We Use Machine Learning & Computer Vision To Help Doctors Identify Melanoma and Breast Cancer.
It all began with proposed machine vs doctor competition. Our friends from largest hospital in Europe were excited about Neural Networks and proposed to conduct competition of real doctors vs specially trained machine in order to identify melanoma and breast cancer using high resolution X-Ray scanners. Within one month such system was created and we conducted competition which was won by the robots. Using Computer Vision and Deep Learning we could quickly identify the anomalies in the image, thus spot existing decease.
Customer Challenge.
Main challenge in healthcare is availability of doctors and their experience in the field. Usually top professionals are concentrated among large cities and it is very hard to cover regions and remote locations. To show you the scale of the problem, there is a shortage of 127,000 eye doctors in India and 45% of the patients loss their vision before diagnosis (data: Google I/O).
So our customer wanted to have backup solution that will help:
To save doctor’s time during their day-to-day work;
To assist doctors during large-scale events (such as Melanoma Day);
To cover areas and regions where doctors are not easily available.
Solution.
For this project we created two layers system. First processing layer did remove ribs and rib’s shadow from the X-Ray image. Second layer did analysis of the image based on provided data set of 100x X-Ray images. The system compares pixel by pixel in order to identify potential danger and highlight it to the doctor, thus saving critical time and eliminating human factor.
Results.
Although the result was quite promising, we do understand that it is just the beginning of much longer journey. We can work on system speed, accuracy and include more scenarios in our work. Currently customer plan to use our algorithm for mass events when hundreds or thousands of images are taken and it is impossible to cover it with existing staff. Good examples are National Melanoma Day in US or European Melanoma Day.
Talking about our future plans, we may look at example (see on the right) of work from Google presented at Google I/O on how deep learning system can predict cardiovascular risk using your retina image. This is the potential way of developing system further.