Hello community,
we currently working on an automatic algorithm for erytrocyte detection in stereo-microscopy images for cell counting. Additionally we want to determine the percentile amount of echinocytes during sample preparation.
In image detection, the usage of haar-cascade-xml databases is a very powerful and popular method for face, body, car etc. detection. However, the training of the model which delivers the xml-file requires a huge dataset of presorted red blood cell images. Unfortunatly, we do not have such a dataset. Maybe someone can provide me an already trained xml-file for red blood cell and/or echinocyte detection?
Thanks in advance!
Kind regards,
Karl T
A pre - trained model of haar cascade on a data set is the algorithm itself which you are trying to develop. Once you get the XML file, you have to just execute it on OpenCV or integrate with the software you are making. Also there is a haar cascade model trained on white blood cells. Check out the link of it.
https://www.semanticscholar.org/paper/Localization-of-white-blood-cell-images-using-Haar-Budiman-Achmad/68a23750c27f2c11f7b937575bb801a971b76637
Dear Karl: Several years ago I developed some software in Visual Basic 5.0 or 6.0 to detect tissue cells stained by immunoperoxidase and developed with diaminobenzydine (DAB). The program did not count the cells, it just found them and stored their location for the pathologist to look at that location and therefore save time. The point is that the software was very easy to write and the results were very good: A picture (.bmp) was scanned in the X and Y dimensions with the program checking if the pixel in question was within the RGB range of colors pre-programmed. If it did, the program painted a red dot on top. At the end of the process, which took only a few seconds, the picture was covered with red spots indicating where the stained cells were. The speed is high because the way Windows(R) works: The analysis is done on real pixels (the ones you see on the screen), not virtual pixels and therefore the max number would be the screen resolution, which is not too much (1366 x 768 = 1,049,088 pixels).
It might be possible to use the red color on the erythrocytes to distinguish them from the background and proceed with other routines to count them.
Another trick that I also used many years ago, for any cells, was to wash them with isotonic glucose, which conducts electricity very poorly and then re-suspend the cells in distilled water. That blew up the cells and released their ions into the water. That changed the conductivity in a way that was proportional to the number of cells. I developed several high sensitivity conductivity readers, but with the number of erythrocytes in any sample it is likely that any commercial conductivity reader will do (you cannot use an Ohm-meter because it polarizes the electrodes; you need to flip them around). You can also be able to detect the ions amperometrically.
Do not hesitate to contact me if needed even though I doubt that I have the software or the schematics of those projects. Good luck, Ricardo
- Badges
- Science topic
- Similar topics
- Biological Science
- Molecular Cell Biology