Teaching Computers to Diagnose Cancers Humans Can’t See
Alam and his collaborators, including Maryellen Giger, professor of radiology at the University of Chicago, published the results of their work in a paper titled “Exploring Deep Parametric Embeddings for Breast Computer Aided Diagnosis.” They submitted the paper and were selected to present their findings at the International Society for Optics and Photonics Medical Imaging conference this past February in Orlando, FL.
Computer-aided diagnosis of cancer is a vital and relatively new field. “It’s so important because of how poorly we understand cancer,” Alam said. Cancer can be difficult even for trained radiologists to identify. While human vision has certain strengths, like the ability to view images holistically and make inferences, computer “vision” has a different set of strengths. Chief among them is the ability to algorithmically identify patterns from images that are impossible for humans to identify, but which may present important data. Using computer-aided image analysis will yield practical benefits in the years to come by supplementing radiologists’ attempts to identify and diagnose cancers.
“In many ways medical imaging technology has grown at a rate that surpasses our ability to meaningfully use it. Computer-aided diagnosis offers a way to really utilize the data that these images have to offer,” Alam explained.
Using a variety of sophisticated techniques, Alam and his collaborators essentially trained computers to analyze images from mammograms and other forms of medical imaging. Alam, whose concentration is computer science, worked to make the computers “smarter, capable of learning and honing their skills over time to make better diagnoses.”
Prepared for the future
It is typical for Simon’s Rock students to intern at top national laboratories. The College devotes significant resources to encourage students to apply their education in settings that will prepare them to be leaders in their fields.
Alam credits the liberal arts education he’s received at Simon’s Rock with helping to make the fellowship such a fruitful experience. “When you’re trying to figure out research possibilities and what might be fruitful, it’s just as important to figure out why those possibilities might not work out. Simon’s Rock has given me the set of skills I need to know how to properly critique an idea. While science often fragments research by very specific subfield specializations, I was able to look at the work we were doing in the context of the history of the field and the economics of health care delivery to understand more holistically what ideas would really be worth pursuing.”
Through his summer fellowship, Alam was able to examine, in practice, the capabilities and limitations of a computer’s ability to learn. Aside from providing exposure to the world of applied science, the experience has influenced Alam’s focus for his Senior Thesis, which will continue his investigation into machine learning and functional programming.