Crowdsourcing Solutions to Cancer: Thank You, Kaggle, Intel, and MobileODT

istock 520613532

“Competition platforms” offer interesting ways to pull together communities to focus on solving a problem. There are several out there. For example, Kaggle provides access to development tools and the compute cycles to run algorithms. They also have some well-written tutorials and training that I highly encourage you to try. (My wife got lost in the analysis of survival rates for passengers of the Titanic – an excellent introduction to machine learning to do at your own pace.)


There is currently a Kaggle competition to develop an algorithm that will help health care workers in rural parts of the world prevent cervical cancer by applying the right treatment in high-risk pre-cancerous situations. See the Intel and MobileODT Cervical Cancer Screening Competition for a more detailed explanation. (The competition offers $100,000 in prize money, by the way.)

My wife is a survivor of cervical cancer, so the story of the extra challenges in rural situations was especially interesting. We have it easy by comparison. Several years ago, my wife went in for a routine PAP smear and it came back atypical. It took several follow-up visits and some additional testing to select her treatment method, but she was able to be treated with a fairly routine hysterectomy and is cancer free today. However, even with all of the resources available to her, navigating the treatment was challenging.

Imagine what it’s like for women in developing countries. Most aren’t even able to come in for regular screenings, much less for treatment follow-ups. Finding the right treatment to prevent cancer in one visit is essential in these situations but hasn’t been realistic for a couple of reasons.


Key EVA technology: cheaper and local thanks to MobileODT

So, some people dared to ask the question: Can we find an inexpensive solution that can be deployed in remote areas?

One of the technologies that MobileODT is building on for this competition is their EVA (Enhanced Visual Assessment) system. They interfaced optical components with standard Android smartphones to create low-cost, easily deployable colposcopy machines.  That opened new access for many rural clinics, and EVA pilots are now in progress in more than 20 countries. When a woman in one of these areas is finally able to make it into her local clinic, she’ll be able get a cervical screening on site.

Next dream: Do it in one visit

Then some people dared to ask the question: Can we do this in one visit?  What would it take?

Screening is a really important first step, but it doesn’t address navigating the course of treatment. Multiple studies show that treatment needs to be immediate or it’s unlikely to happen in time. The challenge is that the best course of treatment depends on the physiology of the patient, and it takes expert guidance to determine that physiology. Here’s where we hope that Machine Learning comes to the rescue. The Kaggle challenge is to develop an algorithm to determine a woman’s cervix type based on images. That information can then be used to navigate treatment in real time.

If you check out this competition, you’ll find training sets and have access to tools. There’s an overview of Intel’s Deep Learning SDK, along with information on how to log into a hands-on remote access compute environment for use by participants.

As of late May, there were 667 teams entered in the competition. We can learn a lot just by looking at the work-in-progress entries that have been submitted. They include code, comments, outputs, and split-off forks. It all feels extremely collaborative – a lot of smart people out to make the world a better place. These will be interesting to study long after the competition is over.

I am proud to be part of an HPC community that is making technology more accessible and approachable. And I am inspired by some of the recent competitions. I hope you are too.

Click here to download your free 30-day trial of Intel Parallel Studio XE


Copyright © 2017 IDG Communications, Inc.