Computer Vision and the fight against Cancer

Pablo Aguirre Solana
Mar 1, 2023
5 min read

Dedicated to Susan Sly one of my greatest mentors in AI

Last summer I met distinguished professor Regina Barzilay at the MIT in the glorious Eastman Hall, through a series of lectures for the program in Machine Learning and Artificial Intelligence. During an entire week, I had the privilege to listen and to learn from her things that ranged from regularization techniques, loss functions and also to the new challenges that AI presents for the medical community. She had such passion and knowledge about the topics she taught. She made the highly abstract mathematical concepts that take years to master tangible, easy and intuitive, which is a capacity that only extremely intelligent people have. But aside from her brilliancy and meteoric personality, what struck me the most about her was the relevance and reverberations her vast research has made in Natural Language Processing and Computer Vision. Research that has been novel in the field and I hope within a short period a significant change for many.

One of the most startling pieces of investigation Professor Barzilay has conducted in Computer Vision and that has a potential to revolutionize medicine and human lives is breast cancer detection. A victim of breast cancer herself having suffered personally the toll of the disease, Barzilay did research on developing algorithms that can learn from data to improve models of disease progression, prevent overtreatment and narrow down to the cure.

Along with her colleague Dr. Connie Lehman, Professor Barzilay collected over 200,000 mammograms at the Massachusetts General Hospital of people who would and would not develop cancer and fed them to train an algorithm called “Mirai” that would scan those mammograms and make a prediction regarding the incidence of breast cancer, with an average accuracy of about 76 of 100 cases, extending these results to several countries in which the algorithm has been tested in 62,000 patients in seven hospitals from–Sweden, Israel, Taiwan, Brazil and the United States-.

The monumental implication of the above is that a whole new world of testing and preventions could allow patients to avoid aggressive treatments before time, filtering those patients who need follow up screening from who don’t, and also this could drive the medical community to have a tool of prevention that can change the dynamics of how breast cancer is being treated and diagnosed currently, increasing the chances for many people to survive because of early prevention.

Unfortunately, it is not all a bed of roses, according to different sources that I read for this article, there might be a staunch reluctance from the Oncologist and Radiologists community to accept “Mirai” as a valid tool for diagnosis and prevention, first for the understandable anxieties that entails AI taking their jobs, and second because their prognosis now would have to compete with a machine, which can grow a lot of tensions and debates in their communities. Not to mention the insurance companies with their intricate policies as whether they will cover for a non-human AI prevention tool, and how. Finally, that Barzilay and her team made the algorithm open-source so any hospital or institution could use it, can be very unappealing for many corporations and big pharma, which can represent limits for its diffusion and universality.

From my perspective, this posits a twofold opportunity, one for business innovation, the other for social care. Both come as a response in which the pace of science and in its AI advancements, such as “Mirai”, outpace the institutional and legal frameworks of medical and insurance communities. Not that these come as irrelevant and one must disregard them, but perhaps it is time to think in alternative ways by which people can benefit and access “Mirai” seamlessly, with little institutional, legal and economic friction.

Thus, as an opportunity for business innovation, retail can be an answer to outpace the conundrums of medical and insurance realities and to reach out to a vast majority of women from different, racial, social and economic backgrounds. This is what I think.

We all have been to an optometrist in a mall or even in a Walmart or Costco. I know, you might think by now, an optometrist is not even close to diagnosis cancer! Yes, you are right, but I am not thinking in the comparability about diagnosis or medical implications, I’m thinking only about the logistics, operations and setup of an optometrist clinic in retail facilities, which can be emulated in all its characteristics but for a space that can be equipped with a mammography machine that under the supervision of a certified physician takes a mammography and runs it through “Mirai” algorithm that delivers a prognosis, which can be flagged in terms of probabilities (prone or not prone to cancer) so that the patient rapidly could seek further advice, treatment and diagnosis from their respective physicians and other clinics. This would in fact shorten the time for women to have an early prognosis relatively seamlessly and serve as a pre-filter to women that need further observation. And second, it would widen geographically the sample of women that could have mammograms before age 40. Last, in terms of costs, retail chains such as Walmart or Costco could afford the investment initially of setting up these cancer-detection units and offer some type of payment installments within their credit card system and be profitable within a few years of operations.

As an opportunity for social care in poor countries such as Mexico and Turkey with a large proportion of their population without access to healthcare, “Mirai” could shorten the gap between diagnosis and treatment as well, increasing the chances of early prevention and reaching out samples of women that are excluded from healthcare. According to Statista, 35.7% of the population in Mexico is considered vulnerable due to lack of access to health services in 2020 and in Turkey according to a survey conducted in 2021 among the population, 45% of interviews declared that one of the biggest problems facing the health care system is access to treatments and long waiting times, *

Aware of the incidence of breast cancer as a public health issue, since 2012 Mexico’s health ministry ideated a solution to increase the coverage of breast cancer detection through mobile units equipped with a mammography machine that could reach villages, towns and poor neighborhoods with little or no health coverage. The problem that this program faced was that, once a mammography took place, women had to go to a health clinic necessarily (not accessible for many) for an interpretation and prognosis from a physician reducing considerably early detection, treatment possibilities and follow up.

Now, let’s image for a moment that these units equipped with “Mirai” algorithm can provide straight forward prognosis with high accuracy, there will still be the same problems regarding health accessibility and medical accompaniment sure, but at least women in poor villages, towns and neighborhoods from Mexico and Turkey could have more certainty and knowledge to where they stand regarding their probabilities of being prone to breast cancer or not, thus, increasing considerably early detection, treatment possibilities and follow up, despite the precarities of its health system. All thanks to a practical deployment of “Mirai”.

It might sound remarkably easy and probably naïve some might think, these alternatives I am suggesting to you, but I am convinced that these conversations need to happen with retail corporations, health ministries and other profit or non-profit institutions that can wage heavily on this brilliant piece of technological innovation, there is no time to waste here, I suppose….

On a final note, in Spanish “Mirai” can be taken as a pun or wordplay because; “mira” means look and “ai” with h in the middle and tilde, “ahí” means there. So, putting it altogether it sounds like the word means “mira-ahí” (mee-rah-ah-ee) which in English means: “look there”.