Department of Mathematics

CoSy Seminars Spring 2016

24 May 
SpeakerVikram Sunkara, FU Berlin, Konrad Zuse Institute Berlin, PostDoc, Systems Pharmacology And Disease Control
Title: The Mathematical Challenges In Modelling the Degradation Process of Osteoarthritis.
Place: Å11167
Time: 12:00 -- 13:00
Local contact person: Stefan Engblom

17 May 
SpeakerAlison Ramage, Department of Mathematics and Statistics University of Strathclyde​
Title: A multilevel preconditioner for data assimilation with 4D-Var
Place: Å11167
Time: 12:00 -- 13:00

15 Mar 
SpeakerCarl-Fredrik Westin, The founding director of the Laboratory of Mathematics in Imaging, and Associate Professor of Radiology, Harvard Medical School, Boston. http://ssba2016.cb.uu.se/speaker.php
TitleMicrostructure imaging of the human brain using diffusion MRI 
Place: MIC Aula
Time: 09:00 -- 10:00
 

26 April (Canceled)
SpeakerAlex Sverdlov, PhD, Senior Expert Statistician, EMD Serono Inc., Billerica MA
TitleResponse-Adaptive Randomization Designs in Clinical Trials: a Worthy Challenge
Place: TBA (maybe in Ångström 11167)
Time:12.00-13.00
Local contact person: Yevgen Ryeznik

Date: 15 Mar 
Speaker: Carl-Fredrik Westin, The founding director of the Laboratory of Mathematics in Imaging, and Associate Professor of Radiology, Harvard Medical School, Boston. http://ssba2016.cb.uu.se/speaker.php
Title: Microstructure imaging of the human brain using diffusion MRI 
Place: MIC Aula
Time: 09:00 -- 10:00
Abstract:
Diffusion magnetic resonance imaging (dMRI) has become an essential tool over the last two decades, because it provides unique insight into both the microstructure and structural connectivity of the brain. In dMRI, each millimeter-size voxel contains information on the micrometer-scale translational displacements of water molecules whose diffusivity is directionally restricted by barriers such as axonal membranes and myelin packing. Recent developments in dMRI can help us understand the structure and function of the human brain and have applications in a wide range of diseases. 

Date: 26th April, 2016
Speaker:  Alex Sverdlov, PhD, Senior Expert Statistician, EMD Serono Inc., Billerica MA
Title: Response-Adaptive Randomization Designs in Clinical Trials: a Worthy Challenge
Abstract:  Response-adaptive randomization is a class of randomization designs for clinical trials for which treatment allocation probabilities are sequentially modified based on accumulating patient outcome data in the trial to achieve selected experimental objectives while maintaining validity and integrity of the trial results. Clinical trials with response-adaptive randomization can frequently be more flexible, more efficient, and more ethical than traditional fixed randomization designs. However, response-adaptive randomization designs are operationally more complex, can have higher vulnerability to experimental bias, and lead to more complex statistical inference than fixed randomization designs.
In this talk, I will give a perspective on the types of clinical trials where response-adaptive randomization can be applied with most net benefit. These scenarios include trials for rare and fatal diseases (such as childhood cancers) and trials for highly contagious diseases (such as Ebola) where a great proportion of the population of patients with the disease will receive treatment as part of the trial protocol and where there is a strong ethical imperative to increase the chance of a trial participant to be assigned to the empirically better treatment while achieving other objectives as well. I will also cover another important area of application of response-adaptive randomization—multi-arm comparative studies (such as dose-ranging phase II clinical trials) where the objective may be to identify “most informative” dose levels such as the minimal efficacious dose (MED) or the dose producing 95% of the treatment effect (ED95). Some recently published theoretical and simulation results will be presented to illustrate utility and merit of response-adaptive randomization in the context of multi-arm clinical trials with time-to-event outcomes. I will also cover some practical aspects of designing response-adaptive randomized trials, statistical software, and recent perspectives of the Health Authorities (FDA, EMA) on response-adaptive randomization.
References:1. Sverdlov, O., Tymofyeyev, Y., & Wong, W. K. (2011). Optimal response‐adaptive randomized designs for multi‐armed survival trials. Statistics in Medicine, 30(24), 2890-2910.
2. Sverdlov, O., Ryeznik, Y., & Wong, W. K. (2014). Efficient and ethical response-adaptive randomization designs for multi-arm clinical trials with Weibull time-to-event outcomes. Journal of Biopharmaceutical Statistics, 24(4), 732-754.
3. Ryeznik, Y., Sverdlov, O., & Wong, W. K. (2015). RARtool—a MATLAB software package for designing response-adaptive randomized clinical trials with time-to-event outcomes. Journal of Statistical Software, accepted; to appear.
4. Sverdlov, O. (editor) (2015). Modern Adaptive Randomized Clinical Trials: Statistical and Practical Aspects. ISBN 9781482239881. Chapman & Hall/CRC Press, Boca Raton, FL, to appear July 29, 2015. http://www.crcpress.com/product/isbn/9781482239881

17 May 
SpeakerAlison Ramage, Department of Mathematics and Statistics University of Strathclyde​
Title: A multilevel preconditioner for data assimilation with 4D-Var
Place: Å11167
Time: 12:00 -- 13:00
Abstract:
Large-scale variational data assimilation problems are commonly found in applications like numerical weather prediction and oceanographic modelling. The 4D-Var method is frequently used to calculate a forecast model trajectory that best fits the available observations to within the observational error over a period of time. One key challenge is that the state vectors used in realistic applications could contain billions or trillions of unknowns so, due to memory limitations, in practice it is often impossible to assemble, store or manipulate the matrices involved explicitly. In this talk we present a limited memory approximation to the Hessian of the linearised quadratic minimisation subproblems, computed using the Lanczos method, based on a multilevel approach. We then use this approximation as a preconditioner within 4D-Var and show that it can reduce memory requirements and increase computational efficiency.

24 May 
SpeakerVikram Sunkara, FU Berlin, Konrad Zuse Institute Berlin, PostDoc, Systems Pharmacology And Disease Control
Title: The Mathematical Challenges In Modelling the Degradation Process of Osteoarthritis.​
Place: Å11167
Time: 12:00 -- 13:00
Local contact person: Stefan Engblom
Abstract:
Osteoarthritis(OA) is a degenerative disease that affects the majority of individuals in the later stages of their life. Osteoarthritis of the knee, hip and spine are particularly common. They induce severe pain and reduce mobility up to the stage where individuals cannot pursue their day-to-day duties. Currently, there is no effective treatment available. Joint replacement is thus the method of last resort to ensure patients’ autonomy and life-quality. One of the major challenges of mathematically modelling OA is in describing the slow time scale of the cartilage degradation. During the process of cartilage degradation individuals do not show any significant symptoms. Individuals unknowingly continue with regular mobility until all cartilage is lost and bone to bone contact is made. At this stage only major surgery is a viable option. Hence, modelling the interim process of cartilage degradation is very critical for possible intervention and therapy design. In this talk we will be exploring the current in vitro studies being conducted to study osteoarthritis in mice. Furthermore,  we will introduce a preliminary multi scale model, were cartilage homeostasis is a balancing act between mechanical forces, mechanical properties and cellular diversity. We will also explore the mathematical challenges in extrapolating results from in vitro studies in mice to design successful therapies for human.