Nargiza Djurabekova - Adria Casamitjana - CMIC seminar series

Nargiza Djurabekova - Adria Casamitjana (Universitat Politècnica de Catalunya, Barcelona Tech)

Adria Casamitjana, Universitat Politècnica de Catalunya, Barcelona Tech

Title: MRI and Machine learning: a tool for Preclinical AD screening.

Abstract: The identification of healthy individuals harboring amyloid pathology constitutes one important challenge for secondary prevention clinical trials in Alzheimer's disease. Consequently, noninvasive and cost-efficient techniques to detect preclinical AD constitute an unmet need of critical importance. We apply machine learning to structural MRI to identify amyloid-positive subjects. Models were trained on public ADNI data and validated on an independent local cohort. Used for subject classification in a simulated clinical trial setting, the proposed method is able to save 60% unnecessary CSF/PET tests and to reduce 47% of the cost of recruitment when used in a simulated clinical trial setting.

This recruitment strategy capitalizes on already acquired MRIs to reduce the overall amount of invasive PET/CSF tests in prevention trials, demonstrating a potential value as a tool for AD screening. This protocol could foster the development of secondary prevention strategies for AD.

Nargiza Djurabekova

Title: Variational framework for dynamic CBCT of the foot and ankle (Co-authors: Andrew Goldberg, Andreas Hauptmann, David Hawkes, Guy Long, Felix Lucka, Marta M. Betcke)

Abstract:

The foot and ankle is a complex structure of 28 bones and 30 joints that reacts to body-weight's pressure. Only recently, new conebeam CT scanners have been developed allowing imaging in seated and upright positions which are both, to a different degree, weight-bearing. This considerably enhanced the diagnostically relevant information and in turn led to better patient outcomes as compared to non-weight-bearing scans. However, much is left to be desired when it comes to interpreting the functional weight-bearing motion from these static scans. This is largely due to the lack of a precise biomechanical model. It would be unreasonable to subject patients to very large radiation doses by scanning the same foot dozens of times in different poses to acquire the said model. Nevertheless, understanding precise dynamical motion would be immensely helpful in performing correction surgeries.  This contribution  aims to reconstruct a dynamic phantom of the foot and ankle structure from what we assume to be two full rotation scans. With the first of these scans being static and the other dynamic, the total radiation dose is only double the regular static scan.

We pose this dynamic problem in the variational framework including the optical flow (also known as scene flow in 3D) and total variation constraints to compensate for the sparse measurements. In this context we consider optical flow to be the X-ray attenuation constancy over all time-frames. This formulation is solved using the Proximal Alternating Linearized Minimization (PALM) and its inertial counterpart iPALM developed for a broad class of nonconvex and nonsmooth optimization problems. Global convergence of both is guaranteed by the Kurdyka-\L ojasiewicz property of the combined data and optical flow terms.