System Bits: Sept. 12

Neural network cautionary tale
As machine learning and neural networks proliferate widely today, there is a need to exercise caution in how they are employed, according to Stanford University researchers Michal Kosinki and Yilun Wang.

In a study conducted recently, they have shown that deep neural networks can be used to determine the sexual orientation of a person, and caution that this may be a threat to the privacy and safety of gay men and women as companies and governments increasingly use computer vision algorithms to detect people’s intimate traits, which is an invasion of privacy.

Machine-learning earthquake prediction
By listening to the acoustic signal emitted by a laboratory-created earthquake, a computer science approach developed by Los Alamos National Laboratory researchers using machine learning can predict the time remaining before the fault fails.

Researchers at Los Alamos National Laboratory have developed a two-dimensional tabletop simulator that models the buildup and release of stress along an artificial fault. In this image, the simulator is viewed through a polarized camera lens, photo-elastic plates reveal discrete points of stress buildup along both sides of the modeled fault as the far (upper) plate is moved laterally along the fault.
(Source: Los Alamos National Laboratory)

Paul Johnson, a Los Alamos National Laboratory fellow and lead investigator on the research said, “At any given instant, the noise coming from the lab fault zone provides quantitative information on when the fault will slip. The novelty of our work is the use of machine learning to discover and understand new physics of failure, through examination of the recorded auditory signal from the experimental setup. I think the future of earthquake physics will rely heavily on machine learning to process massive amounts of raw seismic data. Our work represents an important step in this direction.”

The team believes that not only does the work have potential significance to earthquake forecasting but the approach is far-reaching, applicable to potentially all failure scenarios including nondestructive testing of industrial materials brittle failure of all kinds, avalanches and other events.

The machine learning technique used in this project also identifies new signals, previously thought to be low-amplitude noise, that provide forecasting information throughout the earthquake cycle.

“These signals resemble Earth tremor that occurs in association with slow earthquakes on tectonic faults in the lower crust. There is reason to expect such signals from Earth faults in the seismogenic zone for slowly slipping faults,” Johnson added.

Handheld device identifies cancer in seconds
To help improve treatment and reduce the chances of cancer recurrence, University of Texas at Austin researchers have invented a powerful tool that rapidly and accurately identifies cancerous tissue during surgery, delivering results in about 10 seconds—more than 150 times as fast as existing technology. Called the MasSpec Pen, the handheld instrument gives surgeons precise diagnostic information about what tissue to cut or preserve.

The MasSpec Pen rapidly and accurately detects cancer in humans during surgery, helping improve treatment and reduce the chances of cancer recurrence. (Source: University of Texas at Austin)

In tests on tissues removed from 253 human cancer patients, the MasSpec Pen took about 10 seconds to provide a diagnosis and was more than 96 percent accurate. The technology was also able to detect cancer in marginal regions between normal and cancer tissues that presented mixed cellular composition. The team expects to start testing this new technology during oncologic surgeries in 2018.

This research was accomplished by an interdisciplinary team, merging the fields of chemistry, engineering and medicine. Other main contributors include Thomas Milner, professor of biomedical engineering in UT Austin’s Cockrell School of Engineering and his lab members; Jialing Zhang, research associate at the Eberlin Lab at UT Austin who led the experimental work with other lab members; Anna Sorace, assistant professor at UT Austin’s Dell Medical School; Chandandeep Nagi and Wendong Yu, professors of pathology at Baylor College of Medicine, and Jinsong Liu, professor of pathology at University of Texas MD Anderson Cancer Center.

The team and UT Austin have filed U.S. patent applications for the technology and are now working to secure worldwide patents.

The process works by taking a molecular fingerprint obtained by the MasSpec Pen from an uncharacterized tissue sample, which is then instantaneously evaluated by software, called a statistical classifier, trained on a database of molecular fingerprints that the team gathered from 253 human tissue samples. The samples included both normal and cancerous tissues of the breast, lung, thyroid and ovary.
When the MasSpec Pen completes the analysis, the words “Normal” or “Cancer” automatically appear on a computer screen. For certain cancers, such as lung cancer, the name of a subtype might also appear.

In tests performed on human samples, the device was more than 96 percent accurate for cancer diagnosis. The team has also demonstrated that it accurately diagnoses cancer in live, tumor-bearing mice during surgery without causing any observable tissue harm or stress to the animals.
Physicians can operate the disposable handheld device easily. It requires simply holding the pen against the patient’s tissue, triggering the automated analysis using a foot pedal, and waiting a few seconds for a result. Meanwhile, the pen releases a drop of water onto the tissue, and small molecules migrate into the water. Then the device drives the water sample into an instrument called a mass spectrometer, which detects thousands of molecules as a molecular fingerprint.

The process is also low-impact for patients. When designing the MasSpec Pen, the team made sure the tissue remains intact by coming into contact only with water and the plastic tip of the MasSpec Pen during the procedure.