Skip to main content
Department of Information Technology

Image analysis seminars at Vi3

The Vi3 image analysis seminar series is hosted by the Centre for Image Analysis (CBA) and open to everybody interested. Join us in Theatrum Visuale, room 100155, building 10, Ångström Laboratory (or online at

Each Monday at 14:15-15:00, seminars on current research on image analysis and related topics are given by internal and external speakers. See schedule below.

The CBA seminars are announced at the weekly internal Vi3 information meeting, and also on the Dept. of IT's newsletter läsit

Note that extra seminars may occasionally be held at other times than mentioned above.


Date Time Title Speaker
2022-08-22 14:15-15:00 Increasing Trust in Alzheimer's Disease Forecasting
Alzheimer’s disease (AD) is the most common form of dementia, causing progressive cognitive impairment, disorientation, and memory loss. Despite years of clinical trials, there is currently no cure for AD and, by the time of diagnosis, as much as 60% of brain matter is already lost. Thus, forecasting Alzheimer's disease years ahead of onset is critical for attempts at early treatment, the selection of subjects for clinical trials, and to facilitate neurologists’ study of the disease. Past research on machine learning for Alzheimer’s disease prediction was limited to using cognitive test scores and highly engineered volumetric features, while failing to leverage the potential information found in brain MRIs. Attempts to train 3D and 2D CNNs on the MRIs have been unsuccessful thus far, due to the insufficient amount of data samples available for the massive amount of parameters that need to be learned. Moreover, forecasting AD using standard statistical models, simple MLPs and sequential models is typically limited to 6-12 months windows, model stability and performance dropping significantly for longer time windows. To introduce more complex longitudinal data and detect future disease stages, we propose a sequential deep learning approach that is expressive enough to handle multimodal longitudinal data, with an added unsupervised mutual information CNN encoder to process the 3D MRI scans, while still maintaining stable forecasting performance over forecasting windows of two years or more. We integrate certain domain knowledge via selective features of the disease-relative areas’ volumetric data, cognitive test scores and demographic information. The model is designed to get latent features from the multimodal data during the training process that are informative to the forecasting task. To achieve this, we propose a hybrid model of RNN and CNN. On the RNN side, we study a RNN-like structure introducing latent anticipated features to enhance the forecasting performance. On the CNN side, we train a mutual information based unsupervised encoder and extract latent features from the 3D MRIs as supportive side information. The model we are developing will be capable of performing forecasting tasks 2-year ahead of time, while the instability issue brought by long-term forecasting is addressed by our new mechanism of training different parts of the model separately in different stages. The future goal of the project lies in forecasting the disease on a longer time gap without loss of accuracy and robustness.
Madalina Fiterau
University of Massachusetts Amherst
2022-09-05 14:15-15:00 Title
Raphaela Heil
2022-09-12 14:15-15:00 Title
Eduard Chelebian Kocharyan
2022-09-19 14:15-15:00 Title
Swarnadip Chatterjee
2022-09-26 14:15-15:00 Title
Elisabeth Wetzer
2022-10-03 14:15-15:00 Title
Erik Hallström
2022-10-10 14:15-15:00 No seminar
IT Strategy Days
2022-10-17 14:15-15:00 Title
Eva Breznik
2022-10-24 14:15-15:00 Title
Karl Bengtsson Bernander
2022-10-31 14:15-15:00 Title
Andrea Behanova
2022-11-07 14:15-15:00 Title
2022-11-14 14:15-15:00 Title
2022-11-21 14:15-15:00 Title
2022-11-28 14:15-15:00 Title
2022-12-05 14:15-15:00 Title
2022-12-12 14:15-15:00 Title
2022-12-19 14:15-15:00 Title
Updated  2022-08-11 10:46:57 by Ingela Nyström.