Bi-Weekly Report #4

Facilitator(s): Jason Kennedy

Recorder(s): Suryadyuti Baral

Deliverer(s): Kenza Slaoui, Simeon Kolev

Planner(s): Tanay Bali

See last page for description of roles. Obviously one person can take more than one role or there can be more than one person per role or make your own roles!

0.    Describe briefly what the main goal of your team is (so the peer reviewer has some context). E.g. we are working on image classification for blah de blah. Our goal is blah de blah etc. In the initial part of the semester before your proposal it is ok to put down “we are still coming up with ideas on team project”.

Our team is working on classifying MRI images of 7023 brain scans from a combination of three different datasets. Our goal is to make a model capable of not only identifying the presence of a tumor, but identifying its type as well. We will do this by using several different types of classification algorithms and choosing the one that works the best for our final model.

I.               What was done during the report period regarding the project: If you want to include code include this in the Appendix. Describe what the group did (including contributions of individual team members) with regards to the group project during this report period. Give enough details so I understand what you folks have been doing over the week. Include dates of your meeting(s) and who met on these days.

Our group met on 4/2 and 4/11 to discuss what progress everyone has made on their roles. Here is what everybody has done:

Suryadyuti tried to get KNN to work, but it took to long, so he moved on to making a random forest. He sampled 3,000 images for training and found the ROC to get the best threshold for classification. For test data, the accuracy is 0.9641, the sensitivity is 0.9625, and the specificity is 0.9679.

Kenza tried to use SVM, but it wouldn’t work because the images are too large at 512 by 512 pixels. So she tried to use PCA for dimension reduction, but is currently stuck and will go to office hours to work through some problems she’s having with PCA.

Simeon processed the images through a series of erosions, dilations, image contouring, and cropping. Then he stabilized the distribution of each kind of tumor across the dataset to represent real-world probabilities of each kind of tumor. Next, he classified each image based on the angle the photo was taken and if it had eyes or not. Lastly, he reduced the dataset by hand picking the highest quality images from the dataset and ran them through PCA. He is currently running them through SVM and KNN as well.

Tanay is working on decision trees and has preprocessed the data into matrices. Similar to Kenza, he is struggling with the tree algorithm using the rpart package, and intends to get help soon.

Jason coordinated meeting times, wrote the bi-weekly report, and will write the final report/presentation. Simeon will help with the writing of the final report and presentation as well.

II.             What were obstacles faced if any in working on the project? This could be technical (like not being able to implement or understand particular techniques) or time issues (midterms for other courses etc).

The most common problems this report period were the size of the dataset and confusion about certain steps of a technique. Suryadyuti had to scrap KNN because of the size of the dataset, Simeon had to reduce the dataset by hand, and Kenza is struggling with dimension reduction. We also struggled to meet a second time this report period because people were out of town over the wellness weekend, and we typically meet on Sundays.

III.            What is the plan for the next reporting period including what each team member is planning to work on. Describe goals and potential timelines (“ I plan to finish understanding x to see if it can be implemented for our project by Wednesday etc”. )

There are no specific goals for every person, but we agreed that everybody will have their parts done before April 20th. Once everybody is finished, Jason and Simeon will work together to write the presentation and final paper.