Navid Toufan
Ph.D. Candidate in Electrical and Computer Engineering, University of Manitoba
Generative deep learning for brain MRI
About Me
I’m a Ph.D. candidate in Electrical and Computer Engineering at the University of Manitoba, supervised by Dr. Ahmed Ashraf. My research focuses on deep generative models for medical imaging, specifically how we can synthesize, inpaint, and reconstruct brain MRI when parts of a scan are missing or obscured by pathology.
My path into medical AI has been a winding one. I started in information technology, ran my own IT and home-automation businesses in Kerman, Iran, and later completed an M.Sc. in Artificial Intelligence and Robotics, where I developed evolutionary algorithms for multi-robot path planning. Today I bring that mix of hands-on engineering and optimization research to problems in medical image analysis.
Alongside research, I’ve taught for five terms as a teaching assistant in signal processing, numerical methods, and applied computational intelligence. Helping students get from confusion to understanding is one of the most rewarding parts of my work.
Research
My Ph.D. thesis, Generative Deep Learning for Brain MRI Synthesis, Inpainting, and Missing-Modality Reconstruction, explores how generative models can fill in what a scan doesn’t show.
Healthy-tissue inpainting in brain MRI. For the BraTS 2026 Local Inpainting Challenge, I designed a mask-gated 3D U-Net with void-size conditioning that synthesizes realistic healthy brain tissue in tumour regions of T1 MRI. On the 219-case official validation set, the model achieved an SSIM of 0.832 and a PSNR of 23.37 dB, outperforming both official baselines. A custom on-the-fly mask augmentation scheme provided the largest single improvement, and a background-suppression loss cut tissue synthesized outside the brain by roughly two-thirds. The model reconstructs a full volume in about 0.3 seconds, and the code is openly available on GitHub. This work was accepted at MICCAI 2026.
Missing-modality reconstruction. I’m now extending this work to diffusion-based reconstruction of missing MRI modalities (T1, T2, and FLAIR), so that clinicians and downstream algorithms can work with complete data even when a sequence wasn’t acquired.
Transfer learning for histopathology. Earlier in my Ph.D., I studied transfer learning for classifying breast cancer histopathology images, adapting ImageNet-pretrained CNNs to the BreakHis and IDC datasets and reaching 86.3% accuracy and 0.92 AUC on IDC.
Metaheuristic optimization for robotics. During my M.Sc., I developed Grey Wolf Optimizer–based methods for robot path planning in unknown environments, tested through a custom V-REP simulator plugin.
Research interests: medical image analysis, diffusion models, VAEs, GANs, image inpainting and synthesis, transfer learning, metaheuristic optimization, robotics, and signal processing.
Publications
Toufan, N., & Ashraf, A. (2026). A mask-gated 3D U-Net with void-size conditioning for healthy tissue inpainting in brain MRI. In MICCAI 2026 Challenges: BraTS 2026 Local Inpainting, Lecture Notes in Computer Science. Springer. Accepted, in press. Open-access version
Toufan, N., & Niknafs, A. (2020). Robot path planning based on laser range finder and novel objective functions in grey wolf optimizer. SN Applied Sciences, 2, 1324. https://doi.org/10.1007/s42452-020-3093-5
Teaching
I’ve supported students across electrical engineering and computer science courses at the University of Manitoba:
ECE 4450 Applied Computational Intelligence (Fall 2026): leading weekly Python and PyTorch labs on deep, convolutional, and recurrent neural networks.
Signal Processing 1 (Winter 2024 – present, five terms): guiding students through DSP labs, Fourier analysis, and filter design in MATLAB.
Numerical Methods for Electrical Engineers (2024, 2025, 2026): leading tutorials and coaching students in MATLAB programming and numerical analysis.
COMP 3190 Introduction to Artificial Intelligence (Winter 2024): grading and written feedback.
Awards and Funding
Student Fellowship for Education Purposes, University of Manitoba
Edward R. Toporeck Graduate Fellowship in Engineering, University of Manitoba
Full scholarship, M.Sc. program, Shahid Bahonar University of Kerman (2016–2019)
Top 2% of 30,000+ candidates, Iran’s national Master’s entrance exam in computer engineering (2016)
Education
Ph.D., Electrical and Computer Engineering, University of Manitoba (expected 2028)
M.Sc., Computer Engineering (AI and Robotics), Shahid Bahonar University of Kerman (2019)
B.Sc., Information Technology Engineering
Get in Touch
I’m always happy to talk about generative models, medical imaging, or potential collaborations.
📧 navid.toufan@gmail.com
🔗 LinkedIn · GitHub
📍 Winnipeg, Manitoba, Canada