Mehmet Yigit Avci

I am a PhD student at King's College London, funded by DRIVE-Health and deepc.

I earned my BSc in Electrical & Electronics Engineering from Bogazici University, where I researched Alzheimer's detection using brain connectomes in VAVlab under the guidance of Prof. Burak Acar.

I have worked with Prof. Berkin Bilgic and Prof. Qiyuan Tian at the Athinoula A. Martinos Center for Biomedical Imaging, focusing on the intersection of diffusion MRI and deep learning.

Additionally, I completed an internship with the COMPAI Lab under the supervision of Prof. Julia Schnabel, Dr. Veronika Zimmer, and Dr. Cosmin Bercea, working on unsupervised representation learning for Alzheimer's disease.

Currently, I am supevised by Dr. Jorge Cardoso on AI-based smart orchestration for my PhD studies.

Medical Image Analysis, MRI, Deep Learning, Unsupervised Learning, Multi-Modal Learning

PhD, King's College London 2024-2028

Msc. in Biomedical Computing, TUM 2022-2024

Bsc. in Electrical and Electronics Engineering, Bogazici University 2017-2022

October 2024 - Started my PhD at King's College.

October 2024 - Finished my my Master's Degree in Biomedical Computing at TUM!

October 2024 - Received Best Paper Award for our work on Unsupervised Alzheimer's Disease Analysis in MICCAI EMERGE Workshop!

September 2022 - Started at TUM for my Master's Degree in Biomedical Computing.

September 2022 - Started working as a Machine Learning Intern at deepc, working on benchmarking radiology AI, detection of body part with deep learning and NLP solutions.

June 2022 - Graduated from Bogazici University with a BSc degree Electrical and Electronics Engineering.

May 2022 - Presented my first abstract in ISMRM 2022.

We present MORPHADE (Morphological Autoencoders for Alzheimer's Disease Detection), a novel unsupervised learning approach which uses deformations to allow the analysis of 3D T1-weighted brain images. To the best of our knowledge, this is the first use of deformations with deep unsupervised learning to not only detect, but also localize and assess the severity of structural changes in the brain due to AD. We believe our framework shows promise as a tool towards improved understanding, monitoring and detection of AD.

This work's goal is to provide a simulation framework for routine neuroimaging test data, which allows for "stress testing" of deep segmentation networks against acquisition shifts that commonly occur in clinical practice for T2w FLAIR MRI protocols.

Participated in QUAD Challenge in MICCAI 2022 and got the 4th place with combination of methods DUnet and deepDTI.

Dropout is conventionally used during the training phase as regularization method and for quantifying uncertainty in deep learning. We propose to use dropout during training as well as inference steps, and average multiple predictions to improve the accuracy, while reducing and quantifying the uncertainty. With our method, accuracy can be improved significantly compared to network outputs without dropout, especially when the training dataset is small. Moreover, confidence maps are generated which may aid in diagnosis of unseen pathology or artifacts.

Template is from Jon Barron's awesome website.