This summer I worked full-time developing machine learning algorithms for breast cancer detection at a startup called Higia Technologies.
Higia was founded almost two years ago by a Mexican team led by Julián Ríos, its current CEO. Julián’s mother has survived breast cancer not once, but twice. Her tumor grew from being undetectable by a mammogram to the size of a golf ball and being a serious threat. This led Julián to recognize the importance of creating a method that could aid the early detection of breast cancer.
Since many women in Mexico and other developing countries are not able to get a mammogram screening regularly, he wanted to establish a method that relied on artificial intelligence — not a specialist’s judgement — to screen for breast cancer in communities with little or no access to doctors. Though this mechanism would not necessarily replace the mammogram, it would help identify lesions in need of a specialist’s attention.
And so, Higia Technologies was born.
I contacted the Higia team while I was studying at Minerva in Buenos Aires last semester and expressed my interest in contributing to their mission. At the time, they needed to answer a relatively straightforward question: can breast cancer be detected using temperature data from specific locations in the breast? Once I was brought on board, my job was to find out. I was originally tasked with using machine learning and data from breast thermal imaging studies to create a proof of concept.
When I started, I had broad ideas about what I needed to do. I had obtained some basic knowledge about machine learning from two classes I had taken at Minerva: Mathematical and Computational Models and Information-Based Decisions. In the latter class I learned about the potential pitfalls of building models and drawing conclusions out of data resulting from observational studies, which proved quite useful when we began designing clinical trials.
Despite this exposure, my experience as a developer was limited. To be successful, I also needed to acquire a large amount of technical skills in a short period of time. Excited by this challenge, I learned to do things that I had only read about, such as creating and tuning Neural Networks. My ability to quickly learn and apply these newfound technical skills to my work proved extremely useful. When things didn’t work or I wasn’t sure how to do something, I remembered one of my mantras: chances are someone has solved this problem before… and I can probably find their answer online.
Being able to identify exactly what I needed to learn and how to learn it quickly are skills that I often practice at Minerva. My classmates and I studied the science of learning in our first year. We are often expected to research and acquire skills (such as programming) o n our own, in order to use them to understand broader or more complex subject matter. For example, learning to use the R programming language in order to use data from observational studies to assess the effectiveness of a social program. Getting accustomed to this was exceptionally useful, though it has taken practice — and many hours of frustration — to finally feel comfortable.
I wondered whether I deserved the trust and the responsibilities that came with it. Eventually, I came to the conclusion that what mattered most was not whether I deserved the trust or not, but what I did with it now that I had it.
With my new knowledge and skills, soon I started taking on more responsibilities at Higia, which eventually led me to holding a more important position within the company. We needed to prepare for clinical trials that would test the cancer imaging device and its algorithms. My experience last year with study design at Audible Inc., where I helped design a randomized controlled trial (RCT), and my ability to understand the data that we needed in order to obtain valid results, led to me being appointed the intermediary between Higia and the medical professionals who helped us design the clinical trials protocol.
This was around the time I began having a recurring thought: what Higia had acquired up to this point was trust. Our funding, our mentorship, our prizes, they had all been granted by members of society who trusted us to use these assets to create value and contribute to the fight against breast cancer. When I was given a more important role in the company, I felt a larger part of that trust being placed upon me. This was both exciting and terrifying. I wondered whether I deserved that trust and the responsibilities that came with it. Eventually, I came to the conclusion that what mattered most was not whether I deserved the trust or not, but what I did with it now that I had it.
As a starting point, I applied concepts I learned at Minerva and Audible to the projects I was tasked with. For example, when calculating sample sizes for the clinical trials, I knew how to approach the task so that we would obtain significant results in a study. I also knew to look for pitfalls, such as selection bias (when the individuals in your study are not representative of your target population, because of the manner in which you selected them).
Here’s a simple example problem: you want a model that can classify a set of breasts as either being affected by cancer or not. Hence, you create a study in which women with and without breast cancer use your device and then construct the model using that data. Now, imagine that most of your cancer cases are late stage cancer. Will you end up with a model that gets really good results on your test, but that fails to identify women with early stage breast cancer? What about selection bias? Are women with breast cancer in your sample considerably older than those without cancer? Will you end up with a model that seems to do fine in your study, but that is actually unable to detect breast cancer in young women?
My limited knowledge and experience was of course not enough for me to single-handedly design the whole protocol for the study, but I was able to understand the problem and communicate it properly with our medical advisors. This proved valuable because these individuals helped us ensure that Higia’s protocol satisfied all requirements in order to be approved by an ethics committee so we could actually carry out the tests. In fact, interacting with stakeholders, though not something I had necessarily signed up for, ended up being quite an interesting part of my job.
Medical professionals, hospital directors, government officials, and investors were all part of a complex network — each with their own motivations and concerns. No one was particularly interested in hearing me explain the math behind our algorithms. Imagine sitting at a table, across from people whose decisions will hugely impact the outcome of all your hard work (and that of others). Then, imagine that some of them may be biased to doubt your competence because of your young age. Good luck.
In my first year at Minerva, I learned about effective negotiation in a class called Complex Systems, where I learned to recognize how the agents in a social system, their motivations, and the way in which the system may behave based on the characteristics of their components. Even though it would be much more difficult to conduct such an analysis at the aforementioned table, as opposed to sitting comfortably at my desk finishing an assignment, the courses I took during my first year provided tools and habits that proved to be useful in navigating through such situations.
In the example I described before, identifying and understanding the social system was important, but after that I still needed to actually effectively communicate with these decision makers. To do this, I had to constantly think about who my audience was, and tailor my communication accordingly. What should I emphasize? Should I be as concise as possible or repeat certain ideas? Would it be good to explain a particular idea with a personal story? The answers to these questions depended on who I needed to convey information to, and under which circumstances.
With clinical trials about to start this year, Higia is in a much better position now than it was at the beginning of the summer. A large part of this change is due to the hard work of my co-workers and the invaluable help of our mentors and allies, and some of it is related to the work I did during the summer. I am now continuing on with Higia as Chief Artificial Intelligence Officer while I carry out my third year at Minerva from Seoul.
I have an unconventional job. It is quite interdisciplinary, full of challenges, and has led to immense satisfaction. I think studying at an unconventional university program, where I’m learning to effectively apply skills and concepts to affect real change, has contributed to making me a good fit for it.