Combining Radiology and Pathology for Automatic Glioma Classification.

Area of Science:

• Medical imaging analysis
• Computational pathology
• Oncology research

Background:

• Accurate glioma subtype classification is crucial for effective treatment and patient care.
• Existing classification methods often rely on single data modalities, limiting their comprehensive accuracy.
• Glioma subtypes (glioblastoma, oligodendroglioma, astrocytoma) require distinct therapeutic strategies.

Purpose of the Study:

• To develop an innovative two-stage model for classifying gliomas into three subtypes using multimodal data (radiology and histology).
• To improve the accuracy of glioma subtype diagnosis by integrating information from different imaging techniques.
• To provide a tool that assists pathologists and neurologists in clinical decision-making.

Main Methods:

• A two-stage classification model was designed: stage one identifies glioblastoma, and stage two differentiates astrocytoma from oligodendroglioma.
• Radiological images were processed using 3D models, while histological images utilized 2D models within the two-stage framework.
• An ensemble classification network was employed to integrate features from both radiological and histological modalities.

Main Results:

• The proposed model achieved 1st place in the MICCAI 2020 CPM-RadPath Challenge, demonstrating its state-of-the-art performance.
• High performance metrics were reported on the validation set: balanced accuracy of 0.889, Cohen's Kappa of 0.903, and an F1-score of 0.943.
• The model successfully integrated multimodal data for robust glioma subtype classification.

Conclusions:

• The developed two-stage, multimodal model offers a significant advancement in glioma subtype classification.
• This approach enhances diagnostic accuracy and has the potential to aid clinicians in glioma treatment planning.
• The publicly available code facilitates further research and development in multimodal glioma analysis.