FEATURE DETECTION AND CLASSIFICATION OF MAMMOGRAMS
OUR GOAL
We've developed a MATLAB algorithm to process mammogram images and distinguish malignant tumor images from healthy images.
OUR TEAM
Passionate Engineers
ALICE TRACEY
Engineer
"One thing I learned from this project is how we can't just evaluate different features separately, since different combinations of features impact the algorithm in unique ways. A feature might capture something important about an image, but when combined with other features it could confuse the algorithm by cluttering feature space or making the differences between classes less apparent."
CYRUS NAJARIAN
Engineer
"Doing research to find new filters and features was the most valuable part of this project to me. It felt like I was going above and beyond the basic fundamentals we learned about in class, and gained experience using tools that are professionally used to solve real problems. Going one step further and modifying them to suit our needs both increased my understanding of DSP, and gave me the experience to do similar things moving forward to the future."
ALLISON TICHENOR
Engineer
"From this project I not only learned how all of the different classifiers work but got to see them used in applications which made me learn much more about them than I would've by only learning how each classification model works mathematically. Also I learned that just because you do some type of preprocessing on an image that may "look good", it may not help increase accuracy of the classifiers, i.e. Image Binarization."
DEVIN BEACH
Engineer
I learned that even if certain features extracted from your data works well with a certain classifier, there is a chance that it may not work with a different classier. I expected all of the classifiers to return similar results, but this was not the case. I found it interesting researching and learning why our tree classifiers may have worked well with the features we extracted, and why the other classifiers may not have worked as well.
SEAN HIGGINS
Engineer
"I found it really interesting how vast the resources for machine learning and classification are. Despite the fact that, in our case, many of them were fairly easy to implement, understanding how they worked and what classifiers would or wouldn't classify accurately turned out to be rather challenging, though intriguing."
"Too. many. transforms."