compared with a basic CNN and a CNN with random forest model.
• Internal and external validation cohorts (449 and 766 patients, respectively) were used to assess model performance.

Main Results:

• The transformer-based model achieved a high AUC of 0.9083, outperforming CNN (AUC 0.5937) and CNN with random forest (AUC 0.7548).
• The transformer model demonstrated strong sensitivity (85.71%) and specificity (90.75%), maintaining high performance in internal and external validations.
• The model accurately identified a high percentage of patients with LNM/LVI across validation datasets.

Conclusions:

• A deep learning-based CDSS integrating real-world data can effectively predict LNM/LVI in EGC.
• This system shows potential for guiding clinical treatment strategies for EGC patients.
• The transformer-based approach offers superior performance for predicting LNM/LVI in EGC.