BEGIN:VCALENDAR
VERSION:2.0
PRODID:icalendar-ruby
CALSCALE:GREGORIAN
METHOD:PUBLISH
BEGIN:VTIMEZONE
TZID:America/New_York
BEGIN:DAYLIGHT
DTSTART:20190310T030000
TZOFFSETFROM:-0500
TZOFFSETTO:-0400
RRULE:FREQ=YEARLY;BYDAY=2SU;BYMONTH=3
TZNAME:EDT
END:DAYLIGHT
BEGIN:STANDARD
DTSTART:20191103T010000
TZOFFSETFROM:-0400
TZOFFSETTO:-0500
RRULE:FREQ=YEARLY;BYDAY=1SU;BYMONTH=11
TZNAME:EST
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
DTSTAMP:20200401T113714Z
UID:fc9553c2-e262-4258-a51d-7ffda7474fb7
DTSTART;TZID=America/New_York:20190626T160000
DTEND;TZID=America/New_York:20190626T170000
CREATED:20190620T114830
DESCRIPTION:Probabilistic inference consists of estimating a probability\nd
istribution based on a limited number of randomly sampled\nobservations. W
hen these observations are images\, Euclidean inference\n(assuming no prio
r covariance among voxels) often fails to estimate a\nrepresentative distr
ibution of the data. This problem can be overcome\nby accounting for two c
haracteristics of images: first\, their\nintrinsic smoothness\, which is c
aptured by a local covariance among\nvoxels\; and second\, their topology\
, which captures the fact that the\nobjects represented in the images are
invariant under some families of\ntransformations (e.g.\, multiplicative o
r additive changes of\nappearance\, affine or non-linear spatial deformati
ons).\n\nIn this talk I will show that a set of images can be described by
a\nmean and a distribution of transformations (of a given type)\, such\nt
hat a single transformation from the distribution would map the mean\nimag
e to a sample from the set of images\, and that the particular\ntransforma
tion type depends on the nature of the variability to be\nmodeled. I will
show two practical applications capitalizing on this\nframework: the estim
ation of sensitivity fields in multi-coil MR\nacquisitions\, and the estim
ation of brain templates in computational\nanatomy. I will then show that
by extending the model of prior\ncovariance from capturing local smoothnes
s only\, to having a\nnon-stationary form\, more structured deviations fro
m the mean image\ncan be captured. This concept will be applied to the est
imation of\nshape and appearance variability in the human brain.\n
LAST-MODIFIED:20190620T114830
LOCATION:32-D451
SUMMARY:Generative modeling of medical images
URL:https://calendar.csail.mit.edu/events/220858
END:VEVENT
END:VCALENDAR