Medical Image Analysis Using Artificial Neural Nets in Support of Detection, Diagnosis, and Prognosis

foulchilianAI and Robotics

Oct 20, 2013 (3 years and 11 months ago)

74 views


Syllabus MATH 4/5779: Math Clinic
-

Last update
10/20/2013

1

SYLLABUS

MATH 4/5779


Math Clinic


Medical Image Analysi
s

Using Artificial Neural Nets

in Support of Detection,
Diagnosis
, and

Prognosis


Sponsor:

University of Colorado
Department of Radiation Oncology



Spring Semester
-

2010

Instructors: Weldon A.
Lodwick and Francis Newman

Resea
rch Assistant:
Shoshona Rosskamm

Office:

Weldon Lodwick
,

U
C
-
Denver Building
,

Room 643



Francis Newman
, UC
-
Denver Building
,
Room 609


Shoshona Rosskamm,

UC
-
Denver Building,
Room 609


Telephone:

303.556
.8462 (office), 303.556.
8442 (sec
retary), 303.556.
8550 (fax)


AMC, Francis

720.848.0134

E
-
Mail:

UCD
-
DDC,

Weldon Lodwick
weldon.lodwick@ucdenver.edu


UCD
-
AMC, Francis Newman
francis.newman@ucdenver.edu



UCD
-
DDC, Shoshana Rosskamm

shoshana.rosskamm
@email.ucdenver.edu


Web Site:

Instructors’ website
http://www
-
math.ucd
enver.edu/~wlodwick


Clinic website
http://www
-
math.cu
denver.edu/~clinic
/5779/

Office Hours:

Mondays/Wednesdays 4:25pm


5:25p
m



Tuesdays
11:00am


12:3
0pm

Other times by arrangement


Text:

Readings in neural
networks


general and specific neural networks (ACM,
Auto CM, and J
-
Net) to be handed out

Neural Network Design

by Martin T. Hagan, Howard B. Demuth, and Mark
Beale, PWS Publishing Company, 1996

(optional



selected chapters can be

made
available, the fir
st four chapters can be

downloaded from
http://hagan.ecen.ceat.okstate.edu/nnd.html

... thanks Ed
Gard
)
.


Students with Disabilities:

If you have a disability that requires accommodation
in this course, please see me as soon as possible.
We are
happy to make
appropriate accommodations, provided timely notice is received.


Cell Phones:

You are to turn off your cell phones prior to enter
ing class.


Objective

of this Mathematics Clinic:

The clinic is primarily a pedagogical tool
where one learns applied mathematics by solving problems faced by the
Department of Radiation Oncology
. Working in research teams to develop results
associated wi
th a project (solving a set of problems and presenting the results) is
an integral part of every clinic. Thus, we will try to solve problems that are of
current concern. In particular, for this semester,
our

objectives will be:



To develop a software syst
em utilizing artificial neural networks
,
and/or others that we may uncover over the course of the semester

for


Syllabus MATH 4/5779: Math Clinic
-

Last update
10/20/2013

2

computer assisted

detection of structures of interest,
diagnosis
, and

prognosis
.



T
herefore, upon completion the student should have a go
od
foundation in the
mathematics of artificial neural netwo
rks,
and how to apply them to medical
images.

The
“deliverable” to our sponsor is
MATLAB co
de that

(or makes
good progress such that it)

performs
:


1.

Computer Assisted Prognosis: Non
-
small cell lung can
cer (NSCL) and
limited small cell lung cancer (LSCLC) patients respond, do not respond,
or remain unchanged upon receiving chemotherapy given to reduce the
size of the tumor. We will apply artificial neural networks (ANN’s),
specifically J
-
Net (and other
architectures) to computed tomography (CT)
images from the NSCL and LSCLC patients before and after
chemotherapy to determine if responders, non
-
responders, and unchanged
patients can be predicted from the ANN image analysis. J
-
Net has shown
that it can f
ind “harmonics” in images that are not responses to
chemotherapy and the question is, “Can ANN’s be trained to predict
responses from patients whose images are not in the training set?”

2.

Computer Assisted Detection and Diagnosis: It has been shown that one
can train ANN’s to differentiate between normal versus abnormal lungs
and between different lung pathologies in CT chest images. The goal is to
train ANN’s on CT’s to determine the effectiveness of differential
diagnosis of these conditions. Can the ANN’
s achieve better than 90%
success rate in the differential diagnosis? What does the confusion matrix
look like?

3.

Computer Assisted
Brain Morphology and IQ
Analysis
in Pediatric
Patients after Radiation Treatment of Brain Tumors: After cancer
treatment to t
he central nervous system, pediatric patients may undergo
noticeable changes in bran morphology. Multiple magnetic resonance
images (MRI’s) are taken of pediatric cancer patients after their treatment
to monitor changes. An IQ test may be administered po
st
-
treatment as
well. A semi
-
automatic segmentation routine has been employed on the
post
-
treatment MRI’s to measure brain morphology. That is, volumes of
certain brain regions such as the right cerebral cortex are measured (albeit
imperfectly because of

motion artifacts common in pediatric patients).
Can one determine a correspondence between the changes in brain
morphology and IQ?

4.

Computer Assisted Brain Lesion Detectors: Find all lesions and/or MS
scabs in an MRI or CT.



To accomplish these
objectives, the clinic w
ill split up into several teams

to work
on a semester project.

Each individual
will be working on subtasks le
ading to the
completion of the team

p
roject and each team will have a team leader to
coordinate the tasks.
Software develop
ment involves research to create
, to test,

to

Syllabus MATH 4/5779: Math Clinic
-

Last update
10/20/2013

3

analyze, and to document (
the software
)
.



Prerequisites

Linear algebra, advanced calculus or junior level engineering mathematics or
mathematical physics. The student should have some familiarity with M
ATLAB

o
r be ready to learn quickly.


Assignments

There are three

problem sets which will be questions and calculations associated
with the readings.

These are to be done individually.


Implementation

Each our four

teams will implement one of the following systems either by
writing their own MATLAB application or by using Semeion software: J
-
Net,
Auto CM, ACM
, and probabilistic (Bayesian) ANNs
.


PROPOSED COURSE OUTLINE

The proposed outline is the initial guess of th
e topics that will be fruitful to
investigate. Research is a process of discovery when one does not know, so the
rule is that we will modify our topics during the semester. Thus the proposed
outline will undoubtedly change as we learn more during the sem
ester.




The tentative topics we will cover are:

I.

Introduction

A.

Conduct of the course, expectations, assignments, projects

B.

Problem statement for this semester’s Math Clinic

II.

Artificial Neural Networks

(from
Neural Network Design
)

A.

Example problems and
architectures (Chapters 2/3)

B.

Perceptron (Chapter 4)

C.

Hebbian Learning (Chapter 7)

D.

Backpropagation (Chapter 11)

E.

Associative Learning (Chapter 13)

III.

Specialized Neural Networks

(from articles handed out)

A.

Bayesian Neural Networks, Probabilistic Neural Networks

B.

J
-
Net

C.

ACM

D.

Auto CM


E.

Adaptive Resonance Theory (Chapter 16)

IV.

Other Approaches (it will depend on what is uncovered during the
course of the semester)


Important Dates/Tentative Schedule

Week of:



Topics Covered

January 20
-

February 20

Introduction to the tec
hnical aspects of the clinic
.


Syllabus MATH 4/5779: Math Clinic
-

Last update
10/20/2013

4

February 12



Problem Set 1 due

(5pm)


February 19



Draft Project Proposal


February 26



Written
Project proposal due

methods, materials,

schedule, di
vision of labor (5pm)


March 12

Problem Set 2 due (5pm)








March 17



Progress report

and
annotated bibliography








Written paper and a
15
m
inu
te oral presentation
by





each
team


April 9
, 12, 16, 19, 23, 26
Implementations






April 16



Problem Set 3

due

(5pm)


May 7

Fina
l written reports

(with software)
due

(5
pm
)

M
ay 10 or 12



Presentations
,
30 minutes/group


Projects

You are
, in conjunction with us,

to choose projects from the following list
.
P
rojects could evolve
, be added,
subtracted or

modified during the course
-

not
arbitrarily but as a result of
circumstances.


Project 1:


Develop a computer assisted prognosis system in MATLAB that
applies artificial neural networks (ANN’s), specifically J
-
Net (and other
architectures) to computed tomography (CT) images from the NSCL and LSCLC
patients before and
after chemotherapy to determine if responders, non
-
responders, and unchanged patients can be predicted from the ANN image
analysis.


Project 2:

Develop a computer assisted detection and diagnosis
system in
MATLAB that applies artificial neural networks (AN
N’s)

that one can train
ANN’s to differentiate between normal versus abnormal lungs and between
different lung pathologies in CT chest images. The goal is to train ANN’s on
CT’s to determine the effectiveness of differential diagnosis of these conditions.

Part of the project will be devoted to looking at the applicability of Bayesian
decision networks (probabilistic neural networks).


Project 3:

Develop a MATLAB system to assist in brain morphology and IQ
analysis for pediatric patients who have undergone

radiation treatment of brain
tumors to determine a correspondence between the changes in brain morphology
and IQ.

The techniques that are best suited for this project are one of the things
that the project will investigate. Auto CM would be one that sho
uld be relevant.


Syllabus MATH 4/5779: Math Clinic
-

Last update
10/20/2013

5


Project 4
:

Computer assisted detection, CADe, of brain lesions
, t
his project will
employ ANNs (novel or traditional) to find lesions in MRIs but perhaps in CTs as
well. A substantial database of MRIs with multiple sclerosis (MS) plaques exists
and it is of diagnostic significance to be able to locate all lesions. Some m
ay be
quite small and difficult to find. Likewise we will employ brain MRIs and CTs to
locate small tumors (usually metastatic lesions) that are also difficult to find.
Small tumors may become large tumors and it is therapeutically important to find
the le
sions when they are still small.


OUR APPROACH

We

believe that teaching is a process that inv
olves an active partnership. Our

role
is that of a guide to your learning. Therefore,
we are

responsible to open the way,
to encourage, and to nudge you towar
d your own learning. In t
he context of the
math clinic, we

will try to model the process of applying mathematics to the
medical image analysis. We

will help guide you toward this learning by providing
mathematics
for you to experience. It is our

aim to c
ommunicate mathematics in
a way that is supportive and nurturing of your efforts. Your role is to find a way to
experience and articulate the mathematics that is pres
ented and that you
encounter. We

believe that it
is your responsibility to let us

know wh
en you find
yourself not understanding mathematical concepts that are presented in class.
Once you make this known, it is our responsibility to work

on trying to attain
clarity. We

will try to

be as proactive as possible. We

believe that results on
proj
ects give us the opportunity to clearly see where the areas of mathematical
understanding are and what areas need more attention.


OUTCOMES

By the end of the semester you should be able to read, understand
,

and apply
appropriate methods associated with asp
ects of medical image
analysis

we’ve
stu
died this semester to
model
correctly
and
to
solve associated probl
ems.
Secondly, given a
medical image
analysis

problem
, you should be able to: (i)
translate the description of the problem into an algorithm, (ii) choose and apply
the appropriate software method(s), (iii) obtain the
an appropriate
solution(s), and
(iv) (correctly) interpret and display results. Lastly, by
the end of the semester
you should be able to judge, for yourself, the veracity of statements made in the
areas of our study.



EVALUATION

Each person on a team will execute a project (identify a set of problems, find
solution methods, present the results
and write
-
up the results).

Individualization
of grades are based on the adequacy of fulfillment of the division of labor
associated with your team’s project proposal.

In particular, the following are
components that will be evaluated.

1.

Participation


att
endance and contributi
ng to class interactions and
discussio
ns (10
%)


Syllabus MATH 4/5779: Math Clinic
-

Last update
10/20/2013

6

2.

Annotate
d bibliography


Included in

the Progress

R
eport

(5
%)

3.

Progress

Report

(10%
)

a.

Write
-
up (5%
)

b.

Presentation (
5
%)

4.

Final Presentation (10%)

5.

Final written report (20%)

6.

Final software (1
5

%)


7.

Assignments
(2
0
% total)

8.

Implementation (10%)


** Graduate students will have extended content and be held to higher standards.

The grade assignments are on the 10 percent scale (A = 90%
-
100%, B = 80%
-
89%, C = 70%
-
79%, D = 60
-
69%).


General advice:

Kee
p all materials that
we

turn back in case you think
we

have
not credited you with the points you earned.
we

can only correct your score if
you have what
we

have turned back to you. It is a good idea to xerox anything
that you turn in just in case
we

lose
what you turn in. Please check to make sure
that the points you earned are the points
we

have recorded. Note: The statistics
that
we

have read about correctness of professors in recording grades state that
there is a 6% error rate in our recording of you
r grades. Please make sure that
we

have correctly recorded your points.


POLICIES

Legitimate Excuses:

Legitimate excuses are for reasons that are beyond your
control. You may be required to produce an official, signed excuse. If you are
needed in a wedding,
for example, you must talk to us

prior

to the (blessed)
event. If you are legally arrested, then

this is not a legitimate excuse. For matters
that are within your control, the general rule is that it is n
ot excused. However,
talk to us

prior

to the event

or arrest

or departure
.


ANNOTATED BIBLIOGRAPHY

An annotated bibliography is

a
full bibliograph
ical

list of

citation
s

of the materials
you have read and consulted with notes. The notes you write are more or less
notes to yourself or to someone next semester that may take up your project. In
general you should have at least five citations and the a
nnotation should state
whether or not you found it useful to your project and why.


INSTRUCTIONS FOR PROJECTS


A project consists of:

1.

Proposal



Depending on the size of the class, projects may be coalesced
or deleted. Each
proposed

project

will be divided into tasks and assigned
to each group so that the assignment is equitable. These tasks and
assignments need to b
e written up and submitted to us
.
Once the tasks

Syllabus MATH 4/5779: Math Clinic
-

Last update
10/20/2013

7

have been identified, assigned
,

and approved, a division of labor is written
by each of the groups.

2.

Division of labor



Each group must take their tasks and subdivide them
into subtasks that are assigned to individuals in the group with an
associated due
-
date. A division of labor is a formal contract between the
members of the gro
up. Once the tasks have been approved and a written
division of labor submitted
, the group needs to schedule a meeting with us

so that we can go over the division of labor, its associated responsibilities
and expectations
.

3.

Methods and Materials



is a des
cription of what materials (software)
and the approach(es) (methods, algorithms)

you will use.

4.

Annotated Bibliography


5.

Software

a.

Code
-

the actual computer implementation of the project.
Attention must be paid to efficiency, readability
,

and portability.

b.

Input


the way information is passed to the software
must be
transparent and easily usable

by
the client.

c.

Execution
-

the algorithm as run must correctly perform what it
was designed to do.

d.

Output
-

relev
ant, clear display of solution(s) such as tables,
g
raphs, images, reports/lists
.

e.

Ease


ease of use.

f.

Documentation


an in
-
line and hardcopy of the documentation
on how to use the software needs to be written. Moreover, help
files must be part of the software.

6.

Testing and analysis

a.

Testing
-

this part in
the context of our clinic

consists of running
the software developed on the test problems
.

b.

Analysis

-

the purpose of an analysis is to get you to
critically

evaluate the results obtained from the software as it w
as run on the
test problems.
Part of an ana
lysis is a critique of the software.


* A caveat: Negative results are not prohibited. Negative results can be very
valuable. However the negative results must be robust in that it would be
novel and instructive to the expert community to avoid a particula
r pathway.


7.

Report that will be a chapter in the
Clinic Report


Each team will need
to be responsible for parts of the final report. This will be done in MS
-
Word

(needs to be translatable into PDF
-
Adobe)

or Latex

as lo
ng as we
can merge the files in Lat
ex or Adobe Acrobat.
The final report will
(subject to modifications we uncover) consist of:

a.

Introduction



clinic
instructors

b.

Project 1

i.

Theoretical foundations


theory, application, algorithms

ii.

Software



description


Syllabus MATH 4/5779: Math Clinic
-

Last update
10/20/2013

8

iii.

Results



summary, tables, graphs, images, lists
,

distinguishing features, performance, and

limitations

iv.

Opportunities for further research

v.

Conclusions

vi.

Bibliography

vii.

Appendices



Source code



Test problems, data, example runs



Documentation



Proofs



Other

c.

Project 2
(same

as P
roject 1)



d.

Project N

(same as P
roject 1)

NOTE: The software from all teams needs to be burned onto one disk where each
team will have a directory into which the software is stored. At the head of each
team’s directory entry, there must be a “read
-
me
” file and the software system
must have a “help” command
that assists the user.


Note:

Each problem from your problems sets will be worth the same

1/22
nd

of
20% for undergrads and 1/2
9
th

of 20% for graduates
.

PROBLEM SET 1
:

Due February 12
th

at 5pm

E2.3, E2.4, E3.1,
E4.2,E4.3,E4.4,E4.7,E4.8

Graduate Students add E4.5, E4.9

PROBLEM SET 2:

Due March 12
th

at 5pm

E7.1,

E7.4,


E11.6,

E11.7,

E11.8,

E11.11,


E13.1,E13.2,
E13.6,E13.9

Graduate Students add E7.5, E7.7, E11.9

PROBLEM SET 3:
Due
April 16
th

at 5pm

E16.1, E16.2, E16.7, E16.8

Graduate Students add E16.4, E16.5

IMPLEMENTATIONS OF THE SEMEION APPROACH
:


Team 1
,
Write a MATLAB implementation

or use the Semeion software for

J
-
Net
.

Due April 9
th

at 5pm, class presentation April 12
th


Team
s

2

and 4
, Write a MATLAB implementation
or use the Semeion software,
Auto

CM

(half the teams), pr
obabilistic ANNs (the other half

of the teams).


Due
April 16
th

at 5pm

, class presentation April 19
th


Team 3
, Write a MATLAB implementation or u
se the Semeion softwa
re for
ACM.
Due April
23
rd

at 5pm
, class presentation April 26
th