研究目的
To reduce the time of visual inspection analysis of stool samples for Giardia Lamblia detection by automating the process using a digital image processing algorithm.
研究成果
The algorithm successfully detects Giardia Lamblia parasites with 86% sensitivity and 67% specificity, processing images in under 2 seconds on average. It serves as a support tool for specialists but requires improvements in specificity and robustness for broader application.
研究不足
The algorithm is sensitive to microscope focus calibration, depends on specialist handling, has low specificity due to a small number of negative samples, and could be improved to include other parasites or reagents.
1：Experimental Design and Method Selection:
The algorithm is designed to process digital images from a microscope and camera setup, using HSV color space conversion, Gaussian filtering, thresholding, labeling, size filtering, Canny edge detection, and Hough transform for circle detection to identify parasites based on morphology and size.
2：Sample Selection and Data Sources:
30 digital images of stool samples prepared with lugol reagent were used, with 25 containing Giardia Lamblia and 5 without, acquired using an Amscope MD500 camera and Leica DM500 microscope.
3：List of Experimental Equipment and Materials:
Amscope MD500 camera with adapter 23 to 30.5mm, Leica DM500 optical microscope, microscope slides, lugol reagent, stool samples, workstation with USB interface for image acquisition, and software (Matlab and Python for algorithm implementation).
4：5mm, Leica DM500 optical microscope, microscope slides, lugol reagent, stool samples, workstation with USB interface for image acquisition, and software (Matlab and Python for algorithm implementation).
Experimental Procedures and Operational Workflow:
4. Experimental Procedures and Operational Workflow: Images are acquired in RGB format (800x600 pixels), converted to HSV, saturation component extracted, filtered with a 3x3 Gaussian kernel, thresholded with a fixed value of 0.25, labeled with 4-connectivity, filtered by object size (1000 to 3900 pixels), edges detected with Canny filter, and Hough transform applied for circle detection (radius 23 to 35 pixels). Detected objects are validated based on intensity criteria and marked in the original image.
5：25, labeled with 4-connectivity, filtered by object size (1000 to 3900 pixels), edges detected with Canny filter, and Hough transform applied for circle detection (radius 23 to 35 pixels). Detected objects are validated based on intensity criteria and marked in the original image.
Data Analysis Methods:
5. Data Analysis Methods: Sensitivity and specificity were calculated using true positive, false negative, true negative, and false positive counts from 30 test images, with validation by a microbiology specialist.
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
