Uploaded April 2026 | Updated November 2026, 5 days ago
Machine Learning for Health (ML4H) Seminar Series
April 1, 2026
Talk Title: Advancing from Medical Imaging to Digital Twins
Speaker:
Stephen Aylward
NVIDIA
Abstract:
Medical imaging AI has reached an inflection point. In the first part of this talk, I will reflect on the community practices, technical choices, and open-science principles that helped MONAI become a widely adopted platform for medical imaging AI. With more than 8 million downloads, over 4,000 publication acknowledgements, and contributions from more than 240 developers worldwide, MONAI has supported a broad range of research efforts and regulatory-approved products.
The second part of the talk will ask what comes next. As healthcare moves beyond image interpretation toward systems that can sense, predict, and guide intervention, AI must evolve from perception to action. NVIDIA has described this broader direction as Physical AI: systems that perceive and interact with the world around them. I will argue that, in healthcare, this paradigm becomes even more powerful when coupled with simulation through Digital Twins. Together, Physical AI and Digital Twins enable a “sense, simulate, decide, act” framework for optimizing long-term patient outcomes and for shaping a new generation of intelligent healthcare systems.
I will conclude by highlighting the open-source platforms and collaborative ecosystem now emerging to support this future, along with early signals of technical and translational success.
Bio:
Stephen Aylward is the Global Lead for Strategic Applied Research for Medical Devices at NVIDIA. His collaborations focus developing open-source software and innovative algorithms that leverage accelerated computing to advance medical device R&D in academia and industry. His current research focuses on AI-enabled digital twin technologies for personalized surgical planning. Dr. Aylward’s background includes leading NIH, DARPA, and DoD-funded medical research for over 25 years as well as leading the development of numerous open-source platforms, including the Insight Toolkit (ITK), 3D Slicer, and MONAI. He is also a MICCAI Fellow, an adjunct professor in computer science at the University of North Carolina at Chapel Hill (UNC), chair of the MONAI advisory board, and an advisory board member for the Vanderbilt Institute for Surgery and Engineering and for Johns Hopkins Laboratory for Computational Sensing and Robotics. Previously, he was senior director of Strategic Initiatives at Kitware and a tenured professor in radiology at UNC.
For more information, visit: broadinstitute.org/ml4h
Copyright Broad Institute, 2026. All rights reserved.
Machine Learning for Health (ML4H) Seminar Series
April 1, 2026
Talk Title: Advancing from Medical Imaging to Digital Twins
Speaker:
Stephen Aylward
NVIDIA
Abstract:
Medical imaging AI has reached an inflection point. In the first part of this talk, I will reflect on the community practices, technical choices, and open-science principles that helped MONAI become a widely adopted platform for medical imaging AI. With more than 8 million downloads, over 4,000 publication acknowledgements, and contributions from more than 240 developers worldwide, MONAI has supported a broad range of research efforts and regulatory-approved products.
The second part of the talk will ask what comes next. As healthcare moves beyond image interpretation toward systems that can sense, predict, and guide intervention, AI must evolve from perception to action. NVIDIA has described this broader direction as Physical AI: systems that perceive and interact with the world around them. I will argue that, in healthcare, this paradigm becomes even more powerful when coupled with simulation through Digital Twins. Together, Physical AI and Digital Twins enable a “sense, simulate, decide, act” framework for optimizing long-term patient outcomes and for shaping a new generation of intelligent healthcare systems.
I will conclude by highlighting the open-source platforms and collaborative ecosystem now emerging to support this future, along with early signals of technical and translational success.
Bio:
Stephen Aylward is the Global Lead for Strategic Applied Research for Medical Devices at NVIDIA. His collaborations focus developing open-source software and innovative algorithms that leverage accelerated computing to advance medical device R&D in academia and industry. His current research focuses on AI-enabled digital twin technologies for personalized surgical planning. Dr. Aylward’s background includes leading NIH, DARPA, and DoD-funded medical research for over 25 years as well as leading the development of numerous open-source platforms, including the Insight Toolkit (ITK), 3D Slicer, and MONAI. He is also a MICCAI Fellow, an adjunct professor in computer science at the University of North Carolina at Chapel Hill (UNC), chair of the MONAI advisory board, and an advisory board member for the Vanderbilt Institute for Surgery and Engineering and for Johns Hopkins Laboratory for Computational Sensing and Robotics. Previously, he was senior director of Strategic Initiatives at Kitware and a tenured professor in radiology at UNC.
For more information, visit: broadinstitute.org/ml4h
Copyright Broad Institute, 2026. All rights reserved.







![MIA: Lorin Crawford, Rethinking scale in single-cell foundation models; Primer: Davide DAscenzo
Models, Inference and Algorithms
September 24, 2025
Broad Institute of MIT and Harvard
Meeting: When more isn’t better: rethinking scale in single-cell foundation models
Lorin Crawford
Microsoft Research New England (MSR)
The success of transformer-based foundation models on natural language and images has motivated their use in single-cell biology. Single-cell foundation models have been trained on increasingly larger transcriptomic datasets, scaling from initial studies with 1 million cells to newer atlases with over 100 million cells. In this talk, we will investigate the role of pre-training dataset size and diversity on the performance of single-cell foundation models on both zero-shot and fine-tuned tasks. In the first half, we use a large corpus of 22.2 million cells to pre-train a total of 400 models and evaluate over 6,400 experiments. We show that current methods tend to plateau in performance with pre-training datasets that are only a fraction of the size. This will lead us to the second half of the talk where we evaluate training data composition on model performance. Focusing on a tractable biological system (human hematopoiesis), we train and analyze deep generative models with a variety of training datasets, including cells from adult and developing tissues, disease states, and perturbation atlases. From the performance across these models, we observe that (1) deep generative models generalize poorly to unseen cell types, (2) addition of malignant cells to a healthy cell training corpus does not necessarily improve modeling of unseen malignant cells, and (3) inclusion of an embryonic stem cell transcription factor differentiation atlas in training data improves performance on out-of-distribution tasks. Our results highlight the distinct contributions of different training data types and point towards strategies for optimizing future single cell foundation models.
Primer: Infrastructure and modeling challenges in single-cell omics
Davide DAscenzo
Polytechnic University of Turin
Deep learning is expected to play a major role in advancing the analysis of single-cell omics data, but a number of practical challenges have so far limited its impact. In this talk, we will focus on two such challenges, one at the level of data infrastructure and one at the level of model design. With the rise of very large-scale single-cell experiments, we are now able to use datasets of hundreds of millions of cells. Model training on these datasets is often bottlenecked by the data-loading process of moving data from disk to GPU. To address this, we developed scDataset, a scalable and efficient data-loading solution for quasi-random sampling of single-cell data from disk [1]. We present block sampling and batched fetching strategies that balance I/O efficiency with memory consumption and minibatch diversity, demonstrating how to achieve increased throughput while maintaining sampling quality comparable to true random shuffling. Moving to biological tasks, we focus on cell type annotation. We find that both simple linear models and more complex transformer-based architectures struggle to generalize in out-of-distribution settings. To mitigate this, we introduce a hierarchical cross-entropy loss that incorporates the structure of cell type ontologies [2]. Across model classes, this leads to consistent improvements in performance, suggesting that structured biological priors may be more useful than scaling up model parameters.[1] Davide D’Ascenzo and Sebastiano Cultrera di Montesano. scDataset: Scalable data loading for deep learning on large-scale single-cell omics. arXiv, 2025. [2] Sebastiano Cultrera di Montesano, Davide D’Ascenzo, Srivatsan Raghavan, Ava P. Amini, Peter S. Winter, and Lorin Crawford. Hierarchical cross-entropy loss improves atlas-scale single-cell annotation models. bioRxiv, 2025.
For more information visit: https://www.broadinstitute.org/talks/spring-2025/mia
Copyright Broad Institute, 2025. All rights reserved. MIA: Lorin Crawford, Rethinking scale in single-cell foundation models; Primer: Davide DAscenzo](https://i.ytimg.com/vi/sMTOOAfG10g/mqdefault.jpg)


