CAMIL: channel attention-based multiple instance learning for whole slide image classification.

Area of Science:

• Computational pathology
• Digital pathology
• Machine learning in medicine

Background:

• Whole-slide images (WSIs) are crucial for computational pathology classification.
• Multiple instance learning (MIL) is a robust framework for analyzing WSIs with slide-level labels.
• Existing MIL models often overlook channel dimension variability in instances, limiting information capture.

Purpose of the Study:

• To develop a novel MIL model that addresses the limitations of existing approaches in capturing channel dimension information.
• To enhance the performance of WSI classification by modeling both inter-instance relationships and intra-channel dependencies.

Main Methods:

• Proposed a plug-and-play Multi-scale Channel Attention Block (MCAB) to model interdependencies between channels using local features with varying receptive fields.
• Designed a channel attention-based MIL model (CAMIL) by integrating Transformer layers and MCAB.
• Conducted experiments on Camelyon16, TCGA-NSCLC, and TCGA-RCC datasets.

Main Results:

• The proposed CAMIL model demonstrates superior performance compared to state-of-the-art MIL models across multiple evaluation metrics.
• CAMIL achieves high performance regardless of whether the feature extractor is pretrained on natural images or WSIs.
• Empirical results validate the effectiveness of CAMIL in WSI classification tasks.

Conclusions:

• The developed CAMIL model effectively captures critical information in the channel dimension, leading to improved WSI classification.
• CAMIL offers a significant advancement in computational pathology by enhancing the analysis of gigapixel WSIs.
• The proposed approach provides a new direction for developing advanced MIL models in medical image analysis.