GSN-HVNET: A Lightweight, Multi-Task Deep Learning Framework for Nuclei Segmentation and Classification.

Area of Science:

• Digital pathology
• Computer-aided diagnosis
• Deep learning

Background:

• Nuclei segmentation and classification are crucial for digital pathology image analysis.
• Current deep learning methods often use separate networks, leading to inefficiency and large model sizes.

Purpose of the Study:

• To propose a lightweight deep learning framework (GSN-HVNET) for simultaneous nuclei segmentation and classification.
• To improve computational efficiency and network stability.

Main Methods:

• Developed an encoder-decoder framework with a three-branch decoder for semantic segmentation, horizontal and vertical (HV) distances, and nucleus classification.
• Introduced Residual-Ghost-SN (RGS) and Dense-Ghost-SN (DGS) blocks to reduce computational cost and enhance stability.
• Redefined the classification principle for the CoNSeP dataset for practical pathological diagnosis.

Main Results:

• The proposed GSN-HVNET model achieved superior segmentation and classification accuracy compared to state-of-the-art methods.
• The model demonstrated high computational efficiency, making it suitable for practical applications.
• Instance segmentation results were effectively obtained by combining semantic segmentation and HV distance outputs.

Conclusions:

• GSN-HVNET offers an efficient and accurate solution for simultaneous nuclei segmentation and classification in digital pathology.
• The novel RGS and DGS blocks contribute to reduced computational load and improved performance.
• The framework shows significant potential for advancing computer-aided diagnosis in pathological analysis.