Diagnosing cancer from biopsies in minutes, not weeks

After a computer scientist’s mom got breast cancer, he started a company using AI to find biomarkers in biopsy images and give personalized guidance.

When Dean Bitan’s mother was diagnosed with Stage 4 breast cancer, the Israeli entrepreneur felt paralyzed, The Times of Israel reports.

“But then my best friend told me, ‘Put your emotions aside and act like this is another project of yours,’” Bitan tells ISRAEL21c.

“Standing in the frontlines with my mother, I learned a lot about the gaps in terms of treatment. I knew we can and should do better. So that’s what I decided I was going to do.”

Bitan’s mother, sadly, passed away two years ago. But the result of Bitan’s fortitude would make any Jewish mother proud.

Imagene, the startup Bitan founded in 2020 with Jonathan Zalach and Shahar Porat, has developed technology that can conduct a full molecular analysis on a digitized biopsy image in real time and return a result in minutes, instead of the current situation of a few weeks.

How does Imagene do it? AI, of course.

Imagene’s artificial intelligence is intended to “democratize” personalized medicine, Bitan explains, by allowing oncologists to obtain an accurate diagnosis and then initiate a precisely targeted therapy plan far sooner than ever before.

Oracle founder Larry Ellison led a $21.5 million Series A investment in the company earlier this year.

Ellison’s involvement is not out of character — he and Dr. David Agus founded the Institute for Transformative Medicine in 2016 with a mission to drive interdisciplinary patient-centred research. Agus was so impressed by Imagene’s tech that he sits on the company’s board.

Imagene’s AI searches for biomarkers in a digitized biopsy image to assist pathologists in classifying a tumour as cancerous or not and, if so, what kind.

It’s a big step up from placing a tissue slide under a microscope and examining it manually.

Imagene’s software can be integrated into the pathologist’s current workflow. “We’re agnostic to their settings,” Bitan says. “We don’t want to add any effort to these physicians.”

The workflow with Imagene inside looks like this:

An Imagene report summarizes the findings and points out actionable biomarkers, “including those that match targeted therapies and relevant clinical trials,” Bitan says.

Imagene has developed models that can check for 28 biomarkers in eight organs including the lungs, thyroid, breasts and ovaries, with more on the way. “On average, our specificity and sensitivity are more than 95%,” Bitan notes.

“Today, only around 15% of patients will be diagnosed with a biomarker that can lead to a precision medicine therapy. The goal is to push the boundaries of precision medicine, to make it more accessible and available,” Bitan says. 

“Even in the US, there is a huge gap between the quality of diagnoses in an academic centre and a community setting,” Bitan says. “By identifying actionable biomarkers, between 40 and 50% of lung cancer patients should be able to be diagnosed.”

Agus call Imagene’s approach “theragnostics” – a combination of diagnosis and therapeutics.

The first step in Imagene’s analysis is defining what it’s looking at.

“If you show me a glass, that’s the shape of the glass,” Bitan explains. “But what if I turn it? Now I’m changing the shape. AI changed the whole approach by teaching the computer by example.”

Accordingly, Imagene trains its AI to recognize a mutation in a digitized biopsy image by showing it many pictures of biopsied tissue at different angles.

Imagene is also addressing another cancer diagnosis challenge: It can be difficult to collect enough tissue to do a traditional microscope biopsy. This is the case with 20% of lung biopsies, Bitan says. Even worse, many lung cancer patients do not live long enough to benefit from genomic screening, which takes between 14 and 28 days.

These challenges make it tough for pharmaceutical companies to identify and enroll patients in their precision medicine tests. It’s especially true for patients with rare conditions, where 26% of trials are terminated because of low accrual rates.

“We can assist these companies with a super cost-efficient solution that can screen dozens of patients a day without wasting tissue,” Bitan says.

Imagene’s system does not yet have regulatory clearances, Bitan stresses. In anticipation of entering the US market, Imagene has set up a small office in the Philadelphia area.

Bitan’s hope is that within five years, patients will be able to get a “comprehensive diagnosis at the point of care and then immediately discuss personalized treatment options or clinical trials. Just imagine what patients are dealing with when they have to wait for results to come in a month later – and that’s if precision medicine and NGS [next generation screening] is even offered in their area.”

The Tel Aviv-based company, with 25 full-time staffers, has collaborated with Sheba Medical Center and Tel Aviv Sourasky Medical Center in Israel and with several medical facilities and pharmaceutical companies outside Israel whose identities are confidential.

Imagene and Sheba recently published a paper in Modern Pathology describing the accuracy of AI models in detecting cancerous biomarkers in non-small cell lung cancer.

“Imagene’s deep learning algorithms radically streamline cancer diagnosis and targeted therapy, and we are committed to continue and expand this collaboration to cover more cancer types and biomarkers,” said Prof. Iris Barshack, head of the Institute of Pathology at Sheba.

Bitan, a computer scientist by training, earned his undergraduate degree at age 15 and was a computer programmer during his military service. He notes that fast cancer diagnosis is more about engineering than biology.

“When you need so much data to process and answer the question of whether this is cancer or not, that’s an engineering problem,” he says.

“It’s about the way you collect the data and generate different insights out of it. Of course, you also need biologists and physicians in the picture, but at the end of the day, you need engineering and data science.”

Israel, it turns out, has a real advantage in this kind of bioconvergence. It’s a small country where everyone knows everyone and where there is a regular collaboration between universities and startups.

“Add in some Israeli chutzpah – in the positive sense – and that drives many entrepreneurs to believe they can do things differently,” Bitan says.

He tries never to lose sight of the ultimate end user.

“Behind all the conferences and publications and numbers are real patients, and that’s too easy to forget when you’re looking at millions of slides. When we recruit new employees, the most important thing is that they have a strong desire to do good, to make a better future. That’s why we established Imagene,” he says.

“When I was once asked if I had an opportunity to sit with anyone, who would it be, I said, ‘with the first patient where Imagene saved his or her life.’ This is why we do what we do.”