M4: Multi-proxy multi-gate mixture of experts network for multiple instance learning in histopathology image analysis

Area of Science:

• Computational pathology
• Artificial intelligence in medicine
• Digital pathology

Background:

• Multiple instance learning (MIL) is crucial for whole slide images (WSIs) analysis in computational pathology.
• Current MIL methods often focus on single tasks, limiting efficiency and overlooking task interdependencies.

Purpose of the Study:

• To develop an efficient MIL framework for simultaneous prediction of multiple genetic mutations from WSIs.
• To address the limitations of single-task learning in computational pathology.

Main Methods:

• Proposed an adapted architecture: Multi-gate Mixture-of-experts with Multi-proxy for Multiple instance learning (M4).
• Implemented a multi-gate mixture-of-experts strategy for simultaneous prediction of multiple genetic mutations.
• Introduced a multi-proxy CNN for expert and gate networks to capture patch-patch interactions within WSIs.

Main Results:

• M4 demonstrated significant improvements across five TCGA datasets.
• Achieved superior performance compared to state-of-the-art single-task MIL methods.
• Successfully enabled simultaneous prediction of multiple genetic mutations from WSIs.

Conclusions:

• The M4 framework offers an efficient and effective approach for multi-task learning in WSI analysis.
• M4 captures inter-task relatedness, outperforming single-task methods.
• This work advances computational pathology by enabling simultaneous prediction of multiple biomarkers from WSIs.