Methodology

The gel cards studied in this project are developed by Bio-Rad Laboratories. The gel cards are designed for ABO forward test and the determination of RH1 (D) antigens. Each card has six gel microtubes, where five of them are impregnated with a reagent specific to the erythrocyte antigen to be tested. The remaining microtube is used as control signal for test validity.

After centrifugation,  non-agglutinated Red Blood Cells (RBCs) are collected at the bottom of the microtube, while agglutinated RBCs collect either at the top of the microtube or dispersed throughout the length of the gel, being their position in the gel the intensity indicator of the reaction.

Three different procedures were studied in this project. While the first one extracts predefined region features from segmented image tubes, the other two use the Features from Accelerated Segment Test (FAST) algorithm for corner detection, using detected interest points for extracting features.

The image acquisition was preformed with a digital camera, Microsoft® LifeCam Studio™ , capable of acquiring images at a resolution of 1920 per 1080 pixels with a focal length starting in 10 centimetres up to infinity.
A physical prototype composed of a base, card stand, white background and a light stand was assembled using steel spring wire, paper and glue. As a illumination system, a 23 Watt (168 mA) light source was positioned approximately 190 mm above the card stand using a metal stool as support.

Image Segmentation and Feature Extraction

In the first tested procedure, image segmentation is applied to each image tube to extract a binary
image of the blood test after centrifugation. First, the greyscale tube image is reduced to a binary
image using a threshold computed using Otsu’s method, followed by a morphological opening.
Next, contour information is extracted from each connected cluster of pixels, using the contour to
compute the best fit ellipse based on minimizing Euclidean distance. To remove detection errors,
from the extracted ellipses is selected the one with the bigger area.

For the two other procedures, the Features from Accelerated Segment Test (FAST) algorithm was used to detect interest points in the original, non-segmented tube image. In one method, the position of all detected points is recorded and used for classification, and in the other the centre of gravity is used to compute a centre of detection using the position of all interest points.

Developed Software

In this project two different applications were developed. One was built to analyse gel cards, determining and displaying the micro-tubes test results, using the developed procedure.
The other application was developed to simplify the process of building the different image feature datasets and training the classification algorithms.