A cell-interacting and multi-correcting method for automatic circulating tumor cells detection

Area of Science:

• Oncology
• Computational Biology
• Medical Imaging

Background:

• Circulating tumor cells (CTCs) detection is crucial for cancer diagnosis and prognosis.
• Existing methods struggle with accurate CTC capture and distinguishing CTCs from similar-looking cells.

Purpose of the Study:

• To develop a novel cell-interacting and multi-correcting detector (CMD) for automatic CTC detection in H&E-stained slide images.
• To address challenges in CTC detection, including accurate capture and differentiation from similar cells.

Main Methods:

• Developed a cell-interacting and multi-correcting detector (CMD).
• Incorporated a self-attention module for feature interaction aggregation.
• Utilized a hard sample mining sampler for correcting ambiguous classifications.

Main Results:

• CMD demonstrated superior performance over state-of-the-art methods on a multi-center dataset.
• Ablation experiments confirmed the effectiveness of the self-attention and hard sample mining modules.
• Successfully applied CMD to H&E-stained images for automated CTC detection.

Conclusions:

• The CMD detector significantly enhances the accuracy of CTC detection in H&E-stained images.
• The proposed method offers a promising tool for improving early cancer diagnosis and prognosis.
• The novel modules effectively address key challenges in automated cell detection.