Chamith S. Rajapakse
· Ph.D.VerifiedUniversity of Pennsylvania · Rehabilitation Medicine
Active 1998–2026
About
Chamith S. Rajapakse, Ph.D., is an Associate Professor of Radiology at the University of Pennsylvania's Perelman School of Medicine. His research expertise encompasses medical imaging, artificial intelligence, and orthopedic engineering. Dr. Rajapakse has contributed to the development and application of advanced imaging techniques, including magnetic resonance imaging (MRI), for evaluating liver fat fraction, bone tissue properties, craniofacial structures, and lung parenchymal aeration. His work involves leveraging deep learning segmentation methods and high-resolution imaging to improve diagnostic accuracy and understanding of various medical conditions. With a background in electrical engineering, computer science, and physics, Dr. Rajapakse's interdisciplinary approach supports innovations in biomedical imaging and computational analysis, advancing both clinical and research applications in radiology.
Research topics
- Pathology
- Radiology
- Medicine
- Surgery
Selected publications
Anatomy orientation‑guided auto‑segmentation of bones from hip 3T magnetic resonance imaging
2026-02-12
articleSenior authorArchives of Pathology & Laboratory Medicine · 2026-02-11
articleOpen accessContext.—: Accurate intraoperative assessment of macrovesicular steatosis in donor liver biopsies is critical for transplant decisions but is often limited by interobserver variability and freezing artifacts that can obscure histologic details. Artificial intelligence (AI) offers a potential solution for standardized and reproducible evaluation. Objective.—: To evaluate the diagnostic performance of 2 self-supervised learning (SSL)-based foundation models, Prov-GigaPath and UNI, for classifying macrovesicular steatosis on frozen liver biopsy sections, compared with assessments by surgical pathologists. Design.—: This retrospective study included 131 frozen liver biopsy specimens from 68 donors collected between November 2022 and September 2024. Slides were digitized into whole slide images, tiled into patches, and used to extract embeddings with Prov-GigaPath and UNI; slide-level classifiers were then trained and tested. Intraoperative diagnoses by on-call surgical pathologists were compared with ground truth determined from independent reviews of permanent sections by 2 liver pathologists. Accuracy was evaluated for both a 5-category classification and a clinically significant binary threshold (<30% versus ≥30%). Results.—: For the binary classification, Prov-GigaPath achieved 96.4% accuracy, UNI 85.7%, and surgical pathologists, 89.3% (P = .37). For the 5-category classification, accuracies were lower: Prov-GigaPath, 57.1%; UNI, 50.0%; and pathologists, 64.2% (P = .47). Misclassification occurred mainly in intermediate categories (5% to <30% steatosis). Conclusions.—: SSL-based foundation models performed comparably to surgical pathologists at the clinically relevant threshold of less than 30% versus 30% or greater. These findings support the potential role of AI in standardizing intraoperative evaluation of donor liver biopsies; however, the small sample size limits generalizability and requires validation in larger, balanced cohorts.
Open Forum Infectious Diseases · 2026-01-11
articleOpen accessAbstract Background It remains unclear if cure of hepatitis C virus (HCV) infection with direct-acting antivirals (DAAs) leads to change in liver fat fraction and if this change differs from people without HCV observed over similar time. We evaluated the change in three-dimensional volumetric liver fat fraction by magnetic resonance imaging (MRI) prior to DAA treatment and 18 months following initiation and compared changes in uninfected controls over 18 months. Methods We conducted a cohort study of 35 participants who initiated DAAs and achieved cure and 42 without HCV infection as controls. At enrollment and 18 months later, participants had measurements of height, body weight, and three-dimensional volumetric liver fat fraction and cross-sectional area of abdominal subcutaneous and visceral adipose tissue by MRI. Multivariable linear regression was used to estimate group differences in mean changes in liver fat fraction. Results Mean change in three-dimensional volumetric liver fat fraction between Months 0 and 18 was +2.89% (95% CI: + 0.89%, + 4.88%) among participants with cured HCV and +0.48% (95% CI: -1.02%, + 1.97%) among controls (between group difference p-value=0.05). After adjustment for age, sex, smoking, and change in body mass index, the mean difference in liver fat fraction change between Months 0 and 18 was +3.11% (95% CI: + 0.46%, + 5.76%; p = 0.022) higher for participants with HCV versus controls. Conclusions Three-dimensional volumetric liver fat fraction by MRI significantly increased between Months 0 and 18 following successful treatment of HCV infection, but this finding was not observed from Month 0 to 18 among participants without HCV infection.
Magnetic Resonance in Medicine · 2025-10-29 · 1 citations
articleOpen accessSenior authorABSTRACT Purpose Pediatric craniofacial imaging may involve examination of both the skull and brain tissues via CT and MRI, respectively. DREAMER (Dual Repetition and Echo Acquisition with Multi‐contrast Encoding and Reconstruction) simultaneously acquires solid‐ and soft‐tissue images, potentially providing a rapid, high‐resolution, and radiation‐free protocol whenever bone‐selective, T 1 w, and T 2 w images are required. Methods The DREAMER sequence combines a solid‐state MRI method with phase‐based T 2 encoding to produce a multi‐contrast signal model that enables retrospective customization of image contrast weighting. DREAMER is paired with an iterative image reconstruction algorithm for accelerated and high‐resolution structural imaging of solid‐ and soft‐tissue compartments. Two healthy adult volunteers and two pediatric patients were scanned at 3 T to qualitatively compare soft‐tissue DREAMER image contrasts with their corresponding clinical standards, T 1 w MPRAGE and T 2 w fast spin‐echo (FSE). Two patients also underwent clinical CT to compare the bone‐selective images and skull renderings. Results DREAMER images are self‐registered, high‐resolution, and spatially isotropic. The bone‐selective, T 1 w, and T 2 w images approximate the image contrasts and structural imaging capabilities of their corresponding clinical standards (CT, T 1 w MPRAGE, and T 2 w FSE). Unlike the standard techniques, DREAMER imaging occurs at a single scanner using a single pulse sequence. Conclusion DREAMER combines mechanisms for solid‐ and multiple‐contrast soft‐tissue imaging into a single scan. For craniofacial imaging, DREAMER may consolidate CT and MRI demand, reduce exposure to ionizing radiation, decrease patient examination and wait times, and simplify the radiological workflow.
medRxiv · 2025-09-17
preprintOpen accessAbstract Introduction Accurate intraoperative assessment of macrovesicular steatosis in donor liver biopsies is critical for transplantation decisions but is often limited by inter-observer variability and freezing artifacts that can obscure histological details. Artificial intelligence (AI) offers a potential solution for standardized and reproducible evaluation. To evaluate the diagnostic performance of two self-supervised learning (SSL)-based foundation models, Prov-GigaPath and UNI, for classifying macrovesicular steatosis in frozen liver biopsy sections, compared with assessments by surgical pathologists. Methods We retrospectively analyzed 131 frozen liver biopsy specimens from 68 donors collected between November 2022 and September 2024. Slides were digitized into whole-slide images, tiled into patches, and used to extract embeddings with Prov-GigaPath and UNI; slide-level classifiers were then trained and tested. Intraoperative diagnoses by on-call surgical pathologists were compared with ground truth determined from independent reviews of permanent sections by two liver pathologists. Accuracy was evaluated for both five-category classification and a clinically significant binary threshold (<30% vs. ≥30%). Results For binary classification, Prov-GigaPath achieved 96.4% accuracy, UNI 85.7%, and surgical pathologists 84.0% ( P = .22). In five-category classification, accuracies were lower: Prov-GigaPath 57.1%, UNI 50.0%, and pathologists 58.7% ( P = .70). Misclassification primarily occurred in intermediate categories (5%–<30% steatosis). Conclusions SSL-based foundation models performed comparably to surgical pathologists in classifying macrovesicular steatosis, at the clinically relevant <30% vs. ≥30% threshold. These findings support the potential role of AI in standardizing intraoperative evaluation of donor liver biopsies; however, the small sample size limits generalizability and requires validation in larger, balanced cohorts.
Machine Learning Prediction Of Atypical Femur Fractures In Bisphosphonate-Treated Patients
Scholarly Commons (University of Pennsylvania) · 2025-09-15
otherOpen accessSenior authorBisphosphonates are the standard treatment for osteoporosis, reducing fracture risk by inhibiting bone resorption. However, long-term use has been associated with rare but serious adverse events, such as atypical femur fractures (AFF). This project employs machine learning (ML) to predict the occurrence of these adverse events by analyzing clinical data that includes bisphosphonate type, dosage, treatment duration, demographic factors, and patient comorbidities. Using anonymized electronic medical records from 139,376 encounters (24,675 unique patients), 920 AFF cases were identified. Multiple ML algorithms—including K Nearest Neighbors (KNN), Boosted Tree, Decision Tree, and Bootstrap Forest—were trained and validated to identify the most effective model. The KNN model demonstrated the highest predictive performance, accurately distinguishing between patients with and without AFF. In the test set, it achieved near-perfect classification for non-AFF cases and correctly identified 89.7% of AFF cases. These findings highlight the potential of ML to identify patients at elevated risk for adverse events, enabling clinicians to balance therapeutic benefits with long-term safety considerations. By supporting data-driven, individualized treatment decisions, this approach could improve the clinical management of osteoporosis and minimize the risks associated with extended bisphosphonate use.
Simultaneous solid and multiple-contrast soft tissue musculoskeletal magnetic resonance imaging
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 authorMotivation: MRI and CT are widely used for clinical assessment of musculoskeletal pathology. These modalities provide complementary information, but the scheduling of separate scans creates a logistically cumbersome scenario. Goal(s): To demonstrate the capabilities of a specialized pulse sequence that yields bone-selective and soft-tissue contrasts (T1-weighted and T2-weighted) in a single scan. Approach: DREAMER is a dual-echo UTE sequence combined with a phase-based encoding strategy to simultaneously obtain bone-selective, T1-weighted, and T2-weighted images. Imaging was performed at the knee, calf, and foot in healthy volunteers. Results: DREAMER enables a rapid and radiation-free method to assess bony anatomy and surrounding soft tissue for musculoskeletal applications. Impact: DREAMER performs simultaneous imaging of musculoskeletal solid and soft tissues, potentially obviating the need for additional CT. It can reduce the use of ionizing radiation in clinical imaging and remove logistical complexities related to scheduling examinations.
Implications of inflammation: Bone and joint health with COVID-19
Elsevier eBooks · 2025-01-01
book-chapterModel for Calculating Impact Force for Individualized Hip Fracture Prediction During a Fall
Advances in Orthopedics · 2025-01-01
articleOpen accessSenior authorOsteoporotic‐related weakening of bone is a common cause of hip fractures. The standard of care for the diagnosis and management of osteoporosis is the dual‐energy x‐ray absorptiometry bone mineral density T‐scores. Many individuals considered nonosteoporotic, however, still sustain fractures since these tools do not incorporate vital bone parameters and subject‐specific characteristics. The purpose of this work was to (1) develop a simple analytical model for estimating the force exerted on the femur during a fall (i.e., impact force) based on measurable patient metrics and (2) define a quantifiable fracture risk index by comparing finite‐element‐derived bone strength and impact force, which could be validated in a cohort of human subjects. Aggregated regression models were derived for estimating impact force based on patient age, weight, height, and soft tissue thickness. Patients with a history of hip fractures were then compared to a matched nonfracture group via the bone strength index (BSI), defined as the ratio between bone strength and maximum impact force. The BSI was lower in the fracture group compared to the control group by 0.23 ( p = 0.045). The combination of patient‐specific impact force on the femur during a fall and bone strength could provide additional insights into osteoporotic hip fracture risk alongside standard risk assessments.
Elsevier eBooks · 2025-01-01
book-chapter
Recent grants
NSF · $279k · 2020–2024
Clinical Assessment of Hip Fracture Biomechanics using MRI
NIH · $1.8M · 2015–2022
MRI Assessment of Hip Fracture Risk and Therapy Response in HIV/HCV Coinfection
NIH · $2.7M · 2019–2026
NIH · $491k · 2015
NIH · $167k · 2016
Frequent coauthors
- 48 shared
Félix W. Wehrli
- 25 shared
Jason D. Wink
Penn Presbyterian Medical Center
- 25 shared
Rami D. Sherif
Dallas Plastic Surgery Institute
- 25 shared
Hyun‐Duck Nah
Children's Hospital of Philadelphia
- 25 shared
Youngshin Lim
Cedars-Sinai Medical Center
- 25 shared
Patrick A. Gerety
Creative Commons
- 25 shared
Jesse A. Taylor
Children's Hospital of Philadelphia
- 25 shared
Nadya A. Clarke
University of Pennsylvania
- Resume-aware match score
- Save to shortlist
- AI-drafted outreach
See your match with Chamith S. Rajapakse
PhdFit ranks faculty by your research interests, methods, and publications — grounded in their actual work, not templates.
- Free to start
- No credit card
- 30-second signup