With a rate of 440.5 per 100,000 men and women per year, in 2024, cancer caused over 600,000 deaths in the USA alone. The earlier the symptoms are identified, the higher the patient’s chances of survival. To achieve this goal and maximize treatment outcomes, HistAI, a company behind an all-in-one digital pathology platform, introduces a new way to educate specialists and help them cure patients worldwide.
HistAI represents the open-source Foundation Vision Models, Hibou-L and Hibou-B, which perform multiple tasks, including tumor classification and region of interest (ROI) and cell segmentation. The models empower oncologists to provide patients with accurate diagnostics and efficient treatment planning. “We believe that our top-performing foundation model under the Apache 2.0 license will significantly accelerate innovation in computational pathology, leading to tangible benefits for humanity as a whole,” emphasizes Alex Pchelnikov, CEO.
Combination of Technology and Research
Alex Pchelnikov is not the only visionary in the team. Much of HistAI’s success can be attributed to its COO, Katherine Ivanova. Katherine transitioned from a career in marketing and production (Danone, Snapchat, etc.) to becoming a pioneer in AI-driven healthcare together with Alex. The two met while working at Snapchat, where Katherine was responsible for overseeing multi-million dollar influencer programs. Highly interested in biology and medicine, she joined Pchelnikov in the creation of Snapmole, an AI-based app that classifies over 300 skin conditions.
This project inspired the two to continue their journey in shaping solutions for oncology identification and treatment. Again, Katherine played a crucial role in organizing vital partnerships that enabled HistAI to gather the world’s second-largest pathology dataset by collaborating with top researchers, pathologists, and laboratories. Thus this step was instrumental in developing the foundation models.
“The use of foundation models in cancer research highlights the transformative potential of collaborative AI-driven diagnostics. I believe that by involving a diverse range of researchers in working on these advanced models, we significantly increase the chances of cancer diagnosis and treatment breakthroughs. Meanwhile, our commitment to open access ensures that these cutting-edge tools are available to the research community, empowering scientists and clinicians to drive innovation and achieve groundbreaking results in oncology,” emphasizes Katherine Ivanova.
To reach this accessibility and enable researchers to collaborate without limitations, HistAI had to overcome challenges related to the sheer scale and diversity of the dataset needed to train models. “With over 1.2 billion pathology image patches derived from over 1 million whole slide images, managing and processing this data required meticulous data curation,” claims Dmitry Nechaev, the Chief of Science.
Also, the DINOv2 framework had to be leveraged to train Hibou-B and Hibou-L. This self-supervised learning approach helped to utilize the vast amount of unannotated data at the company’s disposal.
As a result of the work done, the models can significantly enhance the development of downstream algorithms. Just consider that Hibou-L is recognized for its pathology performance, which includes superior capabilities in tumor classification, region of interest (ROI) segmentation, and cell segmentation. Hibou-B, trained using the same extensive dataset, ensures it performs effectively in various complex tasks within the field.
Importantly, HistAI empowers researchers cost-effectively — using the models is times less expensive than creating custom tools. Thus, the company makes the treatment software more accessible to everyone. “We are especially proud that “Open Access to AI Innovations” fits as many researchers and developers as possible, either for free or at a nominal cost,” says Alex Pchelnikov.
Meanwhile, HistAI is not only providing access to these models but is also actively seeking collaboration with academic institutions and research centers. By offering preferential access to its Foundation Vision Models, HistAI aims to foster a community of innovation focused on advancing cancer research and diagnostics. This collaborative effort is expected to yield new insights into tumor behavior, treatment response, and personalized medicine approaches.
The Future of Open Source in Medical AI
As more organizations like HistAI introduce open-source tools and freely provide them, researchers are empowered to explore new frontiers collaboratively and inclusively. Katherine Ivanova, who didn’t just co-found HistAI to enable widespread analysis of biopsy images, but boosted the development of cancer diagnostics and dermatology technologies, believes that her team’s work can set new industry standards and make AI technologies more accessible, enhancing their speed and accuracy. “Our work will push the boundaries of what can be achieved in cancer diagnostics. Thus, patient care will be transformed,” Katherine Ivanova says.