Mine-Hunting Software Identifies Rare Cancer-Related Cells 

The Office of Naval Research (ONR)-funded software developed for locating and detecting undersea mines can assist doctors in identifying and classifying cancer-related cells, according to recent demonstrations by medical researchers. Interestingly, the problem that physicians are often faced with when examining images of human cells is very similar to the Navy’s challenge of detecting undersea mines.

During tissue sample analysis, doctors must sift through hundreds of microscopic images containing millions of cells. To isolate specific cells that hold interest, they use an automated image analysis software toolkit called FARSIGHT, which means Fluorescence Association Rules for Quantitative Insight. Funded by the National Institutes of Health (NIH) and the Defense Advanced Research Projects Agency (DARPA), FARSIGHT technology seeks out cells according to a pattern of examples selected by a physician. However, the results can often be false because the computer mistakenly tags wrong examples due to the small sampling.

Researchers claim that with the addition of ONR’s active learning software, cells are identified more accurately and FARSIGHT’s performance becomes more consistent. This enhanced toolkit requires doctors to select fewer cell samples because the algorithm automatically selects the best example sets to teach the software.

A medical team at the University of Pennsylvania is now applying the FARSIGHT-embedded ONR algorithms to examine tumors from kidney cancer patients. By carefully monitoring endothelial cells that constitute the blood vessels that supply the tumors with oxygen and nutrients, researchers could one day improve drug treatments for varying instances of kidney cancer.

Researchers have previously not studied endothelial cells of human cancer because it is very difficult and time-consuming to accomplish. It can take days, even weeks, for a pathologist to selectively isolate all the endothelial cells present in 100 images. The enhanced FARSIGHT toolkit can bring the same results in only a few hours but with human accuracy.

The Office of Naval Research’s active learning software was originally engineered to modify robotic mine-hunting programs to behave more like a human when there is uncertainty about classifying an unknown object. Using information theory, the software asks humans to provide labels to classify those items. This feature is critical in naval warfare, where traditionally, divers were the only means of identifying unknown objects beneath the ocean.

Elijah Lamond