Small sized centroblasts as poor prognostic factor in follicular lymphoma - Based on artificial intelligence analysis

Area of Science:

• Hematopathology
• Computational Pathology
• Oncology

Background:

• Follicular lymphoma diagnosis relies on centroblast assessment, but observer variability is high.
• Accurate prognostic factors are crucial for effective follicular lymphoma treatment strategies.

Purpose of the Study:

• To quantitatively analyze centroblast morphology in follicular lymphoma using artificial intelligence (AI).
• To correlate centroblast morphological features with clinical prognosis.
• To evaluate AI's potential to improve upon pathologist-based assessments.

Main Methods:

• Whole slide images of 36 follicular lymphoma cases were analyzed.
• An object detection model (YOLOv5) was fine-tuned to segment centroblasts.
• Morphological characteristics (nuclear size, area, length) were quantitatively measured.
• AI-derived features were correlated with patient prognosis (excellent, poor, indeterminate).

Main Results:

• Centroblasts in the poor prognosis group were significantly smaller in nuclear size compared to the excellent prognosis group.
• A mean nuclear area <55 μm² was associated with poorer event-free survival (p < 0.0123).
• AI demonstrated potential to capture morphological features more effectively than human observation.

Conclusions:

• Quantitative morphological analysis of centroblasts using AI can serve as a novel prognostic factor in follicular lymphoma.
• Smaller centroblast size is linked to a poorer clinical outcome.
• AI-driven pathological assessment may enhance diagnostic accuracy and prognostic prediction.