Deep Learning techniques can and have been applied to produce a range of different solutions over the past few years.

Having mastered the Facial Recognition requirements, we decided it was time to see if our Deep Learning platform could create an AI (Artificial Intelligence) to recognise other classifications.

We set one of our Research Engineers the challenge of training an AI to recognise the difference between cancerous and non-cancerous tissue. The data used for the project were histopathology images that were pre-labelled as to whether cancer was present or not. Histopathology images are taken under the microscope and are stained to enhance cellular features in the image.

One of the most popular methods for diagnosing breast cancer on histology images is by detecting mitosis (cells that are in the process of division). Historically a professionally trained physician looks at each individual frame and marks the presence of mitosis, in a very slow and time consuming process. There has been a lot of interest in finding an accurate and automatic solution to this problem. To this end a medical imaging challenge was set in 2014 by the International Conference of Pattern Recognition that concentrated on finding a solution to this problem.

Using the dataset provided for the challenge, we created an AI to detect the occurrence of mitosis. An analysis of the accuracy on a blind test set, demonstrated outstanding results and even suggest a higher level of accuracy that the 2014 competition winner.

This challenge has proven that our Deep Learning platform has the power and capability to be applied to completely unrelated fields to our core area of experience in face recognition.   We feel confident that we would be able to apply our technology to a wide variety of sources with incredibly reliable results in a relatively short period of time.

To find out more about our developments in Deep Learning click here.