A $1 million data science contest organized by Booz Allen Hamilton asked entrants to design deep-learning algorithms to identify lung cancer using just 2,000 images—a small data set in the machine learning world. While the winning entry won’t necessarily be used in clinical settings, the contest highlights the potential for crowd-sourcing inspiration.
The winning team taught a neural network to first identify nodules in low-dose CT images from the National Cancer Institute before diagnosing cancer.
“We think that explicitly dividing this problem into two stages is critical, which seems also to be what human experts would do,” said Zhe Li, a member of the winning team and a student at China’s Tsinghua University, in an interview with MIT Technology Review.
The winning algorithms are available for free online, allowing medical researchers and data scientists to draw from minds outside the industry—what better way to think outside of the (black) box?
Read the full story here: