arrow-right cart chevron-down chevron-left chevron-right chevron-up close menu minus play plus search share user email pinterest facebook instagram snapchat tumblr twitter vimeo youtube subscribe dogecoin dwolla forbrugsforeningen litecoin amazon_payments american_express bitcoin cirrus discover fancy interac jcb master paypal stripe visa diners_club dankort maestro trash

Shopping Cart



Moonshot for Life

Can artificial intelligence bring down cancer deaths?


The death rate from cancer in the US has dropped 27% over the past 25 years, according to annual statistics reporting from the American Cancer Society. This decline translates to about 1.5% per year and more than 2.6 million deaths avoided between 1991 and 2016. Still, in 2019 there will be an estimated 1,762,450 new cancer cases diagnosed and 606,880 cancer deaths in the United States alone. We are interested to find out about new strategies to bring this still huge number down.


Can artificial intelligence bring down cancer deaths?


The death rate from cancer in the US has dropped 27% over the past 25 years, according to annual statistics reporting from the American Cancer Society. This decline translates to about 1.5% per year and more than 2.6 million deaths avoided between 1991 and 2016. Still, in 2019 there will be an estimated 1,762,450 new cancer cases diagnosed and 606,880 cancer deaths in the United States alone. We are interested to find out about new strategies to bring this still huge number down.

Cancer mortality down to 0%

“The numbers are pretty impressive, but my point of view is how can we improve it further, how can we bring cancer mortality down to 0%?”, says Rahul Remanan, Chief Imagination Officer, Ekaveda Corporation. “I am interested to shorten the lead time to diagnosis by using artificial intelligence and deep learning.”

A big hidden AI healthcare community in New York

Rahul is co-leading the New York Healthcare Artificial Intelligence Society that fosters healthcare applications of AI. “There are around 25,000 data scientists in the city. 40% of them have a relation to healthcare. It’s a very large community. New York has 3 major healthcare systems (Presbyterian affiliated to Columbia and Cornell, NYU and Mount Sinai – they are the big hospitals in the city). All of them are working on some form of data science or machine learning projects involving AI”, explains Rahul.

AI-enabled workflow

Rahul and his team are working on visualization techniques and screening algorithms that in the future could prioritize the many images pathologists need to review. “From the enormous amount of images a pathologist has to review during the day, we want to be able to flag those 2 images that are more likely to show e.g. breast cancer characteristics”, explains Rahul. “An AI-enabled workflow, can drastically improve the lead time to diagnosis.” 

Can we trust the algorithm?

In healthcare the stakes are tremendously high. One of the big issues in implementing AI in healthcare is trusting a very complex system and measuring that trust. “Today why an algorithm makes specific conclusions is still vague. Is it because of random chance that this prediction happens and what if we change some parameters, does the prediction remain the same?” According to Rahul one of the real-world adoption issues with artificial intelligence and particularly deep learning is its inability to properly quantify and address uncertainties. “Physicians need to know how certain an algorithm is. If a physician doesn’t know the level of uncertainty, the life of a patient may be at stake. What I am doing in that field is extending this whole idea of what is called deep Bayesian networks. It’s a promising and powerful idea to make deep learning deal with the uncertainties of the real world”, explains Rahul.

Baeysian networks

The original work on obtaining uncertainty in deep learning was done by Yarin Gal, Associate Professor in Machine Learning at the University of Oxford. “His work allows users to classify how much of uncertainty is in a model. It’s a brilliant solution in a sense that it’s one the most practical and scalable solutions of implementing deep Baeysian networks. Everything is going in the direction of Bayesian networks. It addresses one of the fundamental flaws in deep neural networks.  Uncertainty is part everything we do, especially in healthcare. I am trying to combine deep learning with uncertainty predictions. There are still many challenges to solve.” 

Common sense

Artificial intelligence, machine learning and deep learning are pervasive now. It’s all around us. “We have all these incredible tools that are available to do powerful computations, to go through a huge number of images, rank them in a sort of order and rank the probabilities. This can be done at scale very fast, but the key is how do we make it much better, absolutely trustworthy and combine the machine with the human interpreter who adds a layer of common sense”, concludes Rahul.