Detecting lung nodules on CT scans is central to diagnosing cancer and the earlier that can be accomplished the better the outcome. Currently, only about two thirds of detectable nodules are actually spotted by trained professionals. Researchers at University of Central Florida have been working on an artificial intelligence program that can review CT scans and spot nodules with near perfect accuracy.
Their software, called S4ND, relies on deep learning techniques, coupled with more than 1,000 previously captured lung CT scans, to point out suspect lesions. The Central Florida team built a 3D convolutional neural network to power S4ND so that it doesn’t require any input from the user to produce the results. “We used the brain as a model to create our system,” said Rodney LaLonde, one of the researchers involved in the project. “You know how connections between neurons in the brain strengthen during development and learn? We used that blueprint, if you will, to help our system understand how to look for patterns in the CT scans and teach itself how to find these tiny tumors.”
The team compared their system against two existing “state-of-the-art” computer vision and lung nodule detection systems, demonstrating that S4ND has a much better accuracy and efficiency at spotting problem spots.
Next up in moving the technology toward clinical use is getting it to be tried on CT scans of patients coming in for a diagnosis. If everything works out well, the Central Florida researchers believe their software may end up being widely used in a matter of a couple of years.
Paper in arXiv: S4ND: Single-Shot Single-Scale Lung Nodule Detection…