AUTHOR=Wang Linyan , Jiang Zijing , Shao An , Liu Zhengyun , Gu Renshu , Ge Ruiquan , Jia Gangyong , Wang Yaqi , Ye Juan 

TITLE=Self-supervised learning mechanism for identification of eyelid malignant melanoma in pathologic slides with limited annotation

JOURNAL=Frontiers in Medicine

VOLUME=Volume 9 - 2022

YEAR=2022

URL=https://www.frontiersin.org/journals/medicine/articles/10.3389/fmed.2022.976467

DOI=10.3389/fmed.2022.976467

ISSN=2296-858X

ABSTRACT=Purpose：Lack of fine annotated pathologic data has limited the application of deep learning systems (DLS) to the automated interpretation of pathologic slides. Herein, we develop a robust self-supervised learning (SSL) pathology diagnostic system to automatically detect malignant melanoma (MM) in the eyelid with limited annotation. 
Design: Development of a self-supervised diagnosis pipeline based on public dataset, then refined and tested in private real clinical dataset.
Subjects: A. Patchcamelyon (PCam)-a publicly accessible dataset for the classification task of patch-level histopathologic images. B. The Second Affiliated Hospital, Zhejiang University School of Medicine (ZJU-2) dataset - 524307 patches (small sections cut from pathologic slide images) from 192 H&E-stained whole-slide-images (WSIs); only 72 WSIs were labeled by pathologists.
Methods: PCam was used to select a convolutional neural network (CNN) as the backbone for our SSL-based model. This model was further developed in ZJU-2 dataset for patch-level classification with both labeled and unlabeled images to test its diagnosis ability. Then the algorithm retrieved information based on patch-level prediction to generate WSI-level classification results using random forest. A heatmap was computed for visualizing decision-making process.
Main outcome measure(s): The area under the receiver operating characteristic curve (AUC), accuracy, sensitivity, specificity were used to evaluate the performance of the algorithm in identifying MM.
Results：ResNet50 was selected as the backbone of SSL-based model using PCam dataset. This algorithm then achieved an AUC of 0.981, with the accuracy, sensitivity, specificity of 90.9%, 85.2%, 96.3% for patch-level classification of ZJU-2 dataset. For WSI-level diagnosis, the AUC, accuracy, sensitivity, specificity was 0.974, 93.8%, 75.0%, 100% separately. For every WSI, a heatmap was generated based on the malignancy probability.
Conclusions: Our diagnostic system, which is based on SSL and trained with dataset of limited annotation, can automatically identify MM in pathologic slides and highlight MM area in WSIs by a probabilistic heatmap. In addition, this laborsaving and cost-efficient model has the potential to be refined to help diagnose with other ophthalmic and non-ophthalmic malignancies.