Efficient nuclei segmentation model speeds up digital pathology analysis
From UNI2-h to ConvNeXt-T: Lightweight Nuclei Instance Segmentation via Knowledge Distillation
Computer Vision and Pattern Recognition
Summary
High-accuracy models that identify and separate nuclei in pathology images usually run slowly, making them hard to use in real-time medical settings. The authors created a much smaller, faster model by teaching it to mimic a larger, complex one. This small model works almost as well but runs over 20 times faster, analyzing images quickly enough for clinical use. The study also found that more complex techniques aren't needed for this efficient knowledge transfer.
What this means in practice
- •For pathology lab technicians: Use fast lightweight models to analyze tissue images for nuclei segmentation in real-time clinical diagnostics.
- •For medical imaging software developers: Integrate efficient nucleus segmentation models to speed up digital pathology workflows with close to state-of-the-art accuracy.$Commercial implications: Enables development of faster pathology image analysis software that can be marketed to hospitals and diagnostic centers.
Authors
Wenyan Li
Abstract
Nuclei instance segmentation is a core task in digital pathology, yet high-accuracy models rely on large vision transformer (ViT) encoders whose inference speed cannot meet real-time clinical demands. We propose a lightweight scheme that distills the UNI2-h pathology foundation model into a ConvNeXt-Tiny student (Ours-T, 34.7M parameters, 1/20 of the teacher) via output-level knowledge distillation. Ours-T achieves an mPQ of 0.519 on PanNuke (98.8% of the teacher), a zero-shot bPQ of 0.668 on MoNuSeg, and an inference speed of 634.3 img/s, requiring only 0.045 s for full-resolution 1024^2 analysis (21.8x speedup). Experiments further show that multi-scale gated convolution (MALA) yields no gain under ViT encoders, and output-level distillation alone suffices for efficient knowledge transfer.