Eosinophils Instance Object Segmentation on Whole Slide Imaging Using Multi-label Circle Representation.

Area of Science:

• Medical diagnostics
• Computational pathology
• Image analysis

Background:

• Eosinophilic esophagitis (EoE) is a chronic inflammatory condition of the esophagus.
• EoE diagnosis relies on counting esophageal eosinophils (Eos), a labor-intensive manual process.
• Current methods lack efficiency and can be subjective, necessitating automated solutions.

Purpose of the Study:

• To develop an automated method for accurate eosinophil instance segmentation in EoE diagnosis.
• To adapt the CircleSnake model for multi-label segmentation of eosinophils.
• To compare the performance of the proposed model against existing methods like Mask R-CNN.

Main Methods:

• Proposed a multi-label CircleSnake model, extending the original single-label CircleSnake.
• Applied the model to instance segmentation of eosinophils (Eos) in esophageal tissue images.
• Evaluated performance using average precision (AP) and compared with Mask R-CNN and DeepSnake.

Main Results:

• The multi-label CircleSnake model demonstrated superior performance in identifying and segmenting eosinophils.
• Achieved higher average precision (AP) compared to Mask R-CNN and DeepSnake models.
• The automated approach streamlines the assessment process for EoE.

Conclusions:

• The multi-label CircleSnake model offers a promising automated solution for eosinophil segmentation in EoE.
• This advancement can improve diagnostic accuracy and efficiency in clinical practice.
• Publicly available source code facilitates further research and application.