In recent years, human creativity and scientific advancements have given rise to innovative drugs that offer hope for relief and even recovery to millions worldwide battling previously untreatable forms of cancer.
Traditionally, drug discovery involved identifying active ingredients through historical knowledge of folk medicine—aspirin is one such example—or through serendipity, as seen with penicillin, which was discovered by Alexander Fleming in 1928 when he noticed mold on a petri dish. A century later, the process of modern drug discovery has evolved into a sprawling industry, with pharmaceutical companies investing billions of dollars, often with governmental support—as was the case with the first COVID-19 vaccines. However, the journey remains challenging and plagued by inefficiencies.
What if we could improve this process? At Imagene AI, an Israeli startup, technology is being harnessed to transform drug discovery. Founded and led by Dean Bitan, the company aims to enhance oncological care by integrating artificial intelligence (AI) capabilities from Oracle into cancer treatment.
Bitan, a lifelong technologist who began studying computer science at just 15, became interested in cancer research after losing a close relative to the disease. During a conversation with Computer Weekly at Oracle Cloud World in Las Vegas, he explained, “Initially, my focus was on technology and entrepreneurship, without a direct connection to cancer. However, after experiencing personal loss, I educated myself about the disease and recognized significant gaps in treatment options. Many cancer types still lack sufficient drug therapies available for patients.”
This realization propelled the creation of Imagene, with a mission to assist physicians in navigating diagnostics and treatment decisions more effectively.
Can AI Drive Change?
Bitan was curious about the potential of AI early in Imagene’s research phase. “The hypothesis was that we could apply technology to this field. But as an engineer communicating with physicians, I encountered a different perspective,” he notes. “We brainstormed together about how we could tackle the challenges. Some physicians suggested that they could discern patterns in biopsy images based on their intuition and experience, identifying biomarkers that might indicate a patient’s response to treatment. This intuitive process can be enhanced with AI. By applying AI to biopsy images, we could uncover more insights.”
Utilizing a dataset of 630,000 anonymized biopsy images from various cancer types, Bitan and his team created a foundational model aimed at delivering what they call “oncology intelligence.” Known as CanvOI, this model, consisting of 1.1 billion parameters, captures intricate features and patterns in biopsy images that might be elusive to the unaided eye, thereby augmenting researchers’ understanding of various pathological traits.
The overarching goal is to establish a “robust vision data backbone” that can support the development of new applications in oncology research, not only to identify potential new drugs but also to predict how patients with diverse biomarkers may respond, ultimately allowing physicians to tailor cancer care to each patient’s unique profile.
CanvOI Powered by OCI
Imagene’s CanvOI operates on Oracle Cloud Infrastructure (OCI), benefiting from OCI’s AI Infrastructure and Supercluster, which can currently scale to tens of thousands of GPUs for AI tasks, with a planned future capacity exceeding 130,000. Bitan emphasizes that by leveraging Oracle’s computational power alongside Imagene’s innovative approach to digital pathology, CanvOI is already achieving outstanding performance, even with minimally labeled data.
“Oracle uniquely supports AI companies in overcoming these challenges. Their strategic partnership with Nvidia enhances our GPU availability, which translates to increased computing power, essential for our progress,” Bitan explains. “We need partners like Oracle as we navigate this complex journey and achieve critical milestones. Just as we’ve seen rapid advancements with technologies like ChatGPT, we aspire to similar breakthroughs in oncology, particularly with biopsy images, as we continue to expand our models.”
CanvOI is also the foundation of Imagene’s new OISuite, a platform designed to assist researchers and diagnostic developers in addressing a wide range of research questions. This platform alleviates the need for in-depth AI expertise and simplifies data acquisition, promoting innovation while adhering to rigorous data privacy and security standards. All data employed by Imagene’s systems is de-identified prior to use, ensuring compliance with GDPR, HIPAA, and other regulations, alongside implementing zero-trust approaches and regular vulnerability assessments.
Looking Ahead
Imagene is collaborating with medical institutions across multiple countries, including prestigious research centers like Johns Hopkins in Baltimore and Northwestern University in Chicago, as well as prominent cancer facilities in Brazil and Israel. Bitan aims to expand partnerships further, incorporating private labs and reference facilities to maximize cancer research efforts. “In cancer research, we cannot afford to do less than our best; every day matters,” he asserts.
Bitan envisions applying the technology developed at Imagene to other areas of medical research, such as COVID-19 and HIV/AIDS. “As we progress, we will expand our models to encompass various modalities—not just biopsy images but potentially radiology, MRIs, X-rays, microbiomes, and even genome sequencing. This will enable us to address increasingly complex healthcare questions,” he concludes with optimism.