BEGIN:VCALENDAR
VERSION:2.0
PRODID:icalendar-ruby
CALSCALE:GREGORIAN
X-WR-CALNAME:Physical Mathematics Seminar
X-WR-TIMEZONE:Eastern Time (US & Canada)
BEGIN:VEVENT
DTSTAMP:20260908T005447Z
UID:tag:localist.com\,2008:EventInstance_51090467842610
DTSTART:20251028T183000Z
DTEND:20251028T193000Z
DESCRIPTION:Speaker: Mazdak Abulnaga (MIT & Harvard Medical School)\n\nTitl
 e: Machine Learning for Neuroimage Registration\n\nAbstract:\n\nAligning b
 rain images from magnetic resonance imaging (MRI) is a core challenge in n
 euroscience. The alignment problem\, known as registration\, allows us to 
 compare anatomy across individuals\, build population averages\, and study
  how structure and function varies with disease or development. Registrati
 on is posed as a nonlinear optimization problem with two competing objecti
 ves: aligning brain structure while maintaining a smooth deformation field
 . \n\nMachine learning has emerged as a powerful alternative to classical 
 optimization-based registration\, achieving both faster and more accurate 
 alignment. In this talk\, I will first introduce the use of machine learni
 ng for neuroimage registration\, then present two of our recent frameworks
  extending this paradigm.\n\nWe first propose MultiMorph\, a fast and effi
 cient method for constructing anatomical atlases on the fly. Atlases captu
 re the canonical structure of a collection of images and are essential for
  quantifying anatomical variability across populations. MultiMorph is a fe
 edforward model that rapidly produces population-specific atlases in a sin
 gle forward pass for any 3D brain dataset\, without any fine-tuning or opt
 imization. The model is based on a linear group-interaction layer that agg
 regates and shares features within the group of input images.\n\nNext\, I 
 will describe recent work tackling the problem of jointly registering the 
 cortical surface of the brain and the interior volume. While conventional 
 volumetric registration methods such as MultiMorph capture global brain al
 ignment\, they are limited in aligning cortical structure. The cortex is a
  highly folded\, curved surface best represented as a triangular mesh. Euc
 lidean alignment methods fail to capture this geometry. We propose a frame
 work that jointly aligns the cortical and subcortical regions through a un
 ified volume-and-surface-based representation. We do so by jointly learnin
 g a volumetric and spherical alignment\, producing consistent mappings acr
 oss cortical and subcortical regions.\n\nI will conclude by discussing ope
 n research directions of volumetric shape registration in fields beyond ne
 uroimaging\, including computer graphics\, medical imaging\, and computati
 onal biology.
GEO:42.358262;-71.090045
LOCATION:Building 2\, 449
SUMMARY:Physical Mathematics Seminar
URL;VALUE=URI:https://calendar.mit.edu/event/physical-mathematics-seminar-M
 azdak-Abulnaga
CATEGORIES:Conferences/Seminars/Lectures
END:VEVENT
END:VCALENDAR