About

Dr. Akshay Chaudhari is an Assistant Professor of research in the Integrative Biomedical Imaging Informatics at Stanford (IBIIS) section in the Department of Radiology and, by courtesy, in the Department of Biomedical Data Science at Stanford University. He leads the Machine Intelligence in Medical Imaging research group at Stanford, focusing on the intersection of artificial intelligence and medical imaging. Dr. Chaudhari's research integrates machine learning with medical imaging acquisition and analysis to advance the field of medical imaging. He completed his Ph.D. in Bioengineering at Stanford University in 2017, where he developed novel MRI methods for musculoskeletal imaging. His doctoral work was supported by prestigious fellowships including the National Science Foundation Graduate Research Fellowship, the Whitaker Fellowship, and the Siebel Fellowship. Following his Ph.D., Dr. Chaudhari trained as a postdoctoral fellow in Radiology at Stanford University, further combining machine learning techniques with medical imaging. Throughout his career, Dr. Chaudhari has received numerous awards recognizing his contributions, such as the W.S. Moore Young Investigator Award, the Junior Fellow Award, and an Outstanding Teacher Award from the International Society for Magnetic Resonance in Medicine, along with six additional young investigator awards for his work on advanced musculoskeletal medical imaging. In addition to his research, Dr. Chaudhari serves as the Associate Director of Research and Education at the Stanford AIMI Center and is an internal advisory board member of the Precision Health and Integrated Diagnostics Center.

Research topics

• Computer Science
• Artificial Intelligence
• Medicine
• Machine Learning
• Natural Language Processing
• Information Retrieval
• Nuclear medicine
• Data science
• Anatomy
• Radiology
• Mathematics
• Statistics
• Pathology
• Physical therapy
• Simulation
• Human–computer interaction
• Physical medicine and rehabilitation
• Psychology

Selected publications

• Enabling ankle-brachial index prediction from doppler sounds using deep learning

  npj Cardiovascular Health · 2026-04-09

  articleOpen access

  The ability to perform accurate point-of-care assessment of limb perfusion is critical for safe clinical decision-making. A formal ankle-brachial index (ABI) is typically required prior to supervised exercise therapy (SET) for peripheral arterial disease (PAD) or compression therapy for venous stasis ulcers. However, ABI measurements cannot be reported in patients with calcified, non-compressible tibial vessels. In this study, we introduce AutoABI, a deep learning algorithm that classifies ABI categories directly from circulatory Doppler sounds to improve the accessibility of point-of-care ABI assessment. AutoABI was trained and tested on a limited size dataset of 791 recordings from 198 patients and predicts ABI categories of <0.5, 0.5-0.7, 0.7-0.9, and >0.9. The algorithm achieved strong discriminatory performance with average areas under the receiver operating characteristic curve (AUCs) of 0.95, 0.96, 0.94, and 0.97 for the respective ABI ranges. Additional testing demonstrated the ability to predict ABI categories in patients with non-compressible arteries, offering a promising solution for more accessible and reliable PAD assessments.

  PublisherOA PDFDOI

• BMI and Varus Malalignment Compound to Define a High-Risk Phenotype for Compartment-Specific Knee Osteoarthritis Progression

  medRxiv · 2026-04-17

  articleOpen access

  Objectives: Knee osteoarthritis (KOA) is a leading cause of disability, yet which patients will experience structural decline remains unclear. Body mass index (BMI) and lower limb alignment are established risk factors for KOA, but their independent and interactive effects on compartment-specific cartilage loss and total knee replacement (TKR) have not been characterized at scale. Methods: We analyzed 5,832 limbs from 3,016 participants in the Osteoarthritis Initiative followed over 7 years. Cartilage thickness in the weight-bearing medial and lateral femur and tibia was quantified, and lower limb alignment was measured using hip-knee-ankle (HKA) angle obtained from full-limb radiographs. Linear mixed-effects models estimated the independent and interactive effects of BMI and lower limb alignment on longitudinal cartilage thinning, and mixed-effects logistic regression modeled TKR risk. Results: BMI and +10° varus, the rate of medial femur cartilage thinning was 243.5% faster than the reference rate. In the lateral compartment, BMI and valgus alignment were independently associated with faster cartilage thinning, with no significant interaction. TKR risk increased exponentially with HKA deviation (odds ratio [OR] = 1.38 per 1°; ~five-fold at 5° malalignment) but was not associated with BMI. Conclusion: BMI and lower limb alignment influence structural KOA progression through compartment-specific pathways. The multiplicative interaction in the medial compartment identifies high BMI combined with varus malalignment as a discrete high-risk phenotype, with implications for clinical risk stratification and disease-modifying intervention design.

  PublisherDOI

• GaitDynamics: a generative foundation model for analyzing human walking and running

  Nature Biomedical Engineering · 2026-01-05 · 1 citations

  articleOpen accessSenior author
  PublisherOA PDFDOI

• Magnetic resonance imaging provides comparable spinal curvature measurements to computerized tomography

  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: Spinal curvature measurements may be used to better understand scoliosis, lordosis, and low back anatomy. The unknown effect of the type of imaging modality on semi-automatically measured spinal curvature motivated this study, as well as the unknown relationship between image contrast and curvature measurements. Goal(s): To quantify how spinal curvature measurements are impacted by changing imaging modalities, specifically between CT and T1-weighted MR images, and between T1-weighted and T2-weighted MR images. Approach: We compared 20 patients' spinal curvature, calculated four separate ways from their L5-L1 vertebral centroids in each imaging modality. Results: We found no significant differences across modalities in calculating spinal curvature. Impact: This study impacts research done between CT scans and MR images, supporting intra-modality spinal curvature calculations from vertebral centroids calculated semi-automatically. This study also supports the equivalency of spinal curvature calculations from CT scans, T1-weighted and T2-weighted MR images.

  PublisherDOI

• Rapid and robust quantitative cartilage assessment for the clinical setting: deep learning-enhanced accelerated T2 mapping

  Skeletal Radiology · 2025-09-18 · 1 citations

  articleOpen access

  OBJECTIVE: Clinical adoption of T2 mapping is limited by poor reproducibility, lengthy examination times, and cumbersome image analysis. This study aimed to develop an accelerated deep learning (DL)-enhanced cartilage T2 mapping sequence (DL CartiGram), demonstrate its repeatability and reproducibility, and evaluate its accuracy compared to conventional T2 mapping using a semi-automatic pipeline. METHODS: DL CartiGram was implemented using a modified 2D Multi-Echo Spin-Echo sequence at 3 T, incorporating parallel imaging and DL-based image reconstruction. Phantom tests were performed at two sites to obtain test-retest T2 maps, using single-echo spin-echo (SE) measurements as reference values. At one site, DL CartiGram and conventional T2 mapping were performed on 43 patients. T2 values were extracted from 52 patellar and femoral compartments using DL knee segmentation and the DOSMA framework. Repeatability and reproducibility were assessed using coefficients of variation (CV), Bland-Altman analysis, and concordance correlation coefficients (CCC). T2 differences were evaluated with Wilcoxon signed-rank tests, paired t tests, and accuracy CV. RESULTS: Phantom tests showed intra-site repeatability with CVs ≤ 2.52% and T2 precision ≤ 1 ms. Inter-site reproducibility showed a CV of 2.74% and a CCC of 99% (CI 92-100%). Bland-Altman analysis showed a bias of 1.56 ms between sites (p = 0.03), likely due to temperature effects. In vivo, DL CartiGram reduced scan time by 40%, yielding accurate cartilage T2 measurements (CV = 0.97%) with no significant differences compared to conventional T2 mapping (p = 0.1). CONCLUSIONS: DL CartiGram significantly accelerates T2 mapping, while still assuring excellent repeatability and reproducibility. Combined with the semi-automatic post-processing pipeline, it emerges as a promising tool for quantitative T2 cartilage biomarker assessment in clinical settings.

  PublisherOA PDFDOI

• Automated MRI-Based Quantification of Forearm Muscle Health and Associations with Hand Function

  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: Hand function is impaired in many conditions. MRI-derived muscle health markers may improve the evaluation of conditions affecting hand function. Traditional manual segmentation is time-consuming, necessitating automated approaches. Goal(s): Develop an accurate method to automatically assess forearm muscle health (muscle volume, intramuscular fat) and assess their relationship to hand function. Approach: We developed and tested a computer-vision model for automated forearm segmentation using fat-water MRI, then assessed associations between muscle health (volume, intramuscular fat) and hand function (grip strength, dexterity). Results: The computer-vision model achieved high accuracy and good-excellent reliability. Muscle volume was associated with BMI and grip strength. Impact: We developed an accurate, reliable computer-vision model to automatically segment forearm muscles, which will be made openly available. This method can improve clinical assessment of forearm muscle health leading to more efficient evaluation and management of conditions affecting hand function.

  PublisherDOI

• Hamstring Muscle Architecture and Microstructure Changes Following 9-weeks of Nordic Hamstring Exercise Training

  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

  articleSenior author

  Motivation: Evaluate long-term muscle adaptations across the full volume of all four hamstrings in response to Nordic hamstring exercise (NHE) to enhance injury prevention strategies. Goal(s): Examine how 9-weeks of supervised NHE-training affects architecture (volume, fiber length, angle, and curvature) and microstructure (MD, RD, FA, and T2) of Biceps femoris short head (BFsh), Biceps femoris long head (BFlh), Semitendinosus (ST), and Semimembranosus (SM). Approach: 11 subjects underwent MRI scans (Dixon, DTI, and T2) pre and post 9-weeks NHE-training. Results: NHE-training increased hamstring volume with greater hypertrophy in ST and BFsh muscles. Hypertrophy was accompanied by increases in both length and cross-section of muscle fibers. Impact: This study examines architectural and microstructural adaptations of the hamstrings following 9-weeks of Nordic hamstring exercise training. Findings reveal significant, but non-uniform hypertrophy among hamstrings accompanied by increase in length and size of the muscle fibers, advancing injury prevention strategies.

  PublisherDOI

• Comparing List-mode and Count-Mixing Techniques for Deep Learning-Based Disambiguation of AD Radiotracers in PET/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: AD patients require multiple visits for amyloid and tau imaging because PET cannot acquire multiple radiotracers in a single session. Goal(s): DL-based separation of amyloid and tau radiotracers from mixed-dose images could reduce AD patient visits, but current DL applications are hindered by computational costs and other challenges of list-mode dose-mixing. Approach: We propose count-mixing as a compute-efficient alternative for simulating dose-mixing, which can then be used for deep learning (DL)-based radiotracer separation. Results: PET/MR count-mixing can serve as an alternative to list-mode dose-mixing. The approach agrees with list-mode dose-mixing, exhibits enhanced quantitative performance, and equivalent anatomical preservation. Impact: Count-mixing provides a faster, compute-efficient way to generate realistic mixed-dose PET images, enhancing model training and scaling DL applications for radiotracer separation. This approach could enable simultaneous injection of multiple radiotracers in a single acquisition for AD patients.

  PublisherDOI

• Effects of Real-Time Notification of AI-Detected Incidental Coronary Artery Calcium on Statin Prescription: The NOTIFY-PICTURE Trial

  Circulation · 2025-11-10

  letterOpen access
  PublisherOA PDFDOI

• Structured Prompts Improve Evaluation of Language Models

  arXiv (Cornell University) · 2025-11-25

  preprintOpen accessSenior author

  As language models (LMs) are increasingly adopted across domains, high-quality benchmarking frameworks are essential for guiding deployment decisions. In practice, however, frameworks such as Holistic Evaluation of Language Models (HELM) typically evaluate models under a single static prompt configuration, even though model behavior depends strongly on prompt choice. As a result, reported scores can reflect prompt choice as much as model capability. Declarative prompting frameworks such as DSPy offer a scalable way to evaluate models under a set of structured prompting strategies rather than a static prompt configuration. We present a reproducible DSPy+HELM framework for studying how prompt choice impacts reported benchmark outcomes. Using five prompting methods, we evaluate four frontier and two open-source LMs across seven benchmarks against existing HELM baseline scores. By evaluating LMs across a family of prompt configurations, we find that prompt choice can materially impact leaderboard outcomes. In particular, structured prompting improves performance (by 6% on average), alters comparisons (leaderboard rankings shift on 5/7 benchmarks), with most gains coming from introducing chain-of-thought, and little additional benefit from more advanced optimizers. To our knowledge, this is the first study to systematically integrate structured prompting into an established evaluation framework and quantify how prompt choice alone can impact benchmark conclusions. We open-source (i) DSPy+HELM Evaluation (https://github.com/stanford-crfm/helm/pull/3893) and (ii) Prompt Optimization Pipeline (https://github.com/StanfordMIMI/dspy-helm).

  PublisherOA PDFDOI

Recent grants

• Designing Novel MR Imaging Tools to Quantify Lower-Limb Exercise Adaptations in Knee Osteoarthritis

  NIH · $9.5M · 2003–2028

• Opportunistic Atherosclerotic Cardiovascular Disease Risk Estimation at Abdominal CTs with Robust and Unbiased Deep Learning

  NIH · $1.8M · 2023–2028

• Novel Incidental Calcium Evaluation (NICE)

  NIH · $1.4M · 2024–2028

Frequent coauthors

• Garry E. Gold

  76 shared

• Brian A. Hargreaves

  Stanford University

  57 shared

• Arjun Desai

  50 shared

• Matthew P. Lungren

  40 shared

• Feliks Kogan

  38 shared

• Curtis P. Langlotz

  Palo Alto University

  38 shared

• Scott L. Delp

  Stanford University

  35 shared

• Louis Blankemeier

  Stanford University

  34 shared

Labs

• Machine Intelligence for Medical Imaging (MIMI)PI

Education

• Post-Doctoral Scholar, Radiology

  Stanford University

• PhD, Bioengineering

  Stanford University

  2017

• BS, Bioengineering

  University of California San Diego

  2012

Awards & honors

• W.S. Moore Young Investigator Award
• Junior Fellow Award from the International Society for Magne…
• Inducted into the Academy of Radiology’s Council of Early Ca…