The use of artificial intelligence in the healthcare industry is difficult; this is much more evident in oncology, where the stakes are especially high. Biotechnology startup Valar Laboratories goals high but starts small, using a tool that accurately predicts specific treatment outcomes, potentially saving patients useful time. $22 million was raised for the development of recent cancers and therapies.
Every cancer is different, but many have established best practices that have been refined over years of testing. But sometimes meaning going through a given treatment regimen for months to see if it really works.
Bladder cancer is one of them, Valar’s co-founders explained to TechCrunch. A typical first treatment advisable by oncologists, called BCG therapy, has the potential to work like a coin flip – and that is actually quite good! But would not or not it’s nice to not have to flip a coin instantly? This is the problem Valar is trying to unravel.
CEO Anirudh Joshi said the team got here together at Stanford University, where they were researching artificial intelligence support for clinical decision-making. In other words, helping each patients and doctors determine which treatment path to decide on, whether or not it’s one of two or a dozen.
“We learned that the treatment plan for most cancer patients today is really unclear,” Joshi said. “They have options, but it’s hard to say which of them will work well – you only have to try various things. Our goal was to make an informed decision. In the treatment of bladder cancer, only one in two patients responds to straightforward treatment. If we knew which patient was which, we would not have to waste a yr of therapy on something that does not work.”
The first test they developed, called Vesta, focuses on this specific situation. And this is not some theoretical software solution: the team worked with greater than a dozen medical centers around the world to review greater than 1,000 patients and discover what exactly makes them reply to specific therapies.
This process consists of two components: first, a visual artificial intelligence (or computer vision model) trained on hundreds of histological images of cancer patients. These thin slices of affected tissue are increasingly scanned and inspected by experts, although the process will be somewhat approximate.
“This very high-resolution image says a lot about what is happening at the cellular level of the tumor,” explained CTO Viswesh Krishna. “We run our models on this image to extract a very large variety of features, just like a genome panel; we generate hundreds of histological reads [i.e. important image features]and select the most significant ones that pathologists can listen to but cannot quantify. They may see that they are different, but they can not measure the differences between them.
Joshi was careful so as to add that they do not aim to interchange the pathologist but to empower him. You can think of it as a smart microscope that helps the expert make accurate measurements of things reminiscent of cell damage, immune response, and other structures that indicate disease progression or inhibition.
“At the end of the day, the doctor is always in the driver’s seat. It’s just more data and they like it. And doing tests like this provides a grounding outside perspective, and patients really enjoy it,” Joshi said.
The team noted that the imaging component was trained on tons of information and may very well be generalized across multiple domains and cancers; counting lymphocytes in breast cancer tissue is largely the same task as in skin cancer tissue. However, what this counts, or any of the other measurable biomarkers the model can discover, that tells us how likely a patient is to answer treatment is much more limited to specific conditions.
Therefore, the second element of the Valara system is what really must be incorporated in the specific clinical situation. To this end, the company has shown that in the specific case of bladder cancer and standard treatment regimens, its test is a much more accurate predictor of success than any other available indicator.
Risk aspects reminiscent of age, health history, smoking, etc. vary in how they predict specific treatment outcomes, but they are “very rough,” Joshi noted. Valar claims that their AI models “outperform all of these variables [in predictive power]and are independent of them” – meaning they will be used in addition to straightforward risk aspects, not only as an alternative of them.
They also noted that it is necessary that the results are interpretable: the very last thing doctors and patients need is a black box. So if he says that the patient will respond well, this is supported by the statement that “because his immune system does A and his testicles do B, etc.”
Founded in 2021, the company has put a lot of effort into creating an image model and the first clinical model for the mentioned BCG therapy in bladder cancer patients. As Valar noted in a recent announcementthe test identifies individuals who are three times more more likely to not reply to BCG, which suggests (at the discretion of the care team) that trying something else is probably a higher move. If it saves even one month of wasted effort, it may very well be life-changing for some.
Anyone who has survived cancer treatment can inform you that not only is each day of treatment incredibly useful, but self-confidence is also hard to achieve. Valar may not provide reassurance (nearly inconceivable in oncology), but it could possibly be a powerful shot into the spines of caregivers.
Coinciding with the upcoming launch of its first product, Valar has accomplished a $22 million Series A round led by DCVC and Andreessen Horowitz, with participation from Pear VC.
“The fundraising came at the perfect time,” Joshi said. “We were able to complete this validation, and now this funding will help commercialize Vesta while we begin to expand into other types of cancer.”
The founders expressed hope for continued growth, using a industrial laboratory model just like the genomic testing used in recent years. Chief operating officer Damir Vrabac said: “It is very similar to other tests that have come before us, it does not cause any friction in the healthcare system.” This will hopefully allow them to bill insurers and ultimately reduce the cost of care altogether by avoiding unnecessary and ineffective treatment.