Computer-Aided Analysis of Oral Conventional Papanicolaou Cytology Samples

Access the paper here.

Abstract

The conventional Papanicolaou sample preparation technique is the worldwide most commonly used method for cell morphology analysis and cancer detection. It is generally applied to detect cervical cancer, but it can also be useful to detect cancer in oral cavities. However, the manual process of analyzing samples to detect abnormalities is a time-consuming task and is subject to variations in perception from different professionals. Thus, to evaluate possible automation pipelines for this process, this paper compares composed pipelines for three different deep-learning-based image understanding approaches: Semantic Segmentation, Object Detection and Image Classification. We also compare multiple models (ResNet, HRNet, U-Net, Faster R-CNN, and Mask R-CNN). We applied these methods in fields of 8 whole slides images and our results show that the Object Detection with Mask RCNN with ResNet 101 as backbone is the best approach for nuclei localization and classification (0.74 F1 score). Our results reinforce that the computer-aided analysis of oral cytology samples can help to spread this method as a screening exam to detect cancer at early stages and improve the treatment success.

Bibtex

@article{MatiasMultimethodPAPSSRN2022,
    doi = {10.2139/ssrn.4119212},
    url = {https://doi.org/10.2139/ssrn.4119212},
    year = {2022},
    publisher = {Elsevier {BV}},
    author = {Andr{\'{e}} Vict{\'{o}}ria Matias and Jo{\~{a}}o Gustavo Atkinson Amorim and Luiz Antonio Buschetto Macarini and Allan Cerentini and Felipe Perozzo Dalto{\'{e}} and Aldo von Wangenheim},
    title = {Computer-Aided Analysis of Oral Conventional Papanicolaou Cytology Samples},
    journal = {SSRN Electronic Journal}
}