Modern artificial intelligence uses complex algorithms to perform all kinds of tasks in an instant, for example to determine the feeling of a client based on his or her examination or to identify specific features of an image . However, the brightest moments of artificial intelligence come from the creativity with which we use these algorithms. People used the AI for generate new sports, transform scribbles into realistic landscapesand now, MIT has found a way to detect breast cancer up to five years in advance using an in-depth image classification model.
The MIT Computer Science and Artificial Intelligence Laboratory (CSAIL) and the Massachusetts General Hospital (MGH) have used mammograms and known results from more than 60,000 patients to form their new model to the smallest visual detail from the human eye. Well-trained doctors do not miss these predictive patterns simply because they may seem too small to be noticed, but because more subtle patterns just do not attract enough attention. An image classification model that can categorize thousands of scans down to the smallest detail can quickly solve this daunting task.
Regina Barzilay, a professor at MIT (and breast cancer survivor), explains how this new model can improve treatment plans:
Rather than taking a single approach, we can customize the screening for cancer risk in women. For example, a doctor might recommend to a group of women to have a mammogram every two years, while another higher-risk group could undergo additional MRI screening.
When doctors can order mammograms according to the needs of the patient, they can avoid unnecessary exposure to radiation and the cost of potentially unnecessary examinations. Although existing models can accurately identify 18% of patients in the high-risk category, this new model increases this number by up to 31%. Its success is based on the team's approach to its development. For the first time, a model of breast cancer prevention targets women individually. It also takes into account racial diversity, where earlier models focused mainly on white populations. This not only contributes to increased accuracy, but significantly reduces the breast cancer mortality rate among African-American women.
As demonstrated by MIT and HGM, well-trained image classification models can help doctors save lives. Although no AI gives perfect results, image classification algorithms have matured and become reliable in many applications, especially in specific models such as this one. You need a little more than a good idea, relevant data and a little time to create a successful image recognition model. Services like Clarifai, Microsoft Azure, IBM Watson, Vize and others offer free, bespoke customized training platforms that require no programming skills. Thanks to these algorithms, accessible to all, we have all the necessary resources to train AI to solve problems and help others. It takes time and care to safely integrate a successful experience into the practice of diagnostic medicine; This approach will likely see many revisions as it develops outside of a single hospital. But the first results are promising.
A laser tenth of solar power on Earth officially debuted in March when Romanian researchers conducted the first successful test at 10 petawatts (PW). The laser is one of three of an international project in Europe called Extreme Light Infrastructure. To date, it is the most powerful laser ever built and the most concentrated power on the planet.
It is hard to exaggerate the enormity of 10PW, which equates to 10 trillion watts. It's only a few years ago that this site was referring to a simple 1PW laser like a death star. The new is 10 times stronger. In comparison, laser pointers sold in the United States are limited to a maximum of 0.005 watts for security reasons.
The laser in question is part of the Extremely Light Infrastructure Project (ELI), an effort initiated by scientists in Europe in the mid-2000s and led by French scientist and Nobel laureate Gerard Mourou. The project aims to advance not only the research infrastructure for lasers, but also applied sciences.
The program was funded by the European Commission and in just a few years, three countries have been selected to host three new lasers: Romania, Hungary and the Czech Republic. At present, the project has received more than 850 million euros, mainly from the European Regional Development Fund.
The 10PW laser resides in a newly built lab called Extreme Light Infrastructure – Nuclear Physics installation (ELI-NP) in a town called Măgurele, not far from the capital Bucharest. ELI-NP is dedicated to the study of photonuclear physics and its applications. The other two laboratories are called ELI-Beamlines in the Czech Republic, focusing on secondary sources of radiation and short-pulse particles, and ELI-ALPS or Attosecond Light Pulse Source in Hungary.
With 10 PW of power, scientists can literally vaporize matter, opening up new perspectives on what happens during a supernova. This is only one example, albeit rather epic. This type of power in a laser also helps to study how heavy metals are formed.
In terms of more practical research, ELI-NP will work to advance medical research on proton cancer treatment as well as research into new methods of treating radioactive waste. This could also help to create new ways to find and characterize nuclear material, allowing security teams to analyze, for example, incoming shipping containers in search of dangerous and illegal content. The formal research phase of the project is scheduled to begin in early 2020.
The power of lasers has increased in recent decades. "The laws of light-matter interaction change fundamentally because of the predominance of relativistic effects in the dynamics of charged particles under the influence of laser light," according to ELI. From this fundamental change, scientists can develop new methods for generating X-rays, gamma rays, and high-energy particles. These new methods, in turn, open up new opportunities for use in different scientific fields, whether it is medical research or materials science.
As much as anyone would want to imagine a giant satellite antenna pulling from its center (ahem, Death Star), the reality of a 10PW laser is a little more mundane. The laser itself is inside a room and the scientists who operate it are of course behind a computer.
For the laser to have both power and precision, it relies on two systems working together. One is the high-powered laser system, which itself includes two laser arms. They are what the laser pulses deliver. The second piece is the laser beam transport system, which directs the pulses where they need to go with micrometric precision. This second component is not a laser at all, but rather "one-meter wide adjustable aperture mirrors installed in a vacuum system of pipes and housings", according to ELI-NP information. The entire system requires a highly controlled environment with respect to air quality and vibration, at least one image in a clean room, in the scientific sense of the term.
If we could take a look at the inside of the protective chamber while the laser was working, the beam would be visible to the human eye and reddish, although it is close of the infrared radiation limit, according to Dr. Nicolae Zamfir, project manager at ELI-NP. A high energy laser pump would be visible to the eye, too, appearing in green.
Mr. Zamfir also explained the size of the laser beam, about 60 cm or just under two feet in diameter, and the area that he can target: "The beam is focused on mm2 only in the Interaction (chamber), "he added.
If everything goes as planned, the ELI-NP 10PW laser will be even more powerful. According to optics.org (as well as a Job offer for an engineer at ELI-NP – a simple bachelor's degree required, two 10PW lasers will combine together "deliver focused laser intensities up to 1023 watts per square centimeter, at a wavelength of 820 nanometers and a pulse duration of 25 femtoseconds. "The intensity of the current is 1015.
In addition to the three sites in Romania, Hungary and the Czech Republic, ELI plans to create a fourth installation and a laser, with a location to be determined. This should be an order of magnitude stronger than that of Romania.
Top image credit: Getty Images
As the AI hype cycle is built, we have been treated to a plethora of claims about the improvements and breakthroughs that technology can bring. One of the most fundamental – and potentially important – has been the idea that we can use AI to find new drugs and treatments for existing conditions for which current options have been lacking.
This promise has himself now come short. IBM has announced that it will stop selling its Watson AI system as a drug discovery tool. This is a well-publicized retreat for the company, which has aggressively commercialized artificial intelligence, which has proven useful for these purposes and has encountered problems last year when reports indicated that his systems had formulated inappropriate and dangerous recommendations for patients with effect).
While IBM cites slowing sales as a reason for withdrawal, deeper issues are potentially responsible. A recent deep diving by IEEE Spectrum puts the context around these issues. Result: After years of work and a number of moonshot projects, IBM has remarkably little to offer for its efforts. And the company has created some ill-will toward itself, writes IEEE, because it has adopted an aggressive, marketing-focused approach, focused on AI and Watson, promising grandiose achievements that did not accurately describe what the system could really achieve.
Watson seduced the world with his performance on Jeopardy and his ability to analyze the relationships between words rather than treat them as search terms. In theory, Watson could use his engine to sort quantities of medical data in the same way, finding the hidden signal in a system full of noise. Reality did not cooperate. Watson, from IBM, has not been among the few researches into the use of AI to improve outcomes for patients.
The IEEE party is struggling to note that IBM has had to face tremendous challenges in trying to put its artificial intelligence program online and used it effectively for medicine human. Nothing like Watson (or what Watson was supposed to be) has ever existed before. Nobody knew how to build it. Yoshua Bengio, an Artificial Intelligence Researcher at the University of Montreal, summarized efforts to help AI understand medical texts and terminology in the following way: "Our NLP has seen a tremendous improvement compared to five years ago humans. "
Watson's problem was not that it did not work. The problem is that Watson is not doing the right thing Things. Although he quickly learned to ingest and process large amounts of data, he had a hard time identifying the pieces of information contained in a study that could lead physicians to actually change their process. care. This is especially true if the relevant information was incidental to the main subject of the research.
Because patients' data was not always properly formatted or even chronologically ordered, the software was unable to understand the patient history. And the system was unable to compare new cancer patients with old patient databases to uncover hidden treatment patterns, as such practices would not be considered evidence-based. Making a strong recommendation from an evidence-based medicine requires double-blind studies, meta-analyzes, and analyzes of systemic evidence, not an AI system claiming to have found a similarity between different types of patients .
It is not clear what is next for Watson, if anything. The tool has had some success in narrow and customized applications with less ambiguity. But despite the dozens of planned initiatives, waves of media hype and considerable investment, IBM's Watson for Drug Discovery has clearly missed its own goals.
The video was released Wednesday afternoon, in coordination with the conclusion of the filming of the 35th season (!) Of the game show.
In the short sequence, Trebek said: "I wanted to thank you again for your messages of encouragement and support, especially for the many cards I have received from young people.
He also mentions rumors about his health and promises to come back for the next season this fall: "What's wrong with you, I feel good, I'm continuing my therapy, and [the show’s staff] is now working on our next season (…). I look forward to seeing you again in September with all kinds of good things. "
Trebek, 78, welcomed Jepoardy! since the revival of the series in 1984 and the signing of a contract extension in October 2018, which earned him to be the host until 2022.
One of the problems in convincing people to take computer security seriously is that in a nutshell, boring. From time to time, however, someone demonstrates a flaw with the potential to break the walls of boredom surrounding the subject and register with the public consciousness. How do you diagnose this malware running on computer tomography and MRI computers? a lot was.
The scientists in question demonstrated this ability by deploying it themselves, to prove that the security vulnerabilities of medical equipment were a real problem. Hospitals are potentially attractive targets for ransomware and malware more generally because they contain patients. Interrupting the ability to access this data could literally cost lives if the proper systems were penetrated.
The report notes that hospital security systems associated with PACS systems (image archiving and communication systems) used by both CT and MRI scanners are very outdated and often poorly managed. An online search with Shodan.io (a search engine for IoT devices) revealed 1,849 medical image servers (DICOM) and 842 PACS servers exposed to the Internet. Researchers have shown that these services are vulnerable to external attacks and internal penetration. They to write:
As 3D medical scanners provide strong evidence of health status, an attacker with access to a scan would have the power to change the outcome of the patient's diagnosis. For example, an attacker may add or remove evidence of aneurysms, heart disease, blood clots, infections, arthritis, cartilage problems, torn ligaments or tendons, tumors to the brain, heart or spine and other cancers.
The idea that nation-states or other harmful actors could target specific individuals might have seemed far-fetched a decade ago, but not longer. The United States and Israel are believed to have used Stuxnet to neutralize Iranian centrifuges. Thanks to Ed Snowden, we know that the NSA intercepted specific computers to install rootkits while she thought she had reason to do so. An actor recently launched an attack against Asus to try to infect 600 specific systems with malware based on their MAC addresses. It is not known how the attackers managed to hit Asus. The identity of the targets that they were trying to infect is not either. But the idea of a targeted attack intended to sow discord or uncertainty by targeting, for example, those who stand for election can no longer be relegated to the rank of science fiction.
By using machine learning, researchers were able to inject false data into CT scans that were shown to trick medical professionals with the image. 2D images to have It is difficult to manipulate qualified analysts, even when they were created by a digital artist using Photoshop. The authors note that even when an artist is employed, "it is difficult to inject and realistically eliminate cancer."
However, the generation of 3D images of cancer uses an automatic learning model called GAN (Generative Adversarial Network). A GAN opposes two different machines one to the other. The generator is trying to create fake images. The other, the discriminator, tries to identify these images as fake. The results of this long-term training are a generator that finally "learns" to deceive the discriminator by discovering what types of fakes will trigger and will not trigger its detection algorithms. If your discriminator is good enough, what will you get in a 3D CT or MRI image? It's very hard to spot, especially since the people paid to look at it do not expect it to be a problem. wrong in reality. .
The team performed an intrusion test (pen test) and was able to infiltrate and install the Raspberry Pi server in the hospital network. They write:
To verify that we could intercept and manipulate the scans, we scanned a medical dummy (Fig. 5). We found that the dummy's analysis had been sent twice over the network: once in clear text over TCP to an internal web-viewing service and another time to the PACS storage server using of TLSv1.2. However, to our surprise, the payload of the TLS transmission was also clear. In addition, in less than 10 minutes, we obtained the user names and passwords of more than 27 staff and physicians through the multicast Ethernet traffic containing HTTP POST messages sent in clear text. These vulnerabilities have been revealed to hospital IT staff and their PACS software provider.
To test their fake, the team recruited three radiologists and tested them both in a blind and open trial. Radiologists (with an experience of 2, 5 and 7 years) were first tested to determine if they could identify unmodified scans to detect the presence or absence of cancer. All performed well. When given the modified scans – those in which the cancer had been injected or removed – the rate of fake cancer injection was greater than 99%. The success rate of elimination – in which cancer signs were removed from images – was 95.8%.
Informing radiologists of the test in the open scenario has reduced the success rate of 90% cancer injection and cancer. removal success rate at 70 percent. However, this has also resulted in a significant increase in the number of false positives. The AI tools used by radiologists for cancer detection to help them analyze the images were deceived in 100% of the cases.
The authors conclude: "In summary, radiologists and AI are very susceptible to CT-GAN image alteration attacks." Although some cases of cancer may help to alleviate some of them, error rates and risk scores suggest that the attack may not be reported in most cases. "
The threat of deepfakes has mostly been described as attacking truthful public messages or breaches of privacy. Using AI to record fake messages from politicians stating their support for causes they do not advocate, or to insert someone else's face into hardcore pornography is a threat to accurate reporting and copying. private life. There is now evidence that they can be used to modify, even in the places where we need the most accurate data. The risk of attack is low, but the risk of disturbance in case of successful intrusion is quite high.
Featured Image courtesy of Wikipedia
The iconic host of Jeopardy! revealed his diagnosis of stage 4 pancreatic cancer in a video posted on Wednesday on the series' YouTube channel, explaining that he wanted to prevent fans from "reading or hearing exaggerated or inaccurate reporting."
Now, just like 50,000 other people in the United States each year, this week, I was diagnosed with stage 4 pancreatic cancer. Normally, the prognosis is not encouraging, but I'm going to fight , and I will continue to work. And with the love and support of my family and friends and with the help of your prayers, I plan to beat statistics on the low survival rate for this disease.
Despite the diagnosis, Trebek, 78, was optimistic enough to joke that he was contractually obliged to continue working.
"The truth has been said, I have to, because under my contract I have to host Jeopardy! another three years! "I declared." So help me. Keep the faith and we will win. We are going to do it. Thank you."
Fans and other game organizers have taken Twitter to express their support.
The Sajak family is deeply saddened to learn Alex Trebek's fight against cancer. Our hearts go to him and his family. But I do not know anyone who is stronger and more determined and I will never bet against him. We, and the whole country, shoot for you, Alex.
– Pat Sajak (@patsajak) March 6, 2019
We are all drawn for Alex Trebek who just revealed that he was suffering from stage 4 pancreatic cancer. I have to fight and keep working. We are with you, Alex.
– David Muir (@DavidMuir) March 6, 2019
As the daughter of a man who also fights stage 4 cancer with a low survival rate, know this, Mr. Trebek: you can beat the odds. You can overcome the obstacle. You can exceed all expectations. As a Jeopardy spectator, I wish you all the best.
– Katie Nelson (@katienelson210) March 6, 2019
Alex Trebek is a real factor in my book. Crack block cancer on the ground.
– Three year letterman (@ 3YearLetterman) March 6, 2019
I do not even believe in god but I'm going to pray for alex trebek
– Brandy Jensen (@BrandyLJensen) March 6, 2019
I wish good luck to Alex Trebek, who has been as much a part of my life as the sports personalities of my childhood. Developed a geek love for weird anecdotes through Alex and this show and played against my mom every night at college and high school. https://t.co/LJcL3CWDsR
– Darren Rovell (@darrenrovell) March 6, 2019
Answer: Alex Trebek
Question: Who is the most respected TV personality?
Truly a moment when "thoughts and prayers" make sense, millions of people are thinking of you tonight!
– TRINITYPREZ (@ TRINITYPREZ) March 6, 2019
Trebek hosts Jeopardy! since its rebirth in 1984, and renewed his contract in October 2018 to continue hosting until 2022.