Predicting lymph node recurrence in cT1-2N0 tongue squamous cell carcinoma: collaboration between artificial...

Area of Science:

• Oncology
• Pathology
• Artificial Intelligence in Medicine

Background:

• Lymph node recurrence is a critical factor in tongue squamous cell carcinoma (SCC) prognosis.
• Predictive models for lymph node recurrence in early-stage (cT1-2N0) tongue SCC are limited, especially those integrating diverse data types.

Purpose of the Study:

• To develop a highly accurate prediction model for lymph node recurrence in cT1-2N0 tongue SCC.
• To integrate artificial intelligence (AI)-extracted histopathological data with human-assessed clinicopathological information.

Main Methods:

• A dataset of 148 patients with cT1-2N0 tongue SCC was used, split into training and testing sets.
• Prediction models were built using AI on whole slide images (WSIs), clinicopathological data, and a combination of both.
• Weakly supervised learning for WSIs and machine learning for clinicopathological data were employed.

Main Results:

• The combined model achieved an area under the ROC curve (AUC) of 0.991, significantly outperforming models using only WSI (0.826) or clinicopathological data (0.835).
• Histopathological analysis of high-prediction patches revealed increased tumor cells, inflammatory cells, and muscle content in recurrence cases.
• Mixed inflammatory cells, tumor cells, and muscle were more prevalent in recurrence versus non-recurrence cases.

Conclusions:

• Integrating AI-derived histopathological features with clinicopathological information creates a highly accurate model for predicting lymph node recurrence in cT1-2N0 tongue SCC.
• This combined approach offers a powerful tool for improving risk stratification and clinical management of tongue SCC.