Predicting Nottingham grade in breast cancer digital pathology using a foundation model

Area of Science:

• Oncology
• Computational Pathology
• Artificial Intelligence in Medicine

Background:

• The Nottingham histologic grade is vital for breast cancer severity assessment and prognosis.
• Current grading methods are subjective, time-consuming, and prone to inter-observer variability.

Purpose of the Study:

• To develop an AI-based model for automated Nottingham grading of breast cancer from whole-slide images.
• To overcome limitations of traditional subjective grading systems.

Main Methods:

• Developed an AI model using a pathology foundation model and 14 multiple instance learning algorithms.
• Trained and validated the model on 521 H&E-stained breast cancer slides from TCGA database.
• Further validated clinical utility on an additional 597 cases.

Main Results:

• The best model achieved an F1 score of 0.731 and a multiclass AUC of 0.835.
• Gene expression analysis revealed pathways related to cell division and chromosome segregation.
• Predicted grades showed a statistically significant association with 5-year overall survival.

Conclusions:

• The AI-driven automated Nottingham grading system offers an efficient and reproducible tool for breast cancer assessment.
• This system has the potential to standardize histologic grading in clinical practice.