Machine learning researcher working on foundation models and computer vision for medical imaging.
Postdoctoral Research Fellow, Harvard Medical School & Boston Children’s Hospital
I build and evaluate machine learning methods for medical images. My current work adapts 3D segmentation foundation models with only a handful of labeled scans, refines the boundaries they get wrong, and tests models on data they have never seen.
I am a postdoctoral fellow in the Musculoskeletal Digital Innovation & Informatics Program (MDI2) at Boston Children’s Hospital and Harvard Medical School. I received my Ph.D. in Computer Science from UCLA in 2025, advised by Demetri Terzopoulos.
Current
A class-agnostic 3D boundary-refinement network, in the spirit of CascadePSP, that adapts foundation-model bone segmentations using only 1–20 labeled scans. A few-shot scaling study compares it with nnU-Net fine-tuning, and a joint-aware boundary loss gives it the edge on boundary accuracy where bones meet. Now extending to knee MRI.
An image-only classifier extending von Schacky et al. (Radiology, 2020) that pools features over the hip joint space located by a U-Net segmentation, validated on an independent hospital cohort.
Ph.D., UCLA
A vision transformer whose attention is trained on radiologists’ eye-tracking heatmaps, so it learns where experts look; at test time it needs only the image.
Sharpened Segment Anything Model masks in medical images with an active contour model, and studied test-time domain adaptation of large vision models.
Contributed to the external evaluation of 11 AI algorithms for screening mammography; co-authored the external validation of an AI ensemble model and an AI lesion-detection benchmark for tomosynthesis (both in JAMA Network Open); and matched magnified mammograms to specimen radiographs to characterize suspicious microcalcifications.
N. Nakhaei, T. Zhang, D. Terzopoulos, W. Hsu. Proc. SPIE 12927, Medical Imaging 2024: Computer-Aided Diagnosis.
N. Konz, et al., N. Nakhaei, et al., M. A. Mazurowski. JAMA Network Open 6(2): e230524.
W. Hsu, D. S. Hippe, N. Nakhaei, et al., C. I. Lee. JAMA Network Open 5(11): e2242343.
A. Vepa, A. Choi, N. Nakhaei, et al., F. Scalzo. IEEE/CVF Winter Conference on Applications of Computer Vision (WACV 2022).
K. Marathe, C. Marasinou, B. Li, N. Nakhaei, et al., W. Hsu. Computers in Biology and Medicine, 105504.
N. Nakhaei, C. Marasinou, A. Omigbodun, N. Capiro, B. Li, A. Hoyt, W. Hsu. Proc. SPIE 11597, Medical Imaging 2021: Computer-Aided Diagnosis.
C. Marasinou, B. Li, J. Paige, A. Omigbodun, N. Nakhaei, et al., W. Hsu. arXiv:2102.00754 (preprint).
Advisor: Demetri Terzopoulos. Thesis: Novel Techniques for Automated Medical Image Segmentation Spanning Diverse Anatomical Structures and Imaging Modalities.
Three elected terms representing 14,000 graduate and professional students; managed a $300,000 annual budget and initiated the BruinGrad Pass transportation referendum.
Introduction to Algorithms, Introduction to Computer Graphics, Current Topics in Data Structure; Head Teaching Fellow and instructor of the TA Training Seminar (Fall 2023).