Imagene AI

The models are the product. LungOI and the Lung Prediction Panel report gene alterations that CanvOI, Imagene's pathology foundation model, predicts from H&E images, with no sequencing behind the result. OI Suite is software for training and running prediction models on a customer's own patient groups. Without the models there is no test result and no software to sell.

Ask which CanvOI version produces each result, and how a change of model version shows up in reports.

Results go to an ordering clinician, and the controls around a prediction are not described. Only authorized clinicians can order LungOI. Results arrive through a portal with heat maps and summaries of the tissue features behind the prediction.

The research panel is positioned for deciding which samples go to confirmatory molecular testing, and says its results cannot guide treatment. The LungOI page, by contrast, says its results guide the start of treatment and presents a patient treated after an EGFR prediction. It does not say whether sequencing should confirm the result first.

Nothing published states the threshold at which a prediction is reported as positive, when a result is withheld as indeterminate, or how a wrong prediction is handled.

Ask whether LungOI results should be confirmed by sequencing before treatment, the reporting threshold per alteration, and how indeterminate cases are reported.

The model is named and its training data outlined. Version and update practice are not. CanvOI is Imagene's pathology foundation model, developed with Oracle and the Ellison Medical Institute. It is described as trained on more than two million H&E and immunohistochemistry images and tiles from multiple sites across more than 40 organs, with billions of parameters. The 2026 peer reviewed study names version 1.1.

Imagene claims up to 20 percent better performance than other models when data is limited, without publishing the comparison on its pages. No release notes, version history or notice of model changes is published for LungOI or OI Suite customers.

Ask which CanvOI version is in production for each product, and how customers are told when it changes.

Imagene builds its own model. CanvOI was developed with Oracle and the Ellison Medical Institute, and no outside model provider is named or implied between a slide and a result. Imagene describes cloud infrastructure through Oracle, which hosts the systems rather than supplying the model.

No subprocessor list is published, and the parties that handle laboratory images, results and OI Suite customer cohorts are not listed.

Ask for the processors that handle patient images and results, including the cloud host and region, and whether Oracle or the Ellison Medical Institute receive any customer or patient data.

A peer reviewed study supports the predictions, without figures at a stated threshold or any outcome in use. A study in npj Precision Oncology (Rolfo and colleagues, 2026), which Imagene cites, tested CanvOI 1.1 on 968 non small cell lung cancer samples. Areas under the curve were 0.87 for EGFR, 0.96 for ALK, 0.88 for BRAF and 0.83 for MET.

Imagene also describes validation on outside data from more than ten medical centers and a single center prospective study. It cites a case report in the Journal of Clinical Pathology of a patient treated after an EGFR prediction, and names Sheba Medical Center, Tel Aviv Sourasky Medical Center, Northwestern Medicine, Hoag and Ohio State among its partners.

Imagene's pages give no sensitivity, specificity or negative predictive value at the threshold used for reporting. No study shows how often LungOI results change or speed treatment compared with standard testing.

Ask for sensitivity and specificity per alteration at the reporting threshold, the prospective study's design and results, and the share of predictions later confirmed by sequencing.

Imagene gives general assurances that do not answer the questions its models raise. It describes its laboratory system as HIPAA compliant and secured, and its website privacy policy says information is not kept longer than needed.

Nothing published says how long patient slides, images and results are kept, whether clinical cases or customer cohorts are used to train or refine CanvOI, or how training data was de identified. CanvOI is trained on millions of images from multiple sites, and OI Suite customers upload their own cohorts. No safety engineering is published.

Ask how long slides, images and results are kept, whether customer cohorts or clinical cases train Imagene's models, and how training data was de identified.

Imagene claims HIPAA compliance without the documents behind it. The LungOI page describes a HIPAA compliant, secured laboratory information system, and the test is run by a US clinical laboratory that receives patient slides and images.

The only privacy policy published was last updated in December 2020 under Israeli law, and says it covers information collected through the website only. No notice of privacy practices for the laboratory, no business associate position for OI Suite customers and no agreement are published.

Ask for the laboratory's notice of privacy practices, whether Imagene acts as a covered entity or a business associate for each product, and the agreement offered to drug and diagnostics customers.

Imagene publishes no security audit or certification. It describes a HIPAA compliant, secured laboratory information system, which is a compliance claim, not an audit. No SOC 2 report, ISO/IEC 27001 certificate, HITRUST certification, penetration testing statement or trust center appears on its site.

Laboratory certification and accreditation cover testing quality, not information security. Any certifications Oracle's cloud holds are the host's, not Imagene's.

Ask for a current security attestation covering the laboratory portal and OI Suite, with its scope and date.

Imagene makes no device claim, and offers its clinical test as a laboratory developed test. LungOI runs in a CLIA certified, CAP accredited laboratory, and its page states it has not been cleared or approved by the FDA. The Lung Prediction Panel is labeled for research use only, not for diagnosis, and OI Suite is research software. The FDA's databases list no Imagene device.

The LungOI page goes further than the panel. It says the predictions guide the start of treatment, and presents a case where targeted therapy followed an AI predicted mutation. A buyer should weigh that against the test's status.

Ask how LungOI is positioned for treatment decisions made without sequencing, and whether an FDA submission is planned for the test or for companion diagnostics built on CanvOI.

Imagene describes validation across sites, but publishes no results by patient group and no governance process. It describes external validation on data from more than ten medical centers and a prospective study, and a 2026 peer reviewed study reports results by alteration.

Nothing published compares performance across centers, scanners, specimen types, sex, smoking history or ancestry. That matters because lung cancer driver alterations differ in frequency by ancestry and smoking history, which affects how a prediction performs. No responsible AI policy, model governance process or monitoring after release is described.

Ask for results by center, scanner, specimen type, smoking history and ancestry, and how performance is monitored after model updates.

Clinicians can check a prediction, but nothing stands behind it. LungOI results come with heat maps and summaries of the tissue features behind each prediction, which a pathologist can inspect. The research panel is positioned for deciding which samples go on to confirmatory molecular testing.

The published validation reports areas under the curve, which do not give an error rate at the threshold used to report a result. No notice of privacy practices or correction route for laboratory results, warranty, retesting commitment or indemnity is published.

Ask for false positive and false negative rates per alteration at the reporting threshold, and what Imagene does when sequencing contradicts a reported prediction.

Results travel through Imagene's own portal. LungOI results are delivered through an online laboratory portal that notifies the clinician. No record system, laboratory information system or interface standard is named for sending results into a hospital's systems.

For research buyers, OI Suite runs in the cloud, as a hybrid or on premises. Imagene joined Proscia's partner alliance in August 2026 to build AI immunohistochemistry companion diagnostics, without a live integration described.

Ask whether LungOI results can reach your record system as discrete data, and which image management systems OI Suite reads from in production.

Deployment options are named, but no location is committed. OI Suite is offered in the cloud, as a hybrid or on premises, and Imagene describes cloud infrastructure through Oracle. LungOI runs in Imagene's US laboratory, with results on a hosted portal.

Nothing published says in which region images and results are stored, whether data from US patients is processed in Israel, or how customer environments are kept apart.

Ask where LungOI images and results are stored and processed, which Oracle region hosts OI Suite, and whether any processing crosses between the United States and Israel.

No price is published, but how each product is sold can be worked out. Clinicians order LungOI through Imagene's laboratory portal. It has its own billing code and a Medicare payment rate, which is what Medicare pays the laboratory, not a price Imagene states.

The Lung Prediction Panel is sold for research use. OI Suite is licensed by subscription for cloud, hybrid or on premises use, with access arranged by email.

Ask for the LungOI list price and the patient cost outside Medicare, the panel price per slide, and how the OI Suite subscription is priced.

Imagene names a narrow scope and has validation behind the lung products. LungOI and the Lung Prediction Panel are for non small cell lung cancer, working from biopsy H&E images or glass slides. The panel names eight alterations: EGFR, KRAS G12C, ERBB2 and BRAF V600E mutations, ALK, RET and ROS1 fusions, and MET exon 14 skipping. A 2026 peer reviewed study reports results for four of them on 968 samples.

LungOI is a clinical test ordered by oncologists, which is why Imagene is also listed under diagnostics and genomics. OI Suite serves drug company translational research, biomarker discovery and companion diagnostic work, which is why it is also listed under drug discovery. CanvOI is described as trained across more than 40 organs, with breast and colorectal cancer named for OI Suite, but no validation outside lung is published.

Ask for per alteration results on your specimen types, and for validation outside lung cancer before relying on OI Suite elsewhere.