Establishing the Global Standard for Clinical Reasoning in Medical AI

The MARC-Osler Foundation is a nonprofit research consortium dedicated to the independent identification of AI failure modes. We bridge the translation gap between state-of-the-art model performance and real-world clinical safety, with a focus on underrepresented patient populations and complex biological contexts.

Join

What We Are

An Independent Research Engine: We provide the infrastructure for external validation and generalization stress-testing.
A Clinical Commons: Grounded in international datasets from the Global South to ensure AI equity
A Fellowship for Leaders: Empowering MDs and PhDs to provide expert reasoning-level labels for mechanistic biology.

What We Are Not

A Model Developer: We do not build proprietary models; we evaluate them for the public good.
A Performance Leaderboard: We do not optimize for headline metrics like Dice scores alone.
A Commercial Entity: We are a mission-driven scientific organization.

How We Operate

We move beyond simple pattern recognition to evaluate Medical Abstraction. Our process is structured to be reproducible and transparent:

1. The Failure Atlas

We curate corner case datasets specifically designed to identify where models break, focusing on scanner variance, clinical drift, and diverse pathologies.

2. Expert-in-the-Loop Fellowships

Our fellows (MDs and PhDs in underserved regions) add high-value reasoning labels to datasets, translating raw imaging into biologically grounded insights.

3. Federated Stress-Testing

Using privacy-preserving governance, we evaluate AI systems on international datasets. Data remains under local clinical stewardship.

4. Transparent Taxonomy

We publish failure modes regardless of the outcome. We believe that knowing where a model fails is more valuable than knowing where it succeeds.

1. The Failure Atlas

We curate corner case datasets specifically designed to identify where models break, focusing on scanner variance, clinical drift, and diverse pathologies.

2. Expert-in-the-Loop Fellowships

Our fellows (MDs and PhDs in underserved regions) add high-value reasoning labels to datasets, translating raw imaging into biologically grounded insights.

3. Federated Stress-Testing

Using privacy-preserving governance, we evaluate AI systems on international datasets. Data remains under local clinical stewardship.

4. Transparent Taxonomy

We publish failure modes regardless of the outcome. We believe that knowing where a model fails is more valuable than knowing where it succeeds.

1. The Failure Atlas

We curate corner case datasets specifically designed to identify where models break, focusing on scanner variance, clinical drift, and diverse pathologies.

2. Expert-in-the-Loop Fellowships

Our fellows (MDs and PhDs in underserved regions) add high-value reasoning labels to datasets, translating raw imaging into biologically grounded insights.

3. Federated Stress-Testing

Using privacy-preserving governance, we evaluate AI systems on international datasets. Data remains under local clinical stewardship.

4. Transparent Taxonomy

We publish failure modes regardless of the outcome. We believe that knowing where a model fails is more valuable than knowing where it succeeds.

Osler Prize 01

Neuro-Oncology & Clinical Reasoning. Evaluating the generalizability of GBM and pediatric glioma segmentation models across diverse scanner types and clinical settings.

Partners from

Join

An Independent Research Engine: We provide the infrastructure for external validation and generalization stress-testing.

A Clinical Commons: Grounded in international datasets from the Global South to ensure AI equity

A Fellowship for Leaders: Empowering MDs and PhDs to provide expert reasoning-level labels for mechanistic biology.

A Model Developer: We do not build proprietary models; we evaluate them for the public good.

A Performance Leaderboard: We do not optimize for headline metrics like Dice scores alone.

A Commercial Entity: We are a mission-driven scientific organization.

1. The Failure Atlas

We curate corner case datasets specifically designed to identify where models break, focusing on scanner variance, clinical drift, and diverse pathologies.

2. Expert-in-the-Loop Fellowships

Our fellows (MDs and PhDs in underserved regions) add high-value reasoning labels to datasets, translating raw imaging into biologically grounded insights.

3. Federated Stress-Testing

Using privacy-preserving governance, we evaluate AI systems on international datasets. Data remains under local clinical stewardship.

4. Transparent Taxonomy

We publish failure modes regardless of the outcome. We believe that knowing where a model fails is more valuable than knowing where it succeeds.

1. The Failure Atlas

We curate corner case datasets specifically designed to identify where models break, focusing on scanner variance, clinical drift, and diverse pathologies.

2. Expert-in-the-Loop Fellowships

Our fellows (MDs and PhDs in underserved regions) add high-value reasoning labels to datasets, translating raw imaging into biologically grounded insights.

3. Federated Stress-Testing

Using privacy-preserving governance, we evaluate AI systems on international datasets. Data remains under local clinical stewardship.

4. Transparent Taxonomy

We publish failure modes regardless of the outcome. We believe that knowing where a model fails is more valuable than knowing where it succeeds.

1. The Failure Atlas

We curate corner case datasets specifically designed to identify where models break, focusing on scanner variance, clinical drift, and diverse pathologies.

2. Expert-in-the-Loop Fellowships

Our fellows (MDs and PhDs in underserved regions) add high-value reasoning labels to datasets, translating raw imaging into biologically grounded insights.

3. Federated Stress-Testing

Using privacy-preserving governance, we evaluate AI systems on international datasets. Data remains under local clinical stewardship.

4. Transparent Taxonomy

We publish failure modes regardless of the outcome. We believe that knowing where a model fails is more valuable than knowing where it succeeds.

Neuro-Oncology & Clinical Reasoning. Evaluating the generalizability of GBM and pediatric glioma segmentation models across diverse scanner types and clinical settings.

Contact

Contact

Contact