Utku Tahan Email & Phone Number

Eindhoven, Nb, Nl

Eindhoven, North Brabant, Netherlands

Thesis: Low-field MRI offers a compact, cheap, and potentially widely-accessible alternative to its high-field counterpart but unfortunately suffers from low image quality. To overcome this, we propose to improve low-field MRI image quality by probabilistic inference using deep generative models based on score-based diffusion processes. To draw accurate samples from a posterior data distribution given low-signal-to-noise ratio measurements, score-based diffusion models solve reversed stochastic differential equations starting from a tractable (usually Gaussian) prior distribution. We observe that direct posterior sampling yields high-quality images but is also prone to hallucinate features due to high posterior entropy given highly noisy measurements. We propose suppressing such hallucinations by computing the per-pixel mean of multiple posterior image samples and subsequently refining this intermediate estimate with another reverse diffusion process. Our final scheme then cascades several such steps. Finally, we evaluate our proposed method against competitive baselines, including the vanilla diffusion scheme and supervised deep learning methods based on a U-Net and deep unfolded proximal gradient network. We test denoising performance on images obtained under field strengths 0.5T, 0.75T, and 1.0T. The result shows that our proposed improvements beat the baseline diffusion method in terms of reconstruction quality, and they can compete against their fully supervised counterparts in the low-field MRI image denoising task

Eindhoven

I applied deep learning algorithms to low field MRI images to increase their resolution. To do that, I used deep unfolded network where proximal operator replaced by U-Net.

Ankara, Turkey

Simulation of electromagnetic problems such as optical wire transmission in optical nano-couplers.

Ankara, Turkey

I worked as a signal processing engineer internee, and implemented filtering algorithms on FPGA by using Xilinx VHDL

Ankara, Turkey

Master'S Degree, Electrical And Electronics Engineering, 8.28/10.00

Thesis: Low-field MRI is a compact, and cheap alternative to high-field MRI, yet it has poor image quality. We suggest using score-based.

Graduated, Electrical And Electronics Engineering, Gpa: 3.54/4.00

Specialized in signal processing and telecommunication engineering at senior year