Medical image analysis at the Biomedical Imaging Division at KTH
Abstract
Description:
This project collects 3 different subprojects from the Division of Biomedical Imaging at KTH that require the use of Berzelius and are linked to PhD students enrolled in the WASP affiliated PhD program. This is a brief description of them:
1. Deep learning-based breast image analysis (PhD student Zhikai Yang). Goal: A) use diffusion and ViTs for the detection, segmentation, and classification of breast cancer images. B) Generation of mammograms from tomosynthesis data with AI. C) Perform image reconstruction of digital tomosynthesis with primal-dual neural networks.
2. Synthetic generation of longitudinal MRI using aging templates (Previous PhD student Jingru Fu). Goal: use deep learning-based diffeomorphic registration to generate age-specific templates of subjects that can be used to predict future MRI scans of subjects.
3. Reinforcement learning-based tractography (Previous PhD student Fabian Sinzinger). This project is finished
4. Generative models for diffusion MRI (PhD student Sanna Persson). Goal: use generative models to estimate diffusion MRI data with a given b-value (a scanning parameter) from images acquired with a different b-value. These images will be used to predict the mechanical properties of brain tissue.
All subprojects use anonymized data without sensitive information or synthetic data.
Impact:
All sub-projects aim to improve the quality of the images, increase accuracy, or add extra information that can potentially be used by clinicians in the future, which will potentially benefit the healthcare system and society in general.
Continuation projects:
This is a continuation of Berzelius-2024-410.
Relationship with KAW strategic initiatives:
The 4 PhD students from this project are affiliated with WASP and are part of the WASP graduate school. Jingru Fu and Fabian Sinzinger graduated (10-01-2025) and (29-11-2024). They now work in the group as research engineers. Zhikai Yang and Sanna Persson are still with WASP.
