3D ultrasound imaging (freehand)
I can't be the only one working on 3D ultrasound ;-) There are a number of
clinical and engineering issues involved, and I'd be more than happy
to talk shop to other folks out there who are interested.
I've already done some work on image enhancement and segmentation using
active contours/simulated annealing for fairly clean images of, say,
a heart suspended in a bath. This was basically baby work, and fails
completely on thyroid images, which had to be hand segmented. Another
guy here has also done some volume estimation/surface reconstruction work
on thyroids.
More generally, I am trying to use physics-based models (i.e. mass, stiffness,
linear elasticity) for {\em shape} modelling and perhaps even dynamic modelling
of structures. I reckon that a plausible model of how a consultant segments
and visualises is
- he has a prior model, together with typical modes of variation
- he registers this model to the structure being scanned at the mo
- he warps the model using whole-body deformations (pinching/bending/tapering)
to obtain a robust large scale fit
- finally small scale tweaking to fit small features which may not be captured in his library of typical deformations.
our pet (!) radiologist certainly seems to work this way, though i havent
gotten round to probing him about it yet. eg. for the thyroid images, i
personally can't see a sodding boundary there, but he just merrily clicks
away at this "thing" that jumps out at him.
what do the radiologists out there think? how do you work?
any psychovisual literature out there which i should pursue to back up this
approach? Talk to me ;-))
Incidentally, anyone going to VBC'94 in Minnesota in October? I'm doing a
poster there and would be dead happy to meet up with some experts.
Mike Syn
--
-----
Speech, Vision and Robotics Group, Cambridge University Engineering Dept.
