White blood cell detection using saliency detection and CenterNet: A two-stage approach

Area of Science:

• Medical Image Analysis
• Computational Pathology
• Deep Learning in Hematology

Background:

• Accurate white blood cell (WBC) detection is crucial for peripheral blood smear analysis.
• Challenges include cell adhesion, varied staining, and imaging conditions, hindering traditional methods.
• Deep learning, particularly Convolutional Neural Networks (CNNs), offers powerful feature extraction but faces training inefficiencies with large, background-heavy medical images.

Purpose of the Study:

• To develop an efficient and accurate two-stage method for white blood cell (WBC) detection in peripheral blood smear images.
• To address the limitations of standard CNN training on large-scale medical datasets with significant background.
• To improve the localization and classification accuracy of WBCs.

Main Methods:

• A two-stage approach treating WBC detection as a salient object detection task.
• Stage 1: Itti's visual attention model with an adaptive center-surround difference (ACSD) operator to identify regions of interest (ROIs).
• Stage 2: A modified CenterNet model applied to ROIs for precise WBC localization and classification.

Main Results:

• The proposed method demonstrated superior performance compared to existing approaches on two distinct datasets.
• Achieved a state-of-the-art mean Average Precision (mAP) exceeding 98.8%.
• Effectively handled challenges like multi-cell adhesion and varying image conditions.

Conclusions:

• The novel two-stage deep learning strategy significantly enhances WBC detection accuracy and efficiency.
• This approach offers a robust solution for automated peripheral blood smear analysis.
• The method shows promise for clinical applications requiring high-throughput and precise cell identification.