Artificial intelligence modelling in grading breast phyllodes tumours

Area of Science:

• Digital pathology
• Computational pathology
• Artificial intelligence in medicine

Background:

• Breast phyllodes tumors (PT) are rare biphasic neoplasms with benign, borderline, and malignant classifications.
• Accurate grading of PTs is challenging due to complex histological parameters.
• Investigating artificial intelligence (AI) as a diagnostic aid for PT grading is crucial.

Purpose of the Study:

• To explore the potential of AI in stratifying breast phyllodes tumor grades.
• To assess AI's ability to classify PTs based on histological similarities.
• To evaluate AI as a tool for improving diagnostic accuracy in PT grading.

Main Methods:

• Utilized 15 PT whole slide images (WSIs) across benign, borderline, and malignant categories.
• Employed the Yottixel framework for WSI processing and KimiaNet for feature extraction.
• Compared WSI barcodes at various patch sizes to identify histological similarities for grading.

Main Results:

• Achieved a maximum accuracy of 67% for PT grade stratification.
• Optimal performance was observed with a 3000×3000 patch size using majority voting (n=4).
• Demonstrated AI's capability in identifying relevant histological features for tumor grading.

Conclusions:

• AI shows potential for grading breast phyllodes tumors through histological feature matching.
• This study serves as a proof of concept for AI-driven PT grade stratification.
• Further refinement could lead to routine clinical application of AI in PT diagnosis.