Manzano High School (82)/Interim Report

Interim Report
http://mode.lanl.k12.nm.us/get_interim1112.php?team_id=82

Problem Definition:
Alzheimer’s Disease impacts the lives of millions of people each year. It is characterized by loss of cognitive functioning, including memory, personality, behavior, and judgment. An estimated 5.4 million people have Alzheimer’s Disease. Plaques and inflammation in the brain are more common in people with Alzheimer’s Disease. There are currently drug treatments that are helpful but this is a progressive disease that gradually erodes functioning. Researchers continue to look for additional treatments to reduce the size of Alzheimer’s Disease plaques.

The goal is of this project is to contribute to the research of medical treatments for Alzheimer’s Disease. I am writing an image analysis program that will expedite the acquisition of data from magnetic resonance images of transgenic mouse brains for Alzheimer’s Disease research. Researchers at the University of New Mexico spend months finding plaques and analyzing the data they collect of a single image. This program will differentiate the patterns of the plague from the noise and components of the brain to produce a list of plaques, their locations in the brain, their sizes, and their relative intensities.

Problem Solution:
The image analysis program will combine an array of processes in order to produce quantitative data for the University of New Mexico researchers to further analyze. I initially flattened and corrected for the variance of intensity by subtracting the base curvature of the image from the image itself. The program will apply a z-score filter to the image to find the statistical distributions of the plaque. To distinguish plaque from other features of the image, the program will find the difference between the plaque and the surrounding pixels. Then the program will create a binarizing filter to highlight the plaque in the image. Using this binarized image, the program will find the (x, y) coordinates of the plaque using an edge finding algorithm. Then, it will find the average z-score of the plaque and the area in pixels of the plaque. Parallel processes will be used to reduce the computation time of the program.

Progress to Date:
I have been working with Dr. Sillerud, a molecular biology professor at the Univeristy of New Mexico. He gave me the magnetic resonance images to analyze and his existing Alzheimer’s Disease research, which hand counted plaques in the mouse brains. Also, Dr. Sillerud provided me with an overview of the processes required to analyze the images. I produced a program in the MATLAB programming language to accomplish these goals. My program clearly finds many plaque-like patterns. However, it finds over 7,000 plaques, which is not consistent with the 600 found by hand. I am in the process of modifying the original program by reworking the algorithms and parameters for what defines a plaque in the image. Dr. Sillerud regularly discusses my progress and provides feedback to help my program more closely meet the scientific standards established in his previous research. When the program is finished, the results should be reliable with hand gathered information. I have also parallelized the program to speed up running time.

Expected Results:
When the information gathered by the image analysis program is reliable, the Alzheimer’s research team at the University of New Mexico will use this program to collect data on the impacts of a variety of medications on mouse brains. This program will expedite the process, allowing researchers more time to analyze the data and bring medication to people more quickly.

References:
Alzheimer’s Association. (2010). Alzheimer’s Disease Facts and Figures. Alzheimer’s Association, Washington, D.C.: Available at www.alz.org.

Ashe, K.H., and Zahs, K.R. (2010). Probing the biology of Alzheimer’s disease in mice. Neuron 6, 631-645.

MATLAB Central. (2011). Retrieved 2011, from MathWorks: "http://www.mathworks.com/matlabcentral Savitch, W. (2009). Java An Introduction to Problem Solving and Programming. Saddle River, New Jersey: Prentice Hall. Sillerud, L., & Chamberlin, R. (n.d.). SPION-Enhanced MRI Shows That Inhibition of NF-κB Concomitantly Lowers Alzheimer's Plaque Formation and Neuroinflamation in Transgenic Mouse Brain. Unpublished.

Mentor:
Laurel Sillerud

Introduction
Hi,

My name is Drew Einhorn. See the biography on my User Page.

Progress
I see significant progress since your proposal.

Mentors
I am glad to see that you have a mentor actively supporting your work on this ambitious project. However, your project may require an interdisciplinary approach. I am sure your current mentor has all the skills to advise you on clinical and biological issues. You may need additional mentor(s) with expertise in image processing, computer vision, pattern recognition, neural networks, genetic algorithms, etc. Email consult if you need help finding additional mentor(s).

Model

 * You have a very straight forward algorithm.
 * "However, it finds over 7,000 plaques, which is not consistent with the 600 found by hand. I am in the process of modifying the original program by reworking the algorithms and parameters for what defines a plaque in the image."
 * I'm sure you realize that 6,400 false is a serious obstacle to the usability of your results.
 * You might consider neural network based algorithms to distinguish between true positives, and false positives.
 * There is a very large number of these algorithms, and you would start by comparing the results from using a large number of different libraries, rather than coding your own implementation of each of these algorithms.
 * You may need a larger training set for your neural networks, so you might need to ask your mentor for more images with manualy coded results. You can multiply the number of images for training by using a mirror image; rotating by 90, 180, and 270 degrees; shifting left, right, up, and down; zooming in and out.  Depending on the scale of the structures you are attempting to recognize, and the resolution of the images, rotating by other amounts, and zooming, may introduce blurring or other artifacts that might be a problem.
 * Training your neural networks using true negatives, not selected by your initial algorithm, may be useful. Recognizing a disruption of a healthy pattern may be an important part of recognizing a true positive.


 * Are there false negatives? Are all 600 actual plaques identified?


 * Do not get discouraged! Although you might not succeed in creating awe inspiring clinical research tool. A tool that has few false negatives, and makes a significant reduction in the number of false could still be viewed as a successful challenge project.  And, if you reduce the number of false positives to say 1000, it would reduce the amount of human work required, and still by a useful research tool.


 * "I have also parallelized the program to speed up running time."
 * Parallelization and performance optimization is a good thing, but in this case it is premature. Reworking the program to correct false postive/negative issues is often more difficult in the parallelized code.
 * You may want go back to the earlier version. Solve the problems.  Then parallelize the working code.

Face to Face Evaluation
Your next milestone is a |face to face evaluation in February.

Rubrics
The judges will use these rubrics to evaluate your projects. Use them as checklists for what you need to communicate to the judges.


 * Expo Judges Rubric
 * Finalist Judges Rubric

Good Luck!!
You have a project with a lot of potential!