研究目的
To segment red blood cells from the Full-Field OCT data of human skin using deep learning technique for real-time detection and counting.
研究成果
The proposed deep learning model accurately segments red blood cells from Full-Field OCT images, showing promise for real-time detection and counting. This method could be extended for the segmentation of the whole field of view of acquired images, aiding in the analysis of blood flux and flow speed related to immunologic diseases.
研究不足
The study is an exploratory trial with data from a single volunteer, which may limit the generalizability of the results. The signal-to-noise ratio is reduced in the dermis, affecting image quality.
1：Experimental Design and Method Selection:
The study employs a deep learning approach, specifically a modified U-Net model, for segmenting red blood cells from OCT data.
2：Sample Selection and Data Sources:
A skin image volume from a 35-years-old volunteer was used, with images captured in gray format.
3：List of Experimental Equipment and Materials:
Full-Field OCT imaging system.
4：Experimental Procedures and Operational Workflow:
Image data labeling and pre-processing included gamma scaling, Gaussian blurring, thresholding, and clearing erroneous structures. Training and validation datasets were created by pasting segmented RBCs onto different backgrounds.
5：Data Analysis Methods:
Model evaluation was based on Intersection over Union (IoU) as a measure of accuracy.
独家科研数据包，助您复现前沿成果，加速创新突破
获取完整内容
