Blood is a fundamentally important determinant of our health as individuals. Analysis of amounts of different white blood cells in particular can be used to detect cancers include leukemia, lymphoma, and multiple myeloma. Higher-than-normal numbers of lymphocytes or monocytes can indicate the possibility of certain types of cancers. Through identifying different types of white blood cells, important decisions can be made in a medical setting, and medical training can be expediated through automatic classification.
What it does
Albus takes an image of a blood slide under a microscope and identifies each subtype of white blood cell. A confidence score is assigned to each crop of the image and further reading about the subtype is provided. In its current state, it is best suited in a medical practicioner setting for training purposes, but however can be scaled to provide full diagnoses under correct clinical supervision
There are microscopes called FoldScopes that can be added to a phone for just $1 making this website very accessible.
How we built it
We trained a machine learning model to detect and classify white blood cells. These results are sent to a PHP frontend.
Challenges we ran into
Originally, we planned on building a malaria detection app. However, we encountered difficulty training the original model for red blood cells (how malaria is detected).
Accomplishments that we're proud of
Our project works, we are able to differentiate between all 13 types of white blood cells, and crop out the appropriate sections of the images, then provide further information on these subtypes.
What we learned
Your model will never be better than the dataset.
What's next for Albus
With a better dataset, we can take Albus into a number of different directions. This can be used to classify massive amounts of images and potentially automatically provide probabilities of illnesses, to be used in a clinical setting.