About

Mirabela Rusu is an Assistant Professor of Radiology at Stanford University, specializing in Integrative Biomedical Imaging Informatics. She is affiliated with the Center for Artificial Intelligence in Medicine & Imaging (AIMI), where her work focuses on the application of artificial intelligence to healthcare, particularly in medical imaging. Her research involves developing innovative AI methods to improve diagnostic accuracy and patient outcomes, integrating biomedical imaging data with computational techniques. As a faculty member at Stanford, she contributes to advancing the field of AI in medicine through her research, teaching, and participation in various academic initiatives.

Research topics

• Artificial Intelligence
• Computer Science
• Pathology
• Medicine
• Radiology
• Internal medicine
• Computer vision
• Engineering
• Nuclear medicine

Selected publications

• Current Trends and Future Prospects of Radiomics and Machine Learning (ML) Models in Spinal Tumors—A Narrative Review

  Journal of Imaging · 2026-03-19

  articleOpen access

  The intersection between radiomics, the computational analysis of imaging data, and machine learning (ML) may lead to new developments in the diagnosis, prognosis, and management of diseases. For spinal tumors specifically, applications of these fields appear promising. In this educational narrative review, we provide a summary of the current advancements in radiomics and artificial intelligence (AI), as well as applications of both fields in the diagnosis and management of spinal tumors. We also provide a suggested workflow of radiomics and machine learning analysis of spinal tumors for researchers, including a list and description of commonly used radiomic features. Future directions in the field of radiomics and machine learning applications to spinal tumors may involve validating already proposed algorithms with larger datasets, ensuring that all computational applications to patient care maintain high ethical standards, and continuing work in developing novel and highly accurate computational techniques to enhance patient outcomes.

  PublisherOA PDFDOI

• The MAMA-MIA Challenge: Advancing Generalizability and Fairness in Breast MRI Tumor Segmentation and Treatment Response Prediction

  arXiv (Cornell University) · 2026-03-01

  preprintOpen access

  Breast cancer is the most frequently diagnosed malignancy among women worldwide and a leading cause of cancer-related mortality. Dynamic contrast-enhanced magnetic resonance imaging plays a central role in tumor characterization and treatment monitoring, particularly in patients receiving neoadjuvant chemotherapy. However, existing artificial intelligence models for breast magnetic resonance imaging are often developed using single-center data and evaluated using aggregate performance metrics, limiting their generalizability and obscuring potential performance disparities across demographic subgroups. The MAMA-MIA Challenge was designed to address these limitations by introducing a large-scale benchmark that jointly evaluates primary tumor segmentation and prediction of pathologic complete response using pre-treatment magnetic resonance imaging only. The training cohort comprised 1,506 patients from multiple institutions in the United States, while evaluation was conducted on an external test set of 574 patients from three independent European centers to assess cross-continental and cross-institutional generalization. A unified scoring framework combined predictive performance with subgroup consistency across age, menopausal status, and breast density. Twenty-six international teams participated in the final evaluation phase. Results demonstrate substantial performance variability under external testing and reveal trade-offs between overall accuracy and subgroup fairness. The challenge provides standardized datasets, evaluation protocols, and public resources to promote the development of robust and equitable artificial intelligence systems for breast cancer imaging.

  PublisherDOI

• Early detection and differential diagnosis of ovarian cancerous tumors in computed tomography (CT) scans using an nnU-Net-based classifier

  2026-02-15

  articleSenior author
  PublisherDOI

• Differential privacy for medical deep learning: methods, tradeoffs, and deployment implications

  npj Digital Medicine · 2026-01-03

  articleOpen access

  Abstract Differential privacy (DP) is a prominent technique for protecting sensitive patient data in medical deep learning (DL), yet deploying it without compromising clinical utility or equity remains challenging. This scoping review synthesizes applications of DP in medical DL across centralized and federated settings. A structured search identified 74 eligible studies published through March 2025. Across modalities and tasks, DP, especially via DP-SGD, can maintain clinically acceptable performance under moderate privacy budgets ( ϵ ≈ 10), particularly in imaging. However, strict privacy ( ϵ ≈ 1) often leads to substantial accuracy loss, with amplified degradation in smaller or heterogeneous datasets. Only a minority of studies evaluate fairness, and several report that DP can widen subgroup performance gaps. Beyond DP-SGD, alternative mechanisms, including generative modeling, local DP, and hybrid federated designs, are emerging, but reporting of privacy parameters remains inconsistent. We identify key gaps in fairness auditing and standardization, and outline priorities for equitable, clinically robust privacy-preserving DL.

  PublisherOA PDFDOI

• Differential privacy for medical deep learning: methods, tradeoffs, and deployment implications

  npj Digital Medicine · 2026-01-03 · 6 citations

  articleOpen access

  Differential privacy (DP) is a prominent technique for protecting sensitive patient data in medical deep learning (DL), yet deploying it without compromising clinical utility or equity remains challenging. This scoping review synthesizes applications of DP in medical DL across centralized and federated settings. A structured search identified 74 eligible studies published through March 2025. Across modalities and tasks, DP, especially via DP-SGD, can maintain clinically acceptable performance under moderate privacy budgets (ϵ ≈ 10), particularly in imaging. However, strict privacy (ϵ ≈ 1) often leads to substantial accuracy loss, with amplified degradation in smaller or heterogeneous datasets. Only a minority of studies evaluate fairness, and several report that DP can widen subgroup performance gaps. Beyond DP-SGD, alternative mechanisms, including generative modeling, local DP, and hybrid federated designs, are emerging, but reporting of privacy parameters remains inconsistent. We identify key gaps in fairness auditing and standardization, and outline priorities for equitable, clinically robust privacy-preserving DL.

  PublisherOA PDFDOI

• Environment-wide association studies of urologic cancer and kidney disease in the United States

  Stanford Digital Repository · 2026-05-19

  dissertationOpen access
  PublisherDOI

• Evaluating an AI-driven Triaging Workflow for MRI-based Clinically Significant Prostate Cancer Diagnosis: A Simulation Study

  Radiology Imaging Cancer · 2026-04-10

  articleOpen access

  A simulated artificial intelligence–driven triaging workflow for clinically significant prostate cancer diagnosis at MRI triaged 49% of examinations, improving specificity while maintaining radiologist-level sensitivity.

  PublisherDOI

• The role of self-supervised pretraining in differentially private medical image analysis

  Open MIND · 2026-01-27

  preprint

  Differential privacy (DP) provides formal protection for sensitive data but typically incurs substantial losses in diagnostic performance. Model initialization has emerged as a critical factor in mitigating this degradation, yet the role of modern self-supervised learning under full-model DP remains poorly understood. Here, we present a large-scale evaluation of initialization strategies for differentially private medical image analysis, using chest radiograph classification as a representative benchmark with more than 800,000 images. Using state-of-the-art ConvNeXt models trained with DP-SGD across realistic privacy regimes, we compare non-domain-specific supervised ImageNet initialization, non-domain-specific self-supervised DINOv3 initialization, and domain-specific supervised pretraining on MIMIC-CXR, the largest publicly available chest radiograph dataset. Evaluations are conducted across five external datasets spanning diverse institutions and acquisition settings. We show that DINOv3 initialization consistently improves diagnostic utility relative to ImageNet initialization under DP, but remains inferior to domain-specific supervised pretraining, which achieves performance closest to non-private baselines. We further demonstrate that initialization choice strongly influences demographic fairness, cross-dataset generalization, and robustness to data scale and model capacity under privacy constraints. The results establish initialization strategy as a central determinant of utility, fairness, and generalization in differentially private medical imaging.

  DOI

• Cystic Artery Velocity: Evaluation of Performance in the Sonographic Diagnosis of Acute Cholecystitis

  Academic Radiology · 2026-04-01

  article
  PublisherDOI

• The MAMA-MIA Challenge: Advancing Generalizability and Fairness in Breast MRI Tumor Segmentation and Treatment Response Prediction

  ArXiv.org · 2026-03-01

  articleOpen access

  Breast cancer is the most frequently diagnosed malignancy among women worldwide and a leading cause of cancer-related mortality. Dynamic contrast-enhanced magnetic resonance imaging plays a central role in tumor characterization and treatment monitoring, particularly in patients receiving neoadjuvant chemotherapy. However, existing artificial intelligence models for breast magnetic resonance imaging are often developed using single-center data and evaluated using aggregate performance metrics, limiting their generalizability and obscuring potential performance disparities across demographic subgroups. The MAMA-MIA Challenge was designed to address these limitations by introducing a large-scale benchmark that jointly evaluates primary tumor segmentation and prediction of pathologic complete response using pre-treatment magnetic resonance imaging only. The training cohort comprised 1,506 patients from multiple institutions in the United States, while evaluation was conducted on an external test set of 574 patients from three independent European centers to assess cross-continental and cross-institutional generalization. A unified scoring framework combined predictive performance with subgroup consistency across age, menopausal status, and breast density. Twenty-six international teams participated in the final evaluation phase. Results demonstrate substantial performance variability under external testing and reveal trade-offs between overall accuracy and subgroup fairness. The challenge provides standardized datasets, evaluation protocols, and public resources to promote the development of robust and equitable artificial intelligence systems for breast cancer imaging.

  PublisherOA PDF

Recent grants

• Rad-pathomic deep learning models to assist radiologists in differentiating aggressive from indolent prostate cancer on MRI

  NIH · $2.8M · 2022–2026

Frequent coauthors

• Geoffrey A. Sonn

  Stanford University

  84 shared

• Richard E. Fan

  Stanford University

  77 shared

• Simon John Christoph Soerensen

  47 shared

• Pejman Ghanouni

  Stanford University

  42 shared

• Wei Shao

  University of Florida

  34 shared

• Indrani Bhattacharya

  Technical University of Munich

  33 shared

• Sulaiman Vesal

  Stanford Medicine

  28 shared

• Christian A. Kunder

  Palo Alto University

  27 shared