Learning to understand medical images
, Imperial College London
Date: Friday, September 12, 2014
Time: 11:00 AM to 12:00 PM Note: all times are in the Eastern Time Zone
Host: Kayhan Batmanghelich, CSAIL
Contact: Kayhan Batmanghelich, email@example.com
Speaker URL: None
TALK: Learning to understand medical images
Image understanding is an area of research in which the aim is to develop
methods and tools for automatic extraction of semantic information from raw
images. Over the last couple of years, we have been working on a variety of
medical applications using machine learning techniques that enable us to
automatically detect, localize, and segment anatomical structures in
images. In this presentation, I will be giving a couple of examples in
which the same underlying learning framework, randomized decision trees,
has been successfully applied to very different imaging tasks.
The first example will demonstrate our system for automatic localization
and identification of individual vertebrae in pathological spine CT. The
system has been evaluated on a large dataset of patients that has been made
publicly available. In the context of computational spine imaging, we will
discuss how longitudinal registration can greatly benefit from vertebrae
location priors which are automatically extracted from the images using our
In the second example, I will discuss our work on brain tumour segmentation
in multi-channel MRI using contextual features and tissue priors. The
method has been applied to images of patients suffering from Glioblastoma.
Using our approach, we were able to achieve the overall best performance in
the MICCAI 2012 BRATS challenge.
In an outlook, I will present some recent results from ongoing research in
other areas of image understanding such as detection and segmentation of
lesions in traumatic brain injury.
Created by Polina Golland at Wednesday, September 10, 2014 at 10:06 PM.