Leukocyte Image Segmentation Using Novel Saliency Detection Based on Positive Feedback of Visual Perception

Area of Science:

• Computer Vision
• Image Processing
• Biomedical Imaging

Background:

• Salient object detection is crucial for image analysis.
• Nucleated cells are often visually salient but require precise segmentation.
• Existing methods may not fully capture subtle visual cues for accurate cell detection.

Purpose of the Study:

• To develop a novel salient object detection method for nature images.
• To adapt this method for accurate nucleated cell segmentation.
• To leverage microsaccades simulation for improved fixation prediction and iterative refinement.

Main Methods:

• Simulating microsaccades in fixational eye movements for salient object detection.
• Utilizing an existing fixation prediction method for initial area identification.
• Employing ensemble of polyharmonic extreme learning machine (EPELM) for iterative pixel classification and fixation area refinement.
• Accumulating iterative visual stimuli outputs to generate a saliency map.

Main Results:

• The proposed method demonstrates validity across three image databases.
• Successful segmentation of nucleated cells in diverse imaging conditions.
• The iterative pixel sampling-learning-classification procedure effectively refines saliency maps.

Conclusions:

• The novel method, inspired by microsaccades, is effective for salient object detection.
• This approach shows high suitability for segmenting stained nucleated cells.
• The iterative refinement process ensures robust and accurate cell segmentation.