BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Generative Medical Image Segmentation Using Distance Transforms - 
 Lea Bogensperger\, TU Graz
DTSTART:20240919T120000Z
DTEND:20240919T130000Z
UID:TALK220795@talks.cam.ac.uk
CONTACT:Ferdia Sherry
DESCRIPTION:Medical\n image segmentation is a crucial task that relies on 
 the ability to accurately identify and isolate regions of interest in medi
 cal images. Thereby\, generative approaches allow to capture the statistic
 al properties of segmentation masks that are dependent on\n the respective
  structures. We employ two different generative modeling frameworks to rep
 resent the signed distance function (SDF) leading to an implicit distribut
 ion of segmentation masks: a conditional score-based generative model and 
 an image-guided conditional\n flow matching model. The advantage of levera
 ging the SDF is a more natural distortion when compared to that of binary 
 masks. By learning a vector field that is directly related to the probabil
 ity path of a conditional distribution of SDFs\, we can accurately\n sampl
 e from the distribution of segmentation masks\, allowing for the evaluatio
 n of statistical quantities. Thus\, this probabilistic representation allo
 ws for the generation of uncertainty maps represented by the variance\, wh
 ich can aid in further analysis\n and enhance the predictive robustness.
LOCATION:MR2 Centre for Mathematical Sciences
END:VEVENT
END:VCALENDAR