|
The visual assessment of samples in microscopes are important aspects of science and diagnosis.
These often require a huge amount of hours spent looking into the small monocular recognising shapes,
colors, measuring and counting.
A camera with a small computer onto the microscope with the idea that this knowledge about what we are
seeing could be trained onto a cheap micro computer and distributed around the world where trained lab
technicians are not available.
This is enabling advanced diagnosis and understanding of rare or complex
findings in the microscope for the untrained users and also save huge amount of time counting
and collecting data from microscope samples.
Instead of training a new neural network from scratch which takes many hundred of hours on a high powered GPU.
We use something called transfer learning to utilise an all-ready trained network that can recognise images and
adapting this to recognise red and white blood-cells.
This makes the process way faster so we can test the idea in a rapid prototyping environment.
|
