Semantic Segmentation for the Detection of Very Small Objects on Cervical Cell Samples Stained with the AgNOR Technique

Access the paper here.

Abstract

Semantic segmentation has been used widely for cytological imaging for several body parts and different stain types. This work explores deep learning models for the segmentation of very small cellular structures in AgNOR-stained cervical cell images and compares state-of-the-art models (HRNet and SegFormer) with traditional U-nets using ResNets as their backbone. AgNOR has been used as a coadjuvant staining technique for the diagnosis of cervical and other cancer types. Our results presented the transformer model, SegFormer, as the best model to segment nuclei. HRNet is the best model to segment the very small object categories (the NORs — Nucleolar Organizer Regions). HRNet showed to be more suitable for AgNOR image segmentation as a whole, with 84.7% precision, 84.6% recall, and 84.6% F-score for nuclei, and 70.1% precision, 55.4% recall and 61.8% F-score on the detection of NORs.

Bibtex

@article{AtkinsonSegmentationAgNORSSRN2022,
    author= {Jo{\~{a}}o Gustavo Atkinson Amorim and Andr{\'{e}} Vict{\'{o}}ria Matias and Allan Cerentini and Fabiana Botelho de Miranda Onofre and Alexandre Sherlley Casimiro Onofre and Aldo von Wangenheim},
    doi = {10.2139/ssrn.4126881},
    url = {https://doi.org/10.2139/ssrn.4126881},
    year = {2022},
    publisher = {Elsevier {BV}},
    title = {Semantic Segmentation for the Detection of Very Small Objects on Cervical Cell Samples Stained with the {AgNOR} Technique},
    journal = {SSRN Electronic Journal}
}