Peripheral Blood Leukocyte Detection Based on an Improved Detection Transformer Algorithm.

Area of Science:

• Medical Imaging
• Computational Biology
• Artificial Intelligence

Background:

• Current hospital methods for white blood cell detection, like blood cell analyzers and manual microscopy, have limitations in throughput, morphology analysis, and accuracy.
• Existing methods struggle with efficiency and potential for missed detections in white blood cell analysis.

Purpose of the Study:

• To develop an advanced method for detecting peripheral blood leukocytes using Fourier ptychographic microscopy (FPM) and deep learning.
• To enhance the accuracy and efficiency of white blood cell detection in microscopic medical images.

Main Methods:

• Acquired high-resolution, wide-field microscopic images of human peripheral blood cells using FPM.
• Utilized deep convolution generative adversarial networks (DCGANs) for data enhancement and dataset construction.
• Proposed an improved DETR (detection transformer) algorithm with modifications to the feature extraction module and incorporated CIOU loss for enhanced bounding box regression.

Main Results:

• The improved DETR algorithm achieved a mean average precision (mAP) of 0.936 for human peripheral white blood cell detection.
• Demonstrated superior performance compared to other convolutional neural networks in terms of accuracy, parameter count, and inference speed.
• Verified the feasibility of the proposed method for microscopic medical image detection.

Conclusions:

• The integration of FPM and an enhanced deep learning model offers a robust solution for accurate and efficient white blood cell detection.
• The improved DETR algorithm effectively addresses challenges in detecting small white blood cell targets and improves detection accuracy.
• This approach shows significant potential for advancing automated microscopic medical image analysis in clinical settings.