About

Elfar Adalsteinsson is the Eaton-Peabody Professor at MIT, with a research focus that includes AI for Healthcare and Life Sciences, Biological and Medical Devices and Systems, Graphics and Vision. His work leverages computational, theoretical, and experimental tools to develop groundbreaking sensors, energy transducers, and new physical substrates for computation, addressing shared challenges facing humanity. As a prominent figure in electrical engineering and artificial intelligence, he contributes to advancing the integration of AI technologies in healthcare and biological systems, fostering innovations that impact medical devices and systems.

Research topics

• Computer Science
• Artificial Intelligence
• Medicine
• Internal medicine
• Cardiology
• Computer vision
• Physics
• Optics
• Biology
• Algorithm
• Radiology
• Geology

Selected publications

• Fast Multi-Stack Slice-to-Volume Reconstruction via Multi-Scale Unrolled Optimization

  arXiv (Cornell University) · 2026-01-12

  preprintOpen access

  Fully convolutional networks have become the backbone of modern medical imaging due to their ability to learn multi-scale representations and perform end-to-end inference. Yet their potential for slice-to-volume reconstruction (SVR), the task of jointly estimating 3D anatomy and slice poses from misaligned 2D acquisitions, remains underexplored. We introduce a fast convolutional framework that fuses multiple orthogonal 2D slice stacks to recover coherent 3D structure and refines slice alignment through lightweight model-based optimization. Applied to fetal brain MRI, our approach reconstructs high-quality 3D volumes in under 10s, with 1s slice registration and accuracy on par with state-of-the-art iterative SVR pipelines, offering more than speedup. The framework uses non-rigid displacement fields to represent transformations, generalizing to other SVR problems like fetal body and placental MRI. Additionally, the fast inference time paves the way for real-time, scanner-side volumetric feedback during MRI acquisition.

  PublisherDOI

• Fast Multi-Stack Slice-to-Volume Reconstruction via Multi-Scale Unrolled Optimization

  ArXiv.org · 2026-01-12

  articleOpen access

  Fully convolutional networks have become the backbone of modern medical imaging due to their ability to learn multi-scale representations and perform end-to-end inference. Yet their potential for slice-to-volume reconstruction (SVR), the task of jointly estimating 3D anatomy and slice poses from misaligned 2D acquisitions, remains underexplored. We introduce a fast convolutional framework that fuses multiple orthogonal 2D slice stacks to recover coherent 3D structure and refines slice alignment through lightweight model-based optimization. Applied to fetal brain MRI, our approach reconstructs high-quality 3D volumes in under 10s, with 1s slice registration and accuracy on par with state-of-the-art iterative SVR pipelines, offering more than speedup. The framework uses non-rigid displacement fields to represent transformations, generalizing to other SVR problems like fetal body and placental MRI. Additionally, the fast inference time paves the way for real-time, scanner-side volumetric feedback during MRI acquisition.

  PublisherOA PDF

• Why Invariance is Not Enough for Biomedical Domain Generalization and How to Fix It

  ArXiv.org · 2026-04-02

  articleOpen access

  We present MaskGen, a theoretically grounded and deliberately simple approach for domain generalization in 3D biomedical image segmentation. Modern segmentation models degrade sharply under shifts in modality, disease severity, clinical sites, and more, limiting their reliable adoption. Existing generalization methods address this using extreme augmentations, hand-engineered domain statistics mixing, or architectural redesigns that add significant implementation overhead while yielding inconsistent performance across biomedical settings. MaskGen instead presents a principled learning strategy with marginal overhead that utilizes both source-domain image intensities and domain-stable foundation model representations to train robust segmentation models. As a result, MaskGen achieves strong gains in both fully supervised and few-shot segmentation across broad clinical shifts in biomedical studies. Unlike prior approaches, MaskGen is architecture- and loss-agnostic, compatible with standard augmentation pipelines, easy to implement, and tackles arbitrary anatomical regions. Its implementation is freely available at https://github.com/sebodiaz/MaskGen.

  PublisherOA PDF

• Why Invariance is Not Enough for Biomedical Domain Generalization and How to Fix It

  arXiv (Cornell University) · 2026-04-02

  preprintOpen access

  We present MaskGen, a theoretically grounded and deliberately simple approach for domain generalization in 3D biomedical image segmentation. Modern segmentation models degrade sharply under shifts in modality, disease severity, clinical sites, and more, limiting their reliable adoption. Existing generalization methods address this using extreme augmentations, hand-engineered domain statistics mixing, or architectural redesigns that add significant implementation overhead while yielding inconsistent performance across biomedical settings. MaskGen instead presents a principled learning strategy with marginal overhead that utilizes both source-domain image intensities and domain-stable foundation model representations to train robust segmentation models. As a result, MaskGen achieves strong gains in both fully supervised and few-shot segmentation across broad clinical shifts in biomedical studies. Unlike prior approaches, MaskGen is architecture- and loss-agnostic, compatible with standard augmentation pipelines, easy to implement, and tackles arbitrary anatomical regions. Its implementation is freely available at https://github.com/sebodiaz/MaskGen.

  PublisherDOI

• Fetal <scp>MRI</scp>: Radiofrequency Safety Assessment at 3 Tesla

  Journal of Magnetic Resonance Imaging · 2025-04-17 · 2 citations

  articleOpen access

  BACKGROUND: 3-T MRI can improve image quality of fetal imaging compared to 1.5-T MRI. However, concerns exist regarding increased local tissue heating at 3-T. PURPOSE: To assess fetal MRI radiofrequency (RF) safety at 3-T by comparing simulated tissue heating to 1.5-T (using constant RF exposure) and by simulating tissue heating at 3-T using RF exposures from clinical fetal examinations. STUDY TYPE: Retrospective. POPULATION: ). FIELD STRENGTH/SEQUENCE: 3-T, 1.5-T, HASTE, VIBE, TRUFISP, EPI, DTI. ASSESSMENT: Simulated maternal and fetal peak and average SAR, temperature, and peak thermal dose were compared at 3-T and 1.5-T for 60 min 2 W/kg wbSAR using 7 body models and a 16-rung band-pass RF coil. Temperature and thermal dose were simulated in one body model using clinical wbSAR exposures at 3-T. STATISTICAL TESTS: Factorial analysis of variance was performed using 28 maternal and fetal temperature measurements from 7 body models to detect a difference between 3-T and 1.5-T. p < 0.05 was considered statistically significant. RESULTS: For constant RF exposure, we found no difference between 3-T and 1.5-T in peak maternal (1.5-T:40.38 ± 0.21°C; 3-T:40.40 ± 0.20°C; p = 0.85), peak fetal (1.5-T:39.21 ± 0.17°C; 3-T:39.09 ± 0.16°C; p = 0.19), and average maternal (1.5-T:37.32 ± 0.05°C; 3-T:37.33 ± 0.04°C; p = 0.68) temperature. We observed significantly higher average fetal temperatures at 1.5-T (1.5-T:37.75 ± 0.06°C; 3-T:37.70 ± 0.05°C). For 3-T clinical RF exposures, simulated peak temperatures exceeded the recommended limits. However, the thermal dose was below the recommended limit. DATA CONCLUSION: For the same RF coil geometry, local heating was similar at 3-T and 1.5-T for constant RF exposure. Although realistic 3-T RF exposures could cause peak temperatures above the recommended limits, thermal dose was below the recommended limit. EVIDENCE LEVEL: 1. TECHNICAL EFFICACY: Stage 1.

  PublisherOA PDFDOI

• Incorporating realistic fetal motion into RF safety assessment of fetal MRI

  Proceedings on CD-ROM - International Society for Magnetic Resonance in Medicine. Scientific Meeting and Exhibition/Proceedings of the International Society for Magnetic Resonance in Medicine, Scientific Meeting and Exhibition · 2025-09-16

  article

  Motivation: Recent simulation studies using static fetus models show possible fetal heating concerns and the potential for fetal motion to mitigate tissue heating is unknown. Goal(s): To demonstrate the effect of realistic fetal motion on fetal SAR estimations. Approach: We used automatic and manual segmentation, and registration approaches to generate 4 body models of a pregnant subject with different fetal positions. Fetal SAR was simulated in each model. Results: Fetal average and peak SAR changed by up to 23% and 9% respectively and fetal peak SAR was located at different regions in the fetus including the back and left arm. Impact: We demonstrated a pipeline to incorporate realistic fetal motion into SAR estimations for the first time, enhancing RF safety assessment of fetal MRI. Our results show fetal motion is critical for accurate tissue heating estimations in the fetus.

  PublisherDOI

• Robust Fetal Pose Estimation Across Gestational Ages via Cross-Population Augmentation

  Lecture notes in computer science · 2025-09-19 · 1 citations

  book-chapterOpen accessSenior author
  PublisherOA PDFDOI

• Robust Fetal Pose Estimation across Gestational Ages via Cross-Population Augmentation

  ArXiv.org · 2025-09-15

  preprintOpen accessSenior author

  Fetal motion is a critical indicator of neurological development and intrauterine health, yet its quantification remains challenging, particularly at earlier gestational ages (GA). Current methods track fetal motion by predicting the location of annotated landmarks on 3D echo planar imaging (EPI) time-series, primarily in third-trimester fetuses. The predicted landmarks enable simplification of the fetal body for downstream analysis. While these methods perform well within their training age distribution, they consistently fail to generalize to early GAs due to significant anatomical changes in both mother and fetus across gestation, as well as the difficulty of obtaining annotated early GA EPI data. In this work, we develop a cross-population data augmentation framework that enables pose estimation models to robustly generalize to younger GA clinical cohorts using only annotated images from older GA cohorts. Specifically, we introduce a fetal-specific augmentation strategy that simulates the distinct intrauterine environment and fetal positioning of early GAs. Our experiments find that cross-population augmentation yields reduced variability and significant improvements across both older GA and challenging early GA cases. By enabling more reliable pose estimation across gestation, our work potentially facilitates early clinical detection and intervention in challenging 4D fetal imaging settings. Code is available at https://github.com/sebodiaz/cross-population-pose.

  PublisherOA PDFDOI

• Spatiotemporal changes in the uterus during non-labor contractions

  Proceedings on CD-ROM - International Society for Magnetic Resonance in Medicine. Scientific Meeting and Exhibition/Proceedings of the International Society for Magnetic Resonance in Medicine, Scientific Meeting and Exhibition · 2025-09-16

  article

  Motivation: Uteroplacental perfusion can be temporarily compromised by non-labor uterine contractions. Goal(s): We performed a quantitative, image-based analysis of the spatiotemporal uterine, myometrium, and placental changes associated with non-labor contractions. Approach: The temporal volumetric changes in the uterus and myometrium, and their correlation with changes in placental volume and T2*, were determined. We also compared volume and T2* changes in placental cotyledons to the surrounding interstitial spaces. Results: Consistent volume decreases in the uterus and myometrium were observed together with decreases in placental volume and T2*. Significantly more deformation and faster decreases in T2* were observed in placental cotyledons compared to the interstitial spaces. Impact: Non-labor contractions consistently correlate with placental volume and T2* decreases with faster changes in cotyledons. Contractions likely eject maternal blood out of the placenta resulting in transient hypoxic stress. Imaging contractions may provide insight into hypoxic stress for the fetus.

  PublisherDOI

• Motion-Aware Neural Networks Improve Rigid Motion Correction of Accelerated Segmented Multislice MRI

  Proceedings on CD-ROM - International Society for Magnetic Resonance in Medicine. Scientific Meeting and Exhibition/Proceedings of the International Society for Magnetic Resonance in Medicine, Scientific Meeting and Exhibition · 2024-08-14

  articleOpen access

  We demonstrate a deep learning approach for fast retrospective intraslice rigid motion correction in segmented multislice MRI. A hypernetwork uses auxiliary rigid motion parameter estimates to produce a reconstruction network based on the motion parameters that are specific to the input image. This strategy produces higher quality reconstructions than those produced by model-based techniques or by networks that do not use motion estimates. Further, this approach mitigates sensitivity to misestimation of the motion parameters.

  PublisherOA PDFDOI

Recent grants

• NIH Grant R01AG018942

  NIH · $950k · 2005

• Novel MRI Assessment of Placental Structure and Function Throughout Pregnancy

  NIH · $3.6M · 2019–2025

• NIH Grant R01EB007942

  NIH · $1.4M · 2013

• Advanced Fetal Imaging

  NIH · $6.2M · 2014–2021

• Fetal MRI: robust self-driving brain acquisition and body movement quantification

  NIH · $2.8M · 2022–2026

Frequent coauthors

• Lawrence L. Wald

  385 shared

• Adolf Pfefferbaum

  SRI International

  242 shared

• Edith V. Sullivan

  Stanford University

  188 shared

• Borjan Gagoski

  150 shared

• Kawin Setsompop

  129 shared

• Polina Golland

  Massachusetts Institute of Technology

  110 shared

• Berkin Bilgic̦

  103 shared

• Junshen Xu

  100 shared

Education

• Ph.D., Electrical Engineering and Computer Science

  Massachusetts Institute of Technology

  2005

• M.S., Electrical Engineering and Computer Science

  Massachusetts Institute of Technology

  2001

• B.S., Computer Science

  University of Iceland

  1998