Multimodal Ensemble Fusion Deep Learning Using Histopathological Images and Clinical Data for Glioma Subtype...

Area of Science:

• Oncology
• Artificial Intelligence in Medicine
• Digital Pathology

Background:

• Glioma, a common central nervous system malignancy, is classified by the World Health Organization (WHO) into grades II-IV.
• Low-grade glioma (LGG) encompasses WHO grades II and III, while glioblastoma multiforme (GBM) represents WHO grade IV.
• Accurate glioma subtype diagnosis is critical for patient survival and treatment planning.

Purpose of the Study:

• To develop and evaluate a multimodal AI approach for improved glioma subtype classification.
• To integrate histopathology image features with clinical data for enhanced diagnostic accuracy.
• To assess the performance of an ensemble fusion artificial intelligence (EFAI) method in classifying LGG and GBM.

Main Methods:

• Extraction of features from histopathology whole slide images (WSIs) using multiple deep learning models.
• Concatenation of image-derived features with clinical data for multimodal analysis.
• Patch-level classification using machine learning on fused features, followed by ensemble feature selection from top models.

Main Results:

• The proposed EFAI method achieved a classification accuracy of 0.936 and an Area Under the Curve (AUC) of 0.967 on a balanced dataset (240 GBM, 240 LGG).
• Similar performance (accuracy 0.936, AUC 0.967) was observed on an imbalanced dataset (141 GBM, 242 LGG).
• The multimodal ensemble fusion approach significantly outperformed classification using only histopathology images.

Conclusions:

• The developed EFAI approach demonstrates high efficacy in differentiating glioma subtypes.
• Multimodal data integration, combining histopathology images and clinical data, enhances diagnostic performance.
• This AI-driven method shows potential for supporting clinical diagnosis and improving patient outcomes in glioma management.