ICRF backs Israeli AI cancer projects with grants up to $500,000

By Sophie Lindqvist

A five-year, $500,000 grant will test AI-assisted treatment selection for pancreatic cancer at Sheba Medical Center, one of three Israeli projects backed by the Israel Cancer Research Fund (ICRF).

Which teams received ICRF support?

• Ron Kimmel, who founded and leads Technion’s Geometric Image Processing Lab, received $180,000 over three years for an AI model that predicts whether breast cancer patients are likely to benefit from chemotherapy.
• Debbie Anaby, founder of Sheba Medical Center’s Breast Cancer Imaging Lab in Tel Aviv, won an ICRF Career Development Award worth $135,000 over three years for AI analysis of breast MRI scans.
• Talia Golan, who leads Sheba’s Pancreatic Cancer Program and Pancreatic Translational Research Lab, received $500,000 over five years for preclinical testing of AI-guided treatments.

ICRF is an international nonprofit based in New York that funds Israeli cancer researchers. Alan Herman, its executive director and a cancer survivor, called BRCA-mutant breast and pancreatic cancers among oncology’s most urgent challenges. He said the grants support computational methods that can open routes unavailable to traditional research and improve earlier detection, disease characterization and treatment selection.

What changes in breast pathology?

Kimmel’s model analyzes routine pathology samples to estimate chemotherapy benefit, rather than identify cancer in undiagnosed patients. “We’re not trying to diagnose if a patient has cancer; these are already confirmed cases of invasive breast cancer,” he said. The intended decision is whether a confirmed patient will receive too much or too little treatment.

His group uses computer vision, machine learning, medical imaging and scientific computation. Kimmel said AI can analyze large clinical-trial datasets more accurately than human observation, but the report gives no benchmark or error rate for that claim.

The proposed workflow starts with basic-stained tumor slides. Doctors traditionally examine a few slices, then run more detailed tests on selected areas; Kimmel’s model scans all basic slides across a tumor to capture features that selective sampling may miss. Kimmel said his team had shown that the broader view can help determine treatment course, without disclosing cohort size or validation data.

ICRF expects the system, which needs only a digital scanner and internet access, to support advanced pathology analysis in clinics with limited medical resources. That remains a projection: the source describes no prospective trial, deployment or patient outcome.

Can breast imaging reduce false positives?

Anaby’s $135,000 award supports image processing, quantitative image analysis and AI that convert breast MRI into objective measurements of subtle tissue patterns. The team, which works with Sheba breast-radiology head Miri Sklair-Levy, aims to detect very small lesions and separate malignant tumors, benign lesions and normal tissue.

Anaby described MRI as the most sensitive imaging tool for breast cancer detection, especially among BRCA mutation carriers and women with a family history. Its sensitivity also produces benign findings, which leads to additional biopsies and patient anxiety, she said. The program’s operational goal is fewer unnecessary biopsies and less anxiety.

Sheba’s protocol calls for annual MRI among high-risk women from age 25 or 30 through their early 70s. Anaby said 60% to 70% of these MRIs produce false positives. Her grant first supported model development using Israeli patient data, rather than a completed clinical trial.

The next test is external validation in Black, Hispanic and Asian women. Anaby said the Israeli cohort contains a high proportion of Ashkenazi Jewish women with BRCA mutations, so broader data are needed to determine whether the imaging signatures generalize.

How will pancreatic cancer treatment be stratified?

Golan, who was born and raised in South Africa and has worked at Sheba since 2004, has a special interest in inherited BRCA1 and BRCA2 mutations. She said these patients generally respond well to platinum-based chemotherapy and PARP inhibitors, although the source provides no response rates.

Her grant will test new biological agents in preclinical settings using patient-derived models that mimic a patient’s clinical journey. The experiments will examine treatment resistance and combination regimens. Golan expects AI to identify patient subgroups for intensive treatment while allowing others to receive palliative care.

Israel diagnoses about 900 to 1,000 pancreatic cancer cases each year, and roughly 380 patients receive treatment in Golan’s clinic. Globally, around 5% of pancreatic cancer patients have BRCA mutations, with a higher prevalence in Israel.

One in 40 Ashkenazi Jews carries a BRCA1 or BRCA2 mutation, 10 to 20 times the general-population rate, according to the report. Carriers face pancreatic cancer risk two to six times higher than people without those mutations.

Golan cited recent FDA approvals for new drugs but called pancreatic cancer aggressive and its five-year survival low. She also linked rising incidence to preventable factors including obesity and smoking. She described major gains in AI-guided pancreatic research over the previous year or two, including systems that integrate clinical and other patient data to compare likely treatment trajectories across large cohorts.

Her caution is practical: “You need to learn to use AI, but it’s like a monster.” She said users need deep subject knowledge to identify inaccurate predictions and must control the prompts that guide the model. She also said she remains on a learning curve herself.

What must the projects demonstrate next?

The source discloses no cohort size, train-test split, sensitivity, specificity, false-negative rate, benchmark or prospective clinical outcome for the AI models. That leaves key R&D decisions open: whether Anaby’s signatures hold across populations, whether Kimmel’s slide-wide model predicts chemotherapy benefit reliably, and whether Golan’s preclinical models identify useful combinations.

The report was sponsored and produced in partnership with ICRF, with JTA’s native content team producing it. ICRF is therefore the funder and grant partner behind all three projects, and Herman’s claims of potential clinical impact remain grant-supported objectives rather than demonstrated results.

The next credible portfolio gate is measured validation: diverse external cohorts for imaging, documented chemotherapy-response validation for pathology, and preclinical response and resistance data before any clinical trial.