TSHVNet: Simultaneous Nuclear Instance Segmentation and Classification in Histopathological Images Based on...

Area of Science:

• Computational pathology
• Medical image analysis
• Deep learning for oncology

Background:

• Accurate nuclear instance segmentation and classification are crucial for cancer diagnosis and prognosis.
• Existing methods face challenges with similar nuclear appearances, clustered instances, and limited global dependency modeling.

Purpose of the Study:

• To propose TSHVNet, a novel deep learning framework for accurate simultaneous nuclear instance segmentation and classification.
• To improve upon the state-of-the-art HoVer-Net by integrating multi-attention modules (Transformer and SimAM).

Main Methods:

• TSHVNet integrates Transformer attention on the HoVer-Net trunk for long-distance relationships.
• SimAM attention modules are applied to both trunk and branches for 3D channel and spatial weighting.
• The framework was validated on the public PanNuke and CoNSeP datasets.

Main Results:

• TSHVNet demonstrated superior performance compared to existing state-of-the-art methods.
• Nuclear instance segmentation (PQ index) improved by 1.4% (CoNSeP) and 2.8% (PanNuke).
• Nuclear classification (F1_score) increased by 2.4% (CoNSeP) and 2.5% (PanNuke).

Conclusions:

• The proposed TSHVNet effectively addresses challenges in nuclear instance segmentation and classification.
• The integration of multi-attention modules significantly enhances performance.
• TSHVNet shows great potential for advancing cancer diagnosis and prognosis through improved histopathologic image analysis.