Classification of Diffuse Glioma Subtype from Clinical-Grade Pathological Images Using Deep Transfer Learning

Area of Science:

• Neuropathology
• Artificial Intelligence
• Oncology

Background:

• Diffuse gliomas are common primary brain tumors with variable characteristics.
• The 2016 World Health Organization (WHO) guidelines integrate morphology and molecular data for glioma diagnosis.
• Accurate classification is crucial for effective patient treatment and prognosis.

Purpose of the Study:

• To develop and evaluate a deep learning model for automatic classification of diffuse glioma subtypes and grades.
• To utilize whole-slide images from routine clinical practice for glioma analysis.
• To assess the model's performance based on the latest WHO classification standards.

Main Methods:

• A deep transfer learning approach using the ResNet50V2 model was employed.
• The model was trained on whole-slide images of diffuse gliomas.
• Classification of glioma subtypes and grades was performed according to the WHO 2016 guidelines.

Main Results:

• The deep learning model achieved a balanced accuracy of 0.8727 for diffuse glioma subtype classification.
• Majority voting enhanced the classification performance.
• The study demonstrates the feasibility of using deep learning on routine clinical data.

Conclusions:

• Deep learning offers a promising tool for automated glioma classification in neuropathology.
• This technology can support pathologists in making more accurate and efficient diagnoses.
• The findings suggest a significant role for artificial intelligence in the future of diagnostic pathology.