Graphics Reference
In-Depth Information
Advancesinmedicalimagingsystemshavemadesignificantcontributionstomedical
diagnoses and treatments by providing anatomic and functional information about
human bodies that is di
cult to obtain without these techniques. hese modalities
also generate large quantities of noisy data that need modern techniques of com-
putational statistics for image reconstruction, visualization and analysis. his arti-
cle will report recent research in this area and suggest challenges that will need to
be addressed by future studies. Specifically, I will discuss computational statistics
for positron emission tomography, ultrasound images and magnetic resonance im-
ages from the perspectives of image reconstruction, image segmentation and vision
model-based image analysis.
Introduction
2.1
It has been a boon to researchers from many scientific fields to be able to use data vi-
sualization to “see the unseen.” It is rather amazing that researchers can see through
the human body and accurately visualize brain function using cutting edge medical
imaging techniques - something that would have been very di
cult to imagine at
the beginning of the twentieth century (Kevles,
). Indeed, modern medical im-
age modalities have made significant contributions to the understanding, diagnosis
and treatment of biological activities and diseases inside the human body;these con-
tributions are introduced and reviewed in most topics on medical imaging (Suetens,
;Prince and Links,
). As techniques for capturing medical images continue
to advance, it has become more of a challenge to visualize and analyze them, be-
cause they inherently contain a massive amount of noisy data. Modern techniques in
computational statistics arecrucial toextracting useful information fromthevarious
classes of modern medical images. Here,I discuss computational statistical methods
forstudyingmedicalimages,includingimagescapturedbypositronemissiontomog-
raphy (PET), ultrasound images, and magnetic resonance images (MRI), from the
perspectives of image reconstruction, image segmentation, and vision model-based
image analysis.
Computerized tomography (CT) is an important technique for obtaining accu-
rate information about the interior of a human body based on observations detected
outside the body. he precise reconstruction of images of the interior of the human
body from this data is a challenge suited to computational statistics. As the detected
observations are indirectly related to the target image, the tomographic reconstruc-
tion problem is an inverse problem, which is oten ill-posed or ill-conditioned. Due
to the nature of ill-posedness, the reconstruction of, for example, positron emission
tomography (PET) images by maximum likelihood estimation with the EM algo-
rithm (MLE-EM) (Shepp and Vardi,
; Vardi et al.,
) weighted least square
estimation (WLSE), and other methods without regularization, will produce images
withedgeandnoiseartifacts (Fessler,
;Ouyang etal.,
).hus,computational
statistical techniques must be used to integrate and fuse the correlated but incom-
plete structure information with other medical modalities, like X-ray CT, magnetic
