Tweet
Below is an edited selection from this month's ai-medicine digest selection
of dissertation abstracts.
HJW
Date: Thu, 4 Jun 92 17:02:45 EDT
From: humphrey@nlm.nih.gov (Susanne M Humphrey)
Sender: ai-medicine digest, Vol. 4, nr. 2
To: ai-medicine@MED.Stanford.EDU
Subject: Dissertation Abstracts - June 1992 (Long)
The following are citations selected by title and abstract as being
of potential interest to the [Biomch-L] community, resulting from a
computer search, using the BRS Information Technologies retrieval
service, of the Dissertation Abstracts International (DAI) database
produced by University Microfilms International (UMI). Included are UMI
order number, title, author, degree, year, institution; number of pages,
one or more DAI subject descriptors chosen by the author, and abstract.
Unless otherwise specified, paper or microform copies of dissertations
may be ordered from University Microfilms International, Dissertation
Copies, Post Office Box 1764, Ann Arbor, MI 48106; telephone for U.S.
(except Michigan, Hawaii, Alaska): 1-800-521-3042, for Canada:
1-800-268-6090; fax: 313-973-1540. Price lists and other ordering and
shipping information are in the introduction to the published DAI. An
alternate source for copies is sometimes provided. Dissertation titles
and abstracts contained here are published with permission of University
Microfilms International, publishers of Dissertation Abstracts
International (copyright by University Microfilms International), and
may not be reproduced without their prior permission.
AN University Microfilms Order Number ADG92-13428.
AU CHEN, SHIUH-YUNG.
TI SENSOR FUSION AND KNOWLEDGE INTEGRATION FOR MEDICAL IMAGE RECOGNITION.
IN Northwestern University Ph.D. 1991, 274 pages.
SO DAI V52(12), SecB, pp6496.
DE Computer Science.
AB In this dissertation, we describe a medical image understanding
system whose reasoning module employs the profound features of
Dempster-Shafer theory. Given a set of three correlated images
acquired from x-ray CT, $T\sb1$-, and $T\sb2$-weighted modalities at
the same slicing level and angle of a human brain, the proposed
system is capable of mimicking the reasoning process of a human
expert in recognizing the image set based on (1) the knowledge about
sensor characteristics, (2) the knowledge about anatomical
structures, and (3) the knowledge about image processing and analysis
tools. To implement such complicated processes, the blackboard
architecture composed of three major components, knowledge sources,
blackboard data structure, and control is adopted. The proposed
system consists of three phases. In phase one, entities in the form
of regions and curves with associated features are extracted from the
images. The second and the third phases aim at recognizing the
physically meaningful entities in the image set. In phase two, the
system tries to identify the major anatomies and locate the slice in
the model that is most similar to the image set under study. In
phase three, the selected model slice is used to refine the formation
of the identified anatomical structures and extract gray and white
matters.
AN University Microfilms Order Number ADG92-13519.
AU NEIW, HAN-MIN.
TI AN AUTOMATED SYSTEM FOR THREE-DIMENSIONAL REGISTRATION OF MEDICAL
IMAGES.
IN Northwestern University Ph.D. 1991, 163 pages.
SO DAI V52(12), SecB, pp6507.
DE Computer Science. Artificial Intelligence. Health Sciences, Radiology.
AB In this dissertation, we develope an automated three-dimensional
(3-D) medical image registration system. Registration is an image
analysis technique that integrates anatomical and/or functional
information from images acquired by different modalities or at
different times for the improvement of medical diagnosis and
treatment.
Our proposed system is a surface fitting based system with accuracy
on the order of the image pixel sizes. The surface fitting technique
extracts the external surface contours of two or more sets of images,
and employs an optimization scheme to fit the contour sets together.
The proposed automated system contains three main subsystems for
automated contour extraction, surface model matching, and
reformatting correlated images, respectively. Our system has
eliminated the heavy human expert involvement previously required in
all of the image registration systems. In the contour extraction
phase, algorithms equipped with domain-specific knowledge are able to
extract the external surface contours of various brain images without
any assistance from human. In the surface model matching phase, our
system has the ability to circumvent the local minimum problem
encountered quite frequently in many optimization problems.
Moreover, we have added a final matching step to fine-tune the
results from an initial matching step. The error measurement in the
final matching phase is the most accurate measurement available and
it is therefore more effective in guiding the optimization process.
In our system, there is no need for special clinical procedure and
external devices during image acquisition, as such images obtained
from routine clinical practice can be used directly for registration.
Our system also has the ability to handle non-completely overlapping
scans, as such images with different scanning orientations (axial,
sagittal, or coronal) can be registered in a homogeneous manner
without special human assistance.
Our system has been successfully applied to register images acquired
from X-ray computed tomography (CT), magnetic resonance imaging
(MRI), and positron emission tomography (PET). This system is
expected to facilitate the process of employing accurately correlated
medical images for medical treatment and diagnosis.
AN University Microfilms Order Number ADG92-15167.
AU WANG, CHUANMING.
TI A ROBUST SYSTEM FOR AUTOMATED DECOMPOSITION OF THE ELECTROMYOGRAM
UTILIZING A NEURAL NETWORK ARCHITECTURE.
IN Wayne State University Ph.D. 1991, 128 pages.
SO DAI V52(12), SecB, pp6583.
DE Engineering, Electronics and Electrical.
AB Clinical electromyography relies on subjective, qualitative
assessment of the EMG signal. Manual quantitative methods have not
gained widespread use, in part because they are tedious. Several
computer assisted methods have been developed. Most of these methods
require considerable user interaction. Fully automated methods have
not been entirely successful, because of a lack of robustness in the
face of varying and complex signals.
Neural network signal processing methods have recently been shown to
have significant robustness in processing complex, degraded, noisy
and unstable signals. A novel approach to automated EMG signal
decomposition, based on a neural network processing architecture, has
been developed and is presented in this dissertation.
Due to the lack of a priori knowledge of EMG signals, the neural
network must be trained in an unsupervised manner. An unsupervised
neural network classifier, consisting of a multi-layer perceptron
neural network and a pseudo-unsupervised training strategy, is
proposed in this dissertation. The network learns repetitive
appearances of motor unit potential (MUP) waveforms from their
suspected occurrences in filtered EMG signal. The same training
waveforms are fed into the trained neural network and the output of
the neural network is fedback to give rise to a dynamic retrieval net
classifier. Features discovered by the neural network are collected
as feature vectors associated with each input waveform.
Classification is achieved by comparing those feature vectors.
Firing information of each MUP is further used to refine the
classification results of the neural network classifier. Then
individual MUPs are derived and their firing tables are created.
Some potential useful features are extracted from the averaged MUPs
and their firing tables for diagnostic purposes.
The automated decomposition of EMG utilizing an artificial neural
network is robust because its behavior mimics more closely that of
human beings than traditional signal processing and pattern
recognition methods. The proposed method is capable of decomposing
an EMG signal containing up to 11 MUPs.
of dissertation abstracts.
HJW
Date: Thu, 4 Jun 92 17:02:45 EDT
From: humphrey@nlm.nih.gov (Susanne M Humphrey)
Sender: ai-medicine digest, Vol. 4, nr. 2
To: ai-medicine@MED.Stanford.EDU
Subject: Dissertation Abstracts - June 1992 (Long)
The following are citations selected by title and abstract as being
of potential interest to the [Biomch-L] community, resulting from a
computer search, using the BRS Information Technologies retrieval
service, of the Dissertation Abstracts International (DAI) database
produced by University Microfilms International (UMI). Included are UMI
order number, title, author, degree, year, institution; number of pages,
one or more DAI subject descriptors chosen by the author, and abstract.
Unless otherwise specified, paper or microform copies of dissertations
may be ordered from University Microfilms International, Dissertation
Copies, Post Office Box 1764, Ann Arbor, MI 48106; telephone for U.S.
(except Michigan, Hawaii, Alaska): 1-800-521-3042, for Canada:
1-800-268-6090; fax: 313-973-1540. Price lists and other ordering and
shipping information are in the introduction to the published DAI. An
alternate source for copies is sometimes provided. Dissertation titles
and abstracts contained here are published with permission of University
Microfilms International, publishers of Dissertation Abstracts
International (copyright by University Microfilms International), and
may not be reproduced without their prior permission.
AN University Microfilms Order Number ADG92-13428.
AU CHEN, SHIUH-YUNG.
TI SENSOR FUSION AND KNOWLEDGE INTEGRATION FOR MEDICAL IMAGE RECOGNITION.
IN Northwestern University Ph.D. 1991, 274 pages.
SO DAI V52(12), SecB, pp6496.
DE Computer Science.
AB In this dissertation, we describe a medical image understanding
system whose reasoning module employs the profound features of
Dempster-Shafer theory. Given a set of three correlated images
acquired from x-ray CT, $T\sb1$-, and $T\sb2$-weighted modalities at
the same slicing level and angle of a human brain, the proposed
system is capable of mimicking the reasoning process of a human
expert in recognizing the image set based on (1) the knowledge about
sensor characteristics, (2) the knowledge about anatomical
structures, and (3) the knowledge about image processing and analysis
tools. To implement such complicated processes, the blackboard
architecture composed of three major components, knowledge sources,
blackboard data structure, and control is adopted. The proposed
system consists of three phases. In phase one, entities in the form
of regions and curves with associated features are extracted from the
images. The second and the third phases aim at recognizing the
physically meaningful entities in the image set. In phase two, the
system tries to identify the major anatomies and locate the slice in
the model that is most similar to the image set under study. In
phase three, the selected model slice is used to refine the formation
of the identified anatomical structures and extract gray and white
matters.
AN University Microfilms Order Number ADG92-13519.
AU NEIW, HAN-MIN.
TI AN AUTOMATED SYSTEM FOR THREE-DIMENSIONAL REGISTRATION OF MEDICAL
IMAGES.
IN Northwestern University Ph.D. 1991, 163 pages.
SO DAI V52(12), SecB, pp6507.
DE Computer Science. Artificial Intelligence. Health Sciences, Radiology.
AB In this dissertation, we develope an automated three-dimensional
(3-D) medical image registration system. Registration is an image
analysis technique that integrates anatomical and/or functional
information from images acquired by different modalities or at
different times for the improvement of medical diagnosis and
treatment.
Our proposed system is a surface fitting based system with accuracy
on the order of the image pixel sizes. The surface fitting technique
extracts the external surface contours of two or more sets of images,
and employs an optimization scheme to fit the contour sets together.
The proposed automated system contains three main subsystems for
automated contour extraction, surface model matching, and
reformatting correlated images, respectively. Our system has
eliminated the heavy human expert involvement previously required in
all of the image registration systems. In the contour extraction
phase, algorithms equipped with domain-specific knowledge are able to
extract the external surface contours of various brain images without
any assistance from human. In the surface model matching phase, our
system has the ability to circumvent the local minimum problem
encountered quite frequently in many optimization problems.
Moreover, we have added a final matching step to fine-tune the
results from an initial matching step. The error measurement in the
final matching phase is the most accurate measurement available and
it is therefore more effective in guiding the optimization process.
In our system, there is no need for special clinical procedure and
external devices during image acquisition, as such images obtained
from routine clinical practice can be used directly for registration.
Our system also has the ability to handle non-completely overlapping
scans, as such images with different scanning orientations (axial,
sagittal, or coronal) can be registered in a homogeneous manner
without special human assistance.
Our system has been successfully applied to register images acquired
from X-ray computed tomography (CT), magnetic resonance imaging
(MRI), and positron emission tomography (PET). This system is
expected to facilitate the process of employing accurately correlated
medical images for medical treatment and diagnosis.
AN University Microfilms Order Number ADG92-15167.
AU WANG, CHUANMING.
TI A ROBUST SYSTEM FOR AUTOMATED DECOMPOSITION OF THE ELECTROMYOGRAM
UTILIZING A NEURAL NETWORK ARCHITECTURE.
IN Wayne State University Ph.D. 1991, 128 pages.
SO DAI V52(12), SecB, pp6583.
DE Engineering, Electronics and Electrical.
AB Clinical electromyography relies on subjective, qualitative
assessment of the EMG signal. Manual quantitative methods have not
gained widespread use, in part because they are tedious. Several
computer assisted methods have been developed. Most of these methods
require considerable user interaction. Fully automated methods have
not been entirely successful, because of a lack of robustness in the
face of varying and complex signals.
Neural network signal processing methods have recently been shown to
have significant robustness in processing complex, degraded, noisy
and unstable signals. A novel approach to automated EMG signal
decomposition, based on a neural network processing architecture, has
been developed and is presented in this dissertation.
Due to the lack of a priori knowledge of EMG signals, the neural
network must be trained in an unsupervised manner. An unsupervised
neural network classifier, consisting of a multi-layer perceptron
neural network and a pseudo-unsupervised training strategy, is
proposed in this dissertation. The network learns repetitive
appearances of motor unit potential (MUP) waveforms from their
suspected occurrences in filtered EMG signal. The same training
waveforms are fed into the trained neural network and the output of
the neural network is fedback to give rise to a dynamic retrieval net
classifier. Features discovered by the neural network are collected
as feature vectors associated with each input waveform.
Classification is achieved by comparing those feature vectors.
Firing information of each MUP is further used to refine the
classification results of the neural network classifier. Then
individual MUPs are derived and their firing tables are created.
Some potential useful features are extracted from the averaged MUPs
and their firing tables for diagnostic purposes.
The automated decomposition of EMG utilizing an artificial neural
network is robust because its behavior mimics more closely that of
human beings than traditional signal processing and pattern
recognition methods. The proposed method is capable of decomposing
an EMG signal containing up to 11 MUPs.