Artificial Intelligence-Based Histopathology Segmentation for Resource-Constrained Healthcare Systems

Area of Science:

• Medical Imaging
• Computational Pathology
• Artificial Intelligence in Oncology

Background:

• Colorectal cancer (CRC) poses a significant global health challenge, necessitating accurate histopathological analysis for diagnosis.
• Existing deep learning models for tissue segmentation often demand high computational resources, hindering their clinical applicability in diverse healthcare settings.
• Resource-constrained environments require robust diagnostic tools capable of handling variations in staining and infrastructure.

Purpose of the Study:

• To develop a novel, resource-efficient colorectal histopathology segmentation network (RCHS-Net).
• To ensure reliable performance across heterogeneous clinical settings and varying laboratory conditions.
• To provide a scalable solution for AI-assisted cancer diagnosis in real-world healthcare systems.

Main Methods:

• RCHS-Net utilizes a compact multi-scale encoder, a gland context module with atrous convolutions and self-attention, and a feature pyramid decoder.
• Feature-wise linear modulation (FiLM) conditioning enables class-aware segmentation.
• MixStyle augmentation enhances stain domain generalization for improved robustness.

Main Results:

• RCHS-Net achieved a 95.20% Dice coefficient and 91.10% IoU on the EBHI-Seg dataset with only 243,226 parameters.
• On the GlaS dataset, the model attained a 93.39% Dice score and 88.32% IoU.
• The network demonstrated superior performance compared to state-of-the-art methods while maintaining a compact architecture.

Conclusions:

• High-accuracy histopathology segmentation is achievable with compact deep learning architectures like RCHS-Net.
• RCHS-Net offers a practical and scalable solution for AI-assisted cancer diagnosis in complex healthcare environments.
• The developed model supports equitable and globally scalable cancer diagnostics.