Deep Learning Models Differentiate Tumor Grades from H&E Stained Histology Sections.

Area of Science:

• Digital pathology
• Computational oncology
• Medical image analysis

Background:

• Aberrations in tissue architecture are crucial for cancer diagnosis and grading.
• Computer-aided pathology (CAP) models can be developed by extracting aberrant phenotypes and classifying histology.
• Kidney tumor analysis benefits from detailed architectural understanding for prognostic insights.

Purpose of the Study:

• To investigate the application of convolutional neural networks (CNNs) for tumor grading in kidney histology.
• To explore the decomposition of tumor architecture using CNNs for computer-aided pathology.
• To compare the performance of deep versus shallow CNNs in analyzing H&E stained kidney tissue.

Main Methods:

• A training dataset of H&E stained kidney histology images was constructed.
• Images were classified into six categories: normal, fat, blood, stroma, low-grade granular tumor, and high-grade clear cell carcinoma.
• Performance comparison between deep and shallow CNN architectures was conducted.

Main Results:

• CNNs were applied to classify kidney tumor histology and decompose tumor architecture.
• A deeper CNN model showed superior performance over a shallower network.
• The study demonstrated the potential of CNNs for computer-aided pathology in kidney cancer.

Conclusions:

• Deep convolutional neural networks are effective for automated tumor grading and architectural analysis in kidney histology.
• Deeper CNN models offer improved performance for computer-aided pathology tasks compared to shallower counterparts.
• This approach contributes to advancing computational pathology for cancer diagnosis and outcome prediction.