There are currently only 200 radiologists serving Kenya’s entire population of 55 million people. Every year, 5.5 million women among this population are at risk of late breast cancer diagnosis, where they find out they have the illness at stage four or five when it is much harder to treat. When breast cancer is found earlier, however, the chance of survival is about 80%. In response to this challenge and opportunity, Vectorgram exists in order to accelerate the pace and scale at which radiologists can screen patients, enabling them to discover cancer much earlier and giving patients the chance for early intervention.
While Vectorgram’s core product uses AI to help accelerate the screening process, we are also cognizant of the fact that we are not operating in a vacuum and there are many social issues that impact whether people seek out preventive health screenings. Sometimes people don't even know that those services are available near where they live. So even though we will be tackling this problem from the technical side, increasing the capacity and scale at which radiologists can screen more patients, we have also partnered with local nonprofit, community based, social impact organizations, like the Africa Cancer Foundation, that run outreach programs in areas where patients are most likely to miss out on vital healthcare information. We are preparing radiologists to be able to effectively handle the increased demand that these community organisations will ideally create.
The most recent milestone that we have celebrated hitting was publishing our machine learning research in Ghana at the Deep Learning Indaba, which is a continental research conference where the best minds in machine learning, AI, and deep learning meet and then disseminate research and run mentorship sessions. Vectorgram published our research there and we won a couple of awards, including a best poster award for a poster that we presented on the capabilities of our diagnosis platform. This was especially interesting because whenever we publish research that builds on the best in class globally, we call back to global actors like Google or Microsoft that have the resources to explore these problem areas to the fullest.
It was a great milestone for us, especially this month, because we exceeded the benchmarks created by Google for medical imaging classification in breast cancer. It felt great for an organisation like ours, with our limited resources, to even meet the authors who did the research at Google and get their feedback, support and encouragement. This is a pertinent problem that researchers across the globe are working on. Getting to publish our research at that stage, getting all that feedback and credibility as well as just the momentum to keep working on the problem and to target a wider scale was really important and inspiring.
I think any machine learning company works with a bottleneck when it comes to data, but it's especially constraining for us at this point in time, because we are working on clinical validation in Kenya and the datasets that we get from local screening facilities are all physical. That means we take a lot of time to clean the data and preprocess it before we can use it in our machine learning workflows to validate our capabilities in a local context. The first few months we were working on Vectorgram, we went to do our first customer interviews at the hospitals in Kenya and we saw all of the physical films in very large rooms. We decided to pivot and work on building the entire infrastructure set, not just AI powered mammography tools, but also electronic medical records digitisation software because that will help bring all of that data in many facilities that are currently physical. After spending a few weeks on it, however, we felt it best to roll back the pivot to just focus on AI-powered mammography because of how much impact we could achieve on this specific aspect of such a large, interconnected problem. Digitization is still a problem that keeps looming over us because for most Kenyan facilities that we want to partner with to get their historical data, we first have to digitise their records. That's why we're also trying to get third party data licensing partners who would have already digitised, pre-processed and cleaned the data. With more funding, we could make our search more expansive and more extensive with a wider area of reach and bigger datasets to acquire. But right now, the problem is in finding those datasets in the first place. So we are looking to get dataset processing partners who we can team up with to get this data and help increase the scale of our capabilities even further.
Sometimes the problem that you want to solve seems very large and amorphous and you might find yourself working with vague targets at the early stages. Your momentum might feel unfocused, like you’re not heading in a specific direction. I think a lesson that I have picked up so far to help address this is to think of your startup’s absolute end goal. For us, it's helping to save the lives of 70 million African women through early breast cancer diagnosis that is powered by AI tools at Vectogram. So think of that big hairy end goal and then translate that into mid-term action points, and then translate that down further into day-to-day things that you can then build on on a weekly, monthly basis. Then you hope that, with enough execution and consistency, you can end up hitting that big end goal. The vagueness of targets will make any early-stage startup move slower than you should, and speed and momentum are everything for startups in a stage like ours. Translating your end goals into a roadmap and making your daily goals become based on that roadmap is a useful rule of thumb for focusing your momentum in a specific direction.
The partners that Villgro has brought together are pretty remarkable. For many early stage African startups like ours, before you would get exposure to all of these partners at the same time, you would need to go through a couple of incubation programs and have more traction and execution under your belt. Villgro has done the work of several incubation programs in one by bringing all of these partners together. For example, NVIDIA has been really helpful to us as an AI company because they provide us with credits to help train our models on larger scales. We wouldn't be operating at such scales if it wasn't for this collective of partners that Villgro brought together for the AI4H program.
Beyond the resources that all the partners are able to provide, the main benefit we’ve received from Villgro is consistency. We meet our advisors and mentors from Villgro every two weeks. It definitely helps us not only to set actionable and realistic goals, but to be consistent with them because we have experienced mentors who can hold us accountable and step in when they need to help direct our energy and our attention into the right path. This has provided the fuel that Vectorgram needed at this stage to keep on executing in the right direction and to not stop even if the challenges outweigh the positives at certain parts of the journey.