Antibody Supervised Training of a Deep Learning Based Algorithm for Leukocyte Segmentation in Papillary Thyroid...

Area of Science:

• Digital Pathology
• Computational Pathology
• Cancer Research

Background:

• Leukocyte quantification in papillary thyroid carcinoma (PTC) holds prognostic and predictive value.
• Accurate segmentation of leukocytes in histological images is crucial for analysis.

Purpose of the Study:

• To develop and validate a novel method for training a convolutional neural network (CNN) algorithm to segment leukocytes in PTC tissue images.
• To utilize destain-restain immunohistochemistry (IHC) for generating supervised annotations for CNN training.

Main Methods:

• Tissue samples from PTC cohorts were stained with hematoxylin and eosin (HE), destained, and restained with anti-CD45 IHC.
• Registered HE and CD45 IHC images were used to create binary masks of leukocyte locations.
• These masks served as annotations for training a CNN algorithm on digitized HE image tiles.

Main Results:

• The trained CNN achieved a high intersection over union (IoU) of 0.82 for leukocyte detection in HE images.
• The destain-restain IHC guided annotation method resulted in accurate leukocyte segmentation.
• The algorithm was validated on an independent test set of whole slide images (WSIs).

Conclusions:

• The proposed destain-restain IHC guided annotation strategy enables accurate and automated leukocyte segmentation in HE stained PTC images.
• This method offers a robust approach for computational pathology applications in cancer research.
• High-accuracy leukocyte segmentation can aid in prognostic and predictive assessments in papillary thyroid carcinoma.