Extremal Contours: Gradient-driven contours for compact visual attribution
Training-free method that uses gradient-driven smooth contours for compact visual attribution.
Reza Karimzadeh
Reza Karimzadeh
Computer Scientist
Professional Summary
I am a Ph.D. candidate in Computer Science at the University of Copenhagen, specializing in computer vision, deep learning, and medical image analysis. My work focuses on building robust, end-to-end machine learning systems, with expertise in 3D segmentation, self-supervised learning, object detection, domain generalization, and explainable AI. I design and implement solutions across the full ML lifecycle — from data engineering and modeling to evaluation, reproducibility, and deployment.
Alongside research, I have industry and hospital collaboration experience developing production-oriented vision systems and leading small technical teams in interdisciplinary, international environments. I am particularly motivated by applying strong ML engineering and problem-solving skills to real-world products where robustness, interpretability, and measurable impact are central.
If you are interested, please reach out to collaborate 😃
Outside of research, I enjoy solving mechanical puzzles, exploring new technologies, and studying psychology and human behavior. Currently, I am reading evolutionary biology (mostly Richard Dawkins books) out of curiosity about complex adaptive systems and the evolution behind them.
Education
PhD Computer Science
Copenhagen University
MS in Biomedical Engineering
Sharif University of Technology
BS in Biomedical Engineering
Amirkabir University of Technology
Interests
Image Segmentation
Medical Data Analysis
Explainable AI
Self-Supervised Learning
Object Detection
Domain Adaptation
Image Processing
Deep Learning
Machine Learning
Computer Vision
Experience
Machine Learning Intern | Research Intern
Amsterdam UMC | VU AmsterdamDeveloped and validated a Siamese deep learning model to predict cancer treatment response from pre- and post-treatment CT scans (n=502). Built an end-to-end training and cross-validation pipeline using PyTorch and MONAI, collaborating closely with oncologists and radiologists to ensure clinical relevance and interpretability. Led a cross-institutional university–hospital project.Machine Learning Intern | Research Intern (Remote)
Geneva University HospitalBuilt a multi-center deep learning system to predict cancer subtype from PET scans using set-based modeling (Transformers, Deep Sets, MIL, GNN). Achieved ROC-AUC up to 0.91 in cross-validation and improved external performance from 0.81 to 0.87 via test-time adaptation (entropy minimization). Led a fully remote university–hospital collaboration.Deep Learning Engineer
PART Artificial Intelligence StartupDeveloped and deployed YOLO-based object detection models for license plate detection and signature localization, achieving mAP0.5 of 0.97 and 0.94 respectively. Optimized robustness under varying lighting and quality conditions. Containerized and deployed models using Docker with real-time inference APIs in production.Technical Lead | Computer Vision & Signal Processing Engineer
Iran’s National Elites FoundationLed the computer vision group in developing a non-contact heart rate estimation system from RGB video (3.02 MAE). Designed signal processing pipelines including ROI tracking, temporal filtering, and amplitude normalization. Coordinated deployment on mobile and embedded platforms for real-time inference.Engineering Intern
Electro-Xray CompanyPerformed repair and maintenance of medical imaging systems including CT scanners, portable radiology devices, C-Arm systems, mammography units, and OPG systems.Teaching Assistant
Copenhagen University | Sharif University of Technology | Amirkabir University of TechnologyTeaching assistant for courses in Deep Learning, Numerical Methods, Medical Image Analysis and Processing, Image Processing, and Medical Imaging Systems.
Education
PhD Computer Science
Copenhagen UniversityApplications of Deep Learning Algorithms for Medical Data AnalysisMS in Biomedical Engineering
Sharif University of TechnologyWorked on Medical Image Segmentation Using Deep Learning MethodsRead Thesis (Persian)BS in Biomedical Engineering
Amirkabir University of TechnologyDesign and implementation of brain surgery bipolar electrocoagulation simulator using haptic technology.Read Thesis (Persian)
Featured Publications
•
•
1 min read
PiMPiC: An Overlap-Aware Contrastive Learning Framework for 3D Patch-Based Medical Image Segmentation
Contrastive learning framework with overlap-aware modules for improved 3D medical image segmentation.
Reza Karimzadeh
Recent Publications
(2025).
Extremal Contours: Gradient-driven contours for compact visual attribution.
arXiv:2511.01411.
(2025).
PiMPiC: An Overlap-Aware Contrastive Learning Framework for 3D Patch-Based Medical Image Segmentation.
DEMI 2025.
(2025).
CARMA: Collocation-Aware Resource Manager with GPU Memory Estimator.
arXiv:2508.19073.
(2025).
Prediction of Radiological Diagnostic Errors from Eye Tracking Data Using Graph Neural Networks and Gaze-Guided Transformers.
GRAIL 2024.
(2024).
Generating Synthetic Computed Tomography for Radiotherapy: SynthRAD2023 Challenge Report.
Med. Image Anal..
Recent & Upcoming Talks
Extremal Contours
Extremal Contours; Gradient-driven contours for compact visual attribution
PiMPiC Framework
PiMPiC; An Overlap-Aware Contrastive Learning Framework for 3D Patch-Based Medical Image Segmentation