Methods for Segmentation and Classification of Digital Microscopy Tissue Images.

Area of Science:

• Digital pathology
• Computational biology
• Cancer research

Background:

• High-resolution microscopy images offer detailed tissue morphology insights.
• Accurate image analysis is crucial for understanding cancer biology.
• Nuclei segmentation and tissue classification are challenging tasks due to tissue complexity and tumor heterogeneity.

Purpose of the Study:

• To develop accurate and efficient computer algorithms for nuclei segmentation and whole slide tissue image classification.
• To enhance the understanding of cancer biology through advanced image analysis techniques.

Main Methods:

• A multiscale deep residual aggregation network was used for nuclei segmentation, including separating clumped nuclei.
• A deep learning method was employed for patch-level classification, with features fed into a random forest regression model for whole slide classification.

Main Results:

• The segmentation algorithm achieved an accuracy score of 0.78.
• The classification algorithm achieved an accuracy score of 0.81.
• Both algorithms achieved the highest scores in the MICCAI 2017 Digital Pathology challenge.

Conclusions:

• The developed algorithms demonstrate high accuracy and efficiency in nuclei segmentation and whole slide tissue image classification.
• These algorithms represent a significant advancement in digital pathology and cancer research image analysis.
• The success in the MICCAI 2017 challenge highlights the potential of these methods for real-world applications.