This week’s list of data news highlights covers April 20-26, 2019, and includes articles about an autonomous catheter and a replacement for LIDAR in autonomous vehicles.
1. Identifying Patients with PTSD
Researchers from New York University and SRI International, a nonprofit research institute, have developed an algorithm that can detect post-traumatic stress disorder (PTSD) by analyzing audio recordings of patient interviews. The algorithm analyzed more than 40,000 speech characteristics of patients, such as tension in the larynx, finding that 18 of the characteristics, including flatter speech, are markers of PTSD. The algorithm uses these 18 characteristics to identify whether a patient has PTSD with 89 percent accuracy.
Researchers from the Joint Research Centre, the European Commission’s science and knowledge service, have developed an AI system that analyzes tweets to help identify an area’s risk of flooding. The researchers trained a multi-lingual classification system on 7,000 annotated messages. The researchers then tested the system on 14,000 tweets from recent flooding in Calabria, Italy, finding that it accurately identified tweets in areas experiencing actual flooding. The system could help shorten the response time of first responders to areas experiencing flooding.
3. Developing an Autonomous Catheter
Researchers from Harvard University, the University of Strasbourg in France, and the Taipei Veterans General Hospital in Taiwan have developed an autonomous robotic catheter that can navigate to the correct location in a heart by itself. The catheter uses touch and optic sensors and a machine learning algorithm the researchers trained on roughly 2,000 heart tissue images to guide itself. The catheter navigated to the correct location 95 percent of the time during 83 trials on pigs.
4. Converting Brain Signals into Speech
Researchers from the University of California, San Francisco, have developed an AI system that can translate brain signals into speech. The researchers recorded the neural impulses being sent from the brain to muscles while five patients with epilepsy spoke 100 phrases during brain surgery. The researchers fed the recordings into a system that matched the signals to a database of muscles movements and resulting sounds, finding that the system translated the signals into audible speech with 70 percent accuracy. Similar systems could eventually help people with impaired speaking abilities communicate.
5. Developing an Alternative to LIDAR for Autonomous Vehicles
Researchers from Cornell University have developed a system that allows autonomous vehicles to see in 3D without using LIDAR, which can add $10,000 worth of components to a vehicle. Autonomous vehicles need to be able to see in 3D to estimate the distance between it and other objects, and the researchers’ system converts images from inexpensive cameras into a 3D point cloud, which is a set of data points in space, and then uses an algorithm to locate objects with 74 percent accuracy.
6. Using a Smartphone to Carry Out Lab Tests
Healthy.io, a startup based in Tel Aviv, has developed a smartphone-based system that can detect if individuals are at risk of kidney failure by allowing them to perform urine tests at home. Healthy.io’s system uses a smartphone’s camera and machine learning to analyze a dipstick to provide instant analysis, and the system can also detect if a patient has a urinary tract infection by analyzing the dipstick. The system is the first smartphone-based system the U.S. Food and Drug Administration has approved as a Class II clinical-grade diagnostic device, which are devices that could have a moderate to high risk to the patient or user.
7. Perfecting the Brewing Process
Sugar Creek Brewing Company, a craft brewer in North Carolina, has used AI and networked sensors to prevent beer spillage and the inconsistent filling of bottles during the manufacturing process. A camera takes a picture of each bottle that passes through the brewer’s package line, allowing AI to analyze the fill and amount of foam in each bottle. This data and data on the temperature and pH of the beer has helped the brewer save more than $10,000 per month in reduced product loss.
8. Tracking Patient Health Remotely
Current Health, a Scottish startup, has developed a wearable device that uses machine learning to notify doctors when it detects problematic changes in a patient’s health. Patients can wear the device while at home, and the device collects data on a patient’s breathing, oxygen saturation, pulse, temperature, and mobility. The device can send notifications to doctors’ mobile devices and provide its data to electronic health records.
9. Fighting the Opioid Epidemic
New Jersey has developed a tool called the Integrated Drug Awareness Dashboard to help its public safety and law enforcement agencies share opioid-related information. The state’s agencies have struggled to share data, and the dashboard will aggregate law enforcement and healthcare information, including on opioid-related arrests and naloxone administrations, which can prevent overdoses, in one place. By combining data from different agencies, New Jersey can more effectively allocate its resources to different locations.
10. Making Quantum Computers More Functional
A researcher from Google has developed a quantum version of a classical algorithm for quickly multiplying large numbers. While quantum computers can have significantly more processing power than classical computers, it can be inefficient to use quantum computers because they cannot selectively forget information. For example, classical computers will break large numbers in multiplication into smaller figures, multiply the smaller figures to get a new figure, and then delete the smaller figures. A quantum computer cannot forget the smaller figures because it uses qubits, which are entangled, meaning the computers cannot alter one of the qubits without affecting all the other qubits. The researcher’s algorithm reduces the complexity of multiplication on a quantum computer by adding intermediate values in multiplication directly to the output, avoiding the need to store the intermediate values themselves.
Image: Dllu
