Previously (in September update), we have mentioned the problem we faced during the neural network training. Long story short — the dataset was labeled not quite well, neural network’s accuracy fell, so we initiated the relabelling process.
It all went quite well — the neural network is currently being trained on a new, well-labeled dataset. Since it is quite a time consuming process, we expect to know the result of training within next couple of weeks.
The Polygon problem
While we were waiting for our dataset to be relabeled by our partners, we haven’t been sitting around twiddling our thumbs. We have paid attention to the way our neural network showcases its own prediction on each exact area.
In certain cases, our neural network has labeled the image in a very inaccurate way, with straight lines crossing and separating exactly the same label into two new ones, which could potentially provide problems in the nearest future.
The problem of polygons of the same label is fixed now, indicating the correct borders of cancer tumors and other indicators.
Platform for Whole Slide Image labeling
In our previous update we have also mentioned the platform we are currently working on. We aim to build a small team of experienced independent pathologists, who will label the images and will be well rewarded for that.
Using our own platform for image labeling would also allow us to significantly expand the pool of doctors involved in labeling in order to avoid similar mistakes when working with other areas of digital pathology when developing new neural networks of the project.
Since then, we contacted many pathologists who might be interested in cooperation with us. Since we aim to bring onboard only highly experienced specialists, the shortlist of candidates was down to only several specialists, which was subsequently reduced to three candidates we are highly excited to work with in the nearest future. The specialists of 10, 8 and 7 years of work experience in the top-level medical institutions of Russia are about to assist us in tuning our solution on prostate cancer as well as create new solutions in other particular areas of digital pathology.
Application for pilot testing
Since we are about to finish our solution development, it is time for us to look at medical insitutions where Skychain neural network on prostate cancer might be properly tested.
The main hypotheses we would like to test are about the accuracy of our solution and the value it brings to the medical institutions. We want to quantify everything we can: time, money and the most important — lives our technology might potentially save.
We have recently contacted Moscow Innovation Cluster — the governmental body, which helps projects to link up with partners, which are interested in testing new, disrupting technologies in their facilities. We have picked several large medical institutions we are interested in working with and now we are looking forward to the feedback from them and Moscow Innovation Cluster.
Thanks for the support and stay tuned for future updates!
Alexander Oksanenko, Skychain Team
If you have any questions about Skychain, don’t hesitate to write to Alexander Oksanenko on Telegram and on email: email@example.com.