Crf model improves cytology image classification accuracy using stain specific clues
Refining Cytology Predictions with Conditional Random Fields
Computer Vision and Pattern Recognition
Summary
Classifying cells in cytology images is harder than in tissue images because the stains and cell shapes differ a lot. The authors show that improving predictions by sharing information between nearby patches helps fix errors, but older methods designed for tissue don’t work well for cytology. They created CytoCRF, which uses special rules tuned for stained cell details and combines different models to better connect patches. This approach consistently beats earlier methods on ten cytology datasets, even with very few labeled examples.
What this means in practice
- •For pathology ai developers: Improve cytology image classification accuracy in diagnostic tools using stain-specific conditional random fields.$Commercial implications: Enables AI-powered cytology diagnostics software to deliver more accurate cell classifications, enhancing clinical decision making.
- •For medical image analysis engineers: Integrate CytoCRF to refine predictions in multi-stain cytology image datasets with limited annotations.
Authors
Manon Dausort, Tiffanie Godelaine, Karim El Khoury, Maxime Zanella, Christophe De Vleeschouwer, Benoît Macq
Abstract
Vision-language models (VLMs) achieve strong zero-shot (ZS) classification on histology images but do not perform as well on cytology, whose stains and cell morphology differ markedly compared to histology. Conditional random fields (CRFs) can refine noisy VLM predictions by propagating information across patches, but existing CRF frameworks were designed for histopathology and do not transfer to cytology datasets, released as independent patch pools spanning multiple staining protocols. We introduce CytoCRF, which adapts the pairwise terms to cytology by targeting chromatin and cytology-specific staining, and further enrich the neighborhood of each potential term by combining multiple backbones. Across ten cytology datasets, CytoCRF outperforms existing CRF frameworks at every annotation budget, reaching +13.6 percentage points over the best baseline and +33.7 over ZS with only 50 annotations. Combining information from multiple backbones brings further gains, showing that the neighborhood topology matters more than the pairwise potential computed over it.