This week’s list of data news highlights covers June 8-14, 2019, and includes articles about an AI system that can predict if a person will develop psychosis and an AI system that can make humans in photos come “alive.”
Researchers from Emory University and Harvard University have developed an AI system that can predict whether an at-risk individual will develop psychosis with 93 percent accuracy. The researchers trained their system on online conversations of 30,000 Reddit users and on interview data of 40 individuals who were at risk of developing psychosis. The researchers found that higher than normal usage of words related to sound and a higher rate of using words with similar meaning was linked to developing psychosis.
Researchers from the University of Washington and Facebook have developed AI software that can turn humans in images into 3D animated characters. The characters can walk, run, sit, or jump. The software works by labeling each body part to form a map, using the map to build a 3D model of the character, and estimating the bodyweight of the character to help create realistic motion.
A group led by researchers from the Medical University of Vienna have shown that machine-learning algorithms can diagnose benign and malignant skin lesions better than human experts. The researchers tested the ability of more than 500 dermatologists, dermatology residents, and general practitioners and nearly 150 algorithms to diagnose lesions in 30 images. The algorithms, which the researchers trained on 10,000 dermatoscopic images, achieved two more correct diagnoses on average than the human experts, and the top three algorithms averaged 85 percent accuracy.
Researchers from Stanford University have developed an AI tool that automatically redacts information from police reports that imply a suspect’s race. The tool uses computer vision to recognize words and will remove descriptions of race, hair and eye color, and the names of people and neighborhoods. San Francisco is implementing the tool on July 1, and the tool could help mitigate the influence of racial bias when prosecutors decide whether to charge a person with a crime.
Researchers from the Wellcome Sanger Institute, a genomics and genetics research institute in the UK, Emory University, and the U.S. Centers for Disease Control and Prevention have sequenced the DNA of over 20,000 pneumonia strains from infected individuals in 51 countries. The researchers discovered more than 600 different genetic strains and found that the levels of strains not addressed by the standard vaccine increased after the introduction of the vaccine. This data can help researchers predict strains that are likely to emerge in response to vaccine use.
Researchers from the University of Washington used data from robots, satellites, and seals fitted with head sensors to understand why a hole the size of South Carolina occasionally forms in the floating ice in Antarctica’s Weddell Sea. The seals frequently dive to depths as deep as 6,500 feet, providing the researchers access to data they usually could not gather. The data helped the researchers conclude that the hole is a result of intense storms, an underwater mountain, and saltier water.
Researchers from MIT have developed an algorithm that helps robots predict the paths humans will take in structured environments, such as factory floors. The algorithm compares a person’s trajectory with a library of previously recorded trajectories and aligns the trajectories in relation to their distance and timing. This process helps robots better anticipate stops and overlapping paths in a person’s path, which reduces the time robots waste by being overly cautious.
Researchers from Stanford University have developed an AI tool that helps clinicians more accurately identify aneurysms in computed tomography (CT) scans. The researchers trained a neural network on more than 600 scans of the head, teaching the model how to locate areas of the brain where an aneurysm could have occurred. The tool increased clinician’s accuracy identifying aneurysms from 89 percent to 93 percent by highlighting areas in brain scans that are likely to contain an aneurysm.
Researchers from Adobe and the University of California, Berkeley have developed an AI tool that can detect manipulated images of faces. The researchers trained the tool’s neural network on a database that included before and after images of Photoshopped photos. The tool detects edited faces with 99 percent accuracy, compared to 53 percent accuracy for humans.
Researchers from MIT and Massachusetts General Hospital have developed a deep learning model that predicts a woman’s risk of developing breast cancer more accurately than the standard clinical methods, which use factors such as genetics and family history. The researchers trained and tested their model on nearly 90,000 mammograms from 40,000 women, and the model learned to automate breast density measurements, which can indicate breast cancer risk but which radiologists subjectively measure. The researchers found that patients with non-dense breasts that the model assessed as high risk had nearly four times the incidence of breast cancer as patients with dense breasts that the model evaluated as low risk.