White blood cell count is an important health and aging research, but can counting them can be human labour intensive. In this project, I apply machine learning to automatically detect white blood cells from blood smear images, where multiple cells may be in one image, and a bounding box around each cell is required. Given these constraints, I chose the YOLO framework, with custom training.
A random selection of test results are shown below.
The network detected white blood cells well, but due to class imbalance where neutrophil white blood cells repsented ~80% of the samples, all white blood cells were detected as neutrophil. So far, detection has worked, but classification is in progress. The next step is to balance the classes, and see if YOLO is able to distinguish the different types of white blood cells.