Microvessel prediction in H&E Stained Pathology Images using fully convolutional neural networks

Area of Science:

• Oncology
• Pathology
• Computational Biology

Background:

• Pathological angiogenesis is crucial in malignancies, often assessed by microvessel density (MVD) via CD31/CD34 immunostaining.
• Public pathology data predominantly uses Hematoxylin and Eosin (H&E) staining, lacking corresponding immunohistochemistry.
• Manual microvessel identification in H&E images is laborious and prone to observer variability.

Purpose of the Study:

• To develop an automated algorithm for microvessel detection in H&E stained pathology images.
• To address the limitations of manual MVD assessment and the scarcity of immunohistochemistry data.
• To facilitate clinical association studies using readily available H&E images.

Main Methods:

• Utilized fully convolutional neural networks (FCNNs) for microvessel prediction.
• Applied the developed algorithm to H&E stained pathology images.
• Validated the feasibility and performance of the automated detection method.

Main Results:

• Successfully demonstrated the feasibility of the proposed microvessel prediction algorithm on H&E images.
• Identified microvessel features that showed significant association with patient clinical outcomes.
• Provided a novel approach for analyzing angiogenesis in H&E stained samples.

Conclusions:

• This study presents the first algorithm for automated microvessel detection in H&E stained pathology images.
• The developed method offers a valuable tool for prognostic and therapeutic target analysis in oncology.
• Automated analysis of H&E images can unlock vast public datasets for cancer research.