[log in to unmask]" type="cite">Dissertation Defense
Announcement:
To: The George Mason University Community
Candidate: Mekhala Acharya
Program: PhD Bioinformatics & Computational Biology
Date: Tuesday October 23, 2012
Time: 12:00 - 2:00 P.M.
Place: George Mason University
Occoquan Bldg. Room #110-L
Prince William campus
Dissertation Director: Dr. Jason Kinser
Committee Members: Dr. Donald Seto, Dr. Jeffrey Solka, Dr. Geraldine Grant
Title: "Image Analysis of Radiological Images from Patients with Advanced Lung Disease"
The dissertation is on reserve in the Johnson Center Library, Fairfax campus.
The doctoral project will not be read at the meeting, but should be read in advance.
All members of the George Mason University community are invited to attend.
ABSTRACT:
There are diagnostic challenges in the evaluation of a patient with a
known or suspected diagnosis of diffuse infiltrative lung disease or
interstitial lung disease [ILD] because of the extensive possibilities
of diverse potential diagnoses with similar symptoms. High-Resolution
Computed Tomography (HRCT) has changed the diagnostic evaluation of
patients with ILD and is particularly useful in the diagnosis of
idiopathic pulmonary fibrosis (IPF). The characteristic HRCT findings
of IPF are reticular abnormality and honeycombing with basal and
peripheral predominance and the radiographic pattern differs with the
stage of the disease.
The quantification of disease by CT is important to indicate prognosis
and to evaluate progression of the disease or response to treatment. It
is difficult to convey the complex textural information offered by a CT
scan hindered by the lack of user friendly technology for image
analysis. Automated tools are presented which extract information from
the CT images and isolate visual evidence of the disease from healthy
lung tissue. Each CT image is converted to a set of pulse images,
which through collective synchronization of pixels extract pertinent
information of the diseased regions. These pulse streams are used for
training and recall through an associative memory so that entire images
can be segmented.
In spite of the obvious difference in contrast, volume and texture, the
healthy and diseased regions are distinguished and classified using
pulse images. The technique used is successful in classifying the
healthy and diseased portions of the lung. The goal is to train
adequate and varied stages of IPF images and to be able to extract
sufficiently enough information from test images. The algorithm was
tested on HRCT scans procured through INOVA Fairfax Hospital,
Department of Radiology. Two expert radiological reviewers compared the
initial results of the segmentation algorithm with the manual
segmentation of the original scans. Comparison revealed agreement
regarding the presence or absence of honeycombing. Algorithms and
results for the analysis of patients with IPF and healthy patients are
presented.
The absence of gold standards in image processing makes quantification
challenging for early stage images of IPF and blinded images. Thus
medical image processing validation often cannot rely on availability
of true gold standards. Hence lung volumes derived from Pulmonary
Function Tests (PFT) results served as established clinical parameters
and were used as "gold standards" . The results of the segmentation
were compared with the measurements of the pulmonary function tests.
The relationship between image segmentation results and the PFT results
were calculated using linear aggression analysis and Pearson's product
moment correlation. Volumetric measurements of of honeycomb, vascular
and normal regions are found to correlate with results of PFTs in
patients with IPF. The greatest correlation was between honeycomb
regions and forced vital capacity (FVC). The healthy and honeycomb
regions correlated negatively with PFT measure diffusing capacity
(DLco). Results demonstrate that the segmentation of IPF images using
PCNN techniques are useful in extracting quantitative information.
###