About

Yejia Zhang, MD, PhD, is an Associate Professor of Physical Medicine and Rehabilitation at the Hospital of the University of Pennsylvania. He is also a Staff Physician at the Corporal Michael J. Crescenz Department of Veterans Affairs Medical Center in Philadelphia, PA. His clinical expertise includes non-operative care of the spine, acupuncture to treat chronic pain and promote well-being, and wheelchair assessment. His research focuses on cell and molecular biology related to the spine, particularly intervertebral disc degeneration and injury. Zhang's work involves understanding gene expression changes, inflammatory markers, and cellular mechanisms in mouse models of intervertebral disc injury and degeneration.

Research topics

• Immunology
• Genetics
• Cell biology
• Endocrinology
• Biology
• Medicine
• Cancer research

Selected publications

• When Swin Transformer Meets KANs: An Improved Transformer Architecture for Medical Image Segmentation

  2026-04-08

  articleOpen access

  Medical image segmentation is critical for accurate diagnostics and treatment planning, but remains challenging due to complex anatomical structures and limited annotated training data. CNN-based segmentation methods excel at local feature extraction, but struggle with modeling long-range dependencies. Transformers, on the other hand, capture global context more effectively, but are inherently data-hungry and computationally expensive. In this work, we introduce UKAST, a U-Net like architecture that integrates rationalfunction based Kolmogorov—Arnold Networks (KANs) into Swin Transformer encoders. By leveraging rational base functions and Group Rational KANs (GR-KANs) from the Kolmogorov—Arnold Transformer (KAT), our architecture addresses the inefficiencies of vanilla spline-based KANs, yielding a more expressive and data-efficient framework with reduced FLOPs and only a very small increase in parameter count compared to SwinUNETR. UKAST achieves state-of-the-art performance on four diverse 2D and 3D medical image segmentation benchmarks, consistently surpassing both CNN- and Transformer-based baselines. Notably, it attains superior accuracy in data-scarce settings, alleviating the datahungry limitations of standard Vision Transformers. These results show the potential of KAN-enhanced Transformers to advance data-efficient medical image segmentation. Code is available at: https://github.com/nsapkota417/UKAST

  PublisherOA PDFDOI

• Effects of Age on Intervertebral Disc Tissue Morphology and Gene Expression in the ADAM8-Inactivation Mouse

  Cells · 2026-04-20

  articleOpen accessSenior authorCorresponding

  Purpose: To determine which age of mice should be used to compare the effects of ADAM8 mutation on intervertebral disc (IVD) responses to injury. Methods: IVDs of ADAM8 mutant (Adam8EQ) and wild type (WT) mice, aged 3, 10 and 18 months were injured. IVD tissues were harvested 1 week post injury for histological and molecular studies. Results: Histological scores increased with aging in intact IVDs, and there were no differences between Adam8EQ and WT mice (n = 11–28; p > 0.05). Safranin O-staining was less intense in 10-month than in 3-month-old mice, in both intact and injured IVDs (n = 3–15; p < 0.05). Cxcl1, Il6, and Adam8 gene expression levels were higher in the injured tail IVDs of 3-month-old Adam8EQ than WT mice (n = 18–30; p < 0.05); the injury-related differences diminished with increasing age. Conclusions: No histological differences were found between Adam8EQ and WT mouse IVDs at 3, 10 or 18 months of age, in the intact or injured discs. The differences in inflammatory marker gene expression were detectable at age 3 months, but were less evident when the injury occurred at age 10 or 18 months. Therefore, to identify differences in injury responses between WT and Adam8EQ mouse IVDs, 3-month-old mice are superior to older mice.

  PublisherDOI

• Gene Expression Changes Precede Elevated Mechanical Sensitivity in the Mouse Intervertebral Disc Injury Model

  JOR Spine · 2025-02-23

  articleOpen accessSenior authorCorresponding

  Background: Back pain after intervertebral disc (IVD) injury is a common clinical problem. Previous work examining early molecular changes post injury mainly used a candidate marker approach. Methods: In this study, gene expression in the injured and intact mouse tail IVDs was determined with a nonbiased whole transcriptome approach and related to subsequent pain behavior. Mouse tail IVD injury was induced by a needle puncture. Whole murine transcriptome was determined by RNASeq. Transcriptomes of injured IVDs were compared with those of intact controls by bioinformatic methods. Mechanical allodynia was assessed by the Von Frey method. Results: ⟨ 0.01). Ontology study of upregulated genes revealed that leukocyte migration was the most enriched biological process, and network analysis showed that Tnfa had the most protein-protein interactions. The most enriched downregulated pathways were related to the pattern specification process. Mechanical allodynia persisted at the 4-week end point. Conclusion: The RNASeq data revealed numerous early genes that participate in inflammation and repair processes post IVD injury. Mechanical allodynia followed these gene expression changes.

  PublisherOA PDFDOI

• TNFAIP8/TIPE2 inactivation modulates extracellular matrix organization gene expression and preserves the intervertebral disc structure in mice

  Genes & Diseases · 2025-08-30

  articleOpen accessSenior authorCorresponding
  PublisherDOI

• Author Correction: Universal conditional networks (UniCoN) for multi-age embryonic cartilage segmentation with sparsely annotated data

  Scientific Reports · 2025-03-13 · 1 citations

  erratumOpen access
  PublisherOA PDFDOI

• Modest Effects of Age on Inflammatory Markers and Histological Features in the Murine Intervertebral Disc Injury Model

  American Journal of Physical Medicine & Rehabilitation · 2025-05-23 · 1 citations

  articleSenior author

  OBJECTIVE: The aim of the study was to determine the effects of age on inflammatory markers and histological features in the injured mouse tail intervertebral disc. DESIGN: Inflammatory marker gene (e.g., Cxcl1 , Il6 , Adam8 , and Tipe2 ) expression and morphological changes (histological score and % red in Safranin O staining) in the injured intervertebral discs are recorded in 3-, 10-, and 18-mo-old mice 1 wk after injury. RESULTS: The injured intervertebral discs had higher histological scores (more degenerative) than intact discs at all ages post injury ( P < 0.01). However, there was no significant difference among the histological scores of injured intervertebral discs from mice of three ages ( P > 0.05). Expression of inflammatory marker genes (e.g., Cxcl1 and Il6 ) was elevated in the injured compared with intact discs in mice of all ages ( P < 0.01). The injury-induced increase in gene expression was greater in 10-mo and 18-mo-old mouse discs than in the 3-mo-old mice ( P < 0.01). CONCLUSIONS: The intervertebral discs responded to injuries similarly, regardless of age. Because of the modest age-related differences in injury effects, using mice of the same age in experiments is essential, unless one aims to examine age-related differences. Given the costs of maintaining an aged mouse colony, the necessity of using old animals may need justification.

  PublisherDOI

• A Rare Case of Psychogenic Nonepileptic Seizure Following Transcranial Magnetic Stimulation

  Cureus · 2025-04-23

  articleOpen accessSenior author

  Psychogenic nonepileptic seizures (PNES), also referred to as functional seizures, are events that mimic epileptic seizures but are not triggered by abnormal electrical activity in the brain. According to the International Statistical Classification of Diseases (ICD)-11, PNES are classified as dissociative disorders. Transcranial magnetic stimulation (TMS) is a non-invasive brain stimulation therapy commonly used to treat major depression, particularly in cases where other treatments have not been effective. PNES has not been associated with TMS previously. Here, we describe a 52-year-old Caucasian male who presented to the acupuncture clinic with multiple episodes of sudden loss of consciousness following TMS for a long history of major depression. The episodes of unconsciousness occurred up to five times per day. During an electroencephalograph (EEG) session, the patient had an episode that included poor balance, "shaking," head nodding, and a robotic/slowed voice, although no epileptic activity was captured on EEG. His illness was therefore diagnosed as PNES activity. He underwent treatment with body acupuncture and auricular acupressure and improved, with reduced number and duration of episodes. PNES following TMS has not been reported previously. A strong magnetic field can potentially disrupt normal neurotransmission and neuronal metabolism, resulting in PNES. The beneficial effects of acupuncture have been documented, but the mechanism of action has not been elucidated.

  PublisherOA PDFDOI

• Cell Instance Segmentation: The Devil Is in the Boundaries

  IEEE Transactions on Medical Imaging · 2025-10-14 · 1 citations

  article

  State-of-the-art (SOTA) methods for cell instance segmentation are based on deep learning (DL) semantic segmentation approaches, focusing on distinguishing foreground pixels from background pixels. In order to identify cell instances from foreground pixels (e.g., pixel clustering), most methods decompose instance information into pixel-wise objectives, such as distances to foreground-background boundaries (distance maps), heat gradients with the center point as heat source (heat diffusion maps), and distances from the center point to foreground-background boundaries with fixed angles (star-shaped polygons). However, pixel-wise objectives may lose significant geometric properties of the cell instances, such as shape, curvature, and convexity, which require a collection of pixels to represent. To address this challenge, we present a novel pixel clustering method, called Ceb (for Cell boundaries), to leverage cell boundary features and labels to divide foreground pixels into cell instances. Starting with probability maps generated from semantic segmentation, Ceb first extracts potential foreground-foreground boundaries (i.e., boundary candidates) with a revised Watershed algorithm. For each boundary candidate, a boundary feature representation (called boundary signature) is constructed by sampling pixels from the current foreground-foreground boundary as well as the neighboring background-foreground boundaries. Next, a lightweight boundary classifier is used to predict its binary boundary label based on the corresponding boundary signature. Finally, cell instances are obtained by dividing or merging neighboring regions based on the predicted boundary labels. Extensive experiments on six datasets demonstrate that Ceb outperforms existing pixel clustering methods on semantic segmentation probability maps. Moreover, Ceb achieves highly competitive performance compared to state-of-the-art cell instance segmentation methods. The code is available at: https://github.com/pxliang/Ceb.

  PublisherDOI

• Z-scheme MIL-88A@LaFeO3 heterojunction-enabled synergistic photo-Fenton system for acetamiprid degradation: mechanism and toxicity assessment

  Journal of Environmental Management · 2025-09-12 · 3 citations

  article1st author
  PublisherDOI

• Universal conditional networks (UniCoN) for multi-age embryonic cartilage segmentation with sparsely annotated data

  Scientific Reports · 2025-01-31 · 1 citations

  articleOpen access

  Osteochondrodysplasia, affecting 2-3% of newborns globally, is a group of bone and cartilage disorders that often result in head malformations, contributing to childhood morbidity and reduced quality of life. Current research on this disease using mouse models faces challenges since it involves accurately segmenting (precisely delineating) the developing cartilage in 3D micro-CT images of embryonic mice. Tackling this segmentation task with deep learning (DL) methods is laborious due to the big burden of manual image annotation, expensive due to the high acquisition costs of 3D micro-CT images, and difficult due to embryonic cartilage's complex and rapidly changing shapes. While DL approaches have been proposed to automate cartilage segmentation, most such models have limited accuracy and generalizability, especially across data from different embryonic age groups. To address these limitations, we propose novel DL methods that can be adopted by any DL architectures-including Convolutional Neural Networks (CNNs), Transformers, or hybrid models-which effectively leverage age and spatial information to enhance model performance. Specifically, we propose two new mechanisms, one conditioned on discrete age categories and the other on continuous image crop locations, to enable an accurate representation of cartilage shape changes across ages and local shape details throughout the cranial region. Extensive experiments on multi-age cartilage segmentation datasets show significant and consistent performance improvements when integrating our conditional modules into popular DL segmentation architectures. On average, we achieve a 1.7% Dice score increase with minimal computational overhead and a 7.5% improvement on unseen data. These results highlight the potential of our approach for developing robust, universal models capable of handling diverse datasets with limited annotated data, a key challenge in DL-based medical image analysis.

  PublisherOA PDFDOI

Recent grants

• NIH Grant K08HD049598

  NIH · $566k · 2011

• NIH Grant I21RX001896

  NIH · 2018

• ADAM8 in Intervertebral Disc Degeneration

  NIH · $364k · 2018–2021

Frequent coauthors

• Howard S. An

  Rush University Medical Center

  124 shared

• D. Greg Anderson

  Wolaita Sodo University

  81 shared

• Dessislava Markova

  Rothman Institute

  79 shared

• Ana Chee

  Rush University Medical Center

  66 shared

• Peng Shi

  Children's Hospital of Fudan University

  59 shared

• Motomi Enomoto‐Iwamoto

  50 shared

• Lutian Yao

  China Medical University

  47 shared

• Ling Qin

  40 shared

Labs

• Yejia Zhang LabPI