Scalable Detection of Fossil Palynomorphs in Multifocal Digital Microscopy Images

TL;DR

The paper introduces an automated pipeline for detecting palynomorphs in whole slide images using multifocal digital microscopy. It enables scalable analysis by efficiently decomposing high-resolution images into 2D tiles, applying advanced object detection models like RF-DETR, and optimizing I/O for fast inference, reducing analysis time from days to under an hour.

Why It Matters

Automated detection allows palynologists to analyze large-scale geological samples with high precision, enabling more comprehensive studies of past climates and ecosystems. This advancement supports broader and faster research in paleoenvironmental reconstruction and stratigraphy.

Key Findings

• A method for compressing and decomposing multifocal digital slide images into manageable 2D tiles for efficient processing.
• RF-DETR achieved an AP@50 of 0.879 on palynomorph detection, outperforming other models in benchmarking.
• An optimized I/O pipeline and output synthesis algorithm enabled fast inference, making whole-slide analysis feasible at scale.