When it comes to biomedical research, there are hundreds of research papers being published every day. Yet it can be difficult to predict what research will make it out of the lab setting and lead to clinical applications. Recently, a machine learning model developed by the Office of Portfolio Analysis, or OPA, at the National Institutes of Health (NIH) was able to determine the likelihood of a biomedical research case being used in clinical trials or guidelines. According to the OPA, the citation of a research article in a clinical trial is an early indicator of translational progress or the use of research findings as a potential treatment for disease.
As reported by AI Trends, the researchers at the OPA created a new metric for their machine learning model to use, dubbed Approximate Potential to Translate, or APT. According to the OPA Director, George Santangelo, bio-medicinal translation can be predicted based on the reaction of the scientific community to the research papers that a project based on. Santangelo said that there are distinct trajectories for the flow of knowledge which can predict the success or failure rate of a paper influencing clinical research.
The creation of the APT metric coincides with the release of the NIH’s second version of the iCite tool. iCite is a browser-based application that provides information about journal publications based on their specific field of analysis. Moving forward, the iCite tool will return the APT values for queries.
The process of adapting laboratory research into clinical applications is a complex tasks that often takes years. Attempts have been made to expedite this process, due to the many variables involved in the task, it can be difficult to assess the translational process. As explained by Santangelo, machine learning algorithms are a powerful tool that could
enable clinicians to better understand which research papers are likely to prove useful in the clinic. As the team of researchers experimented with and refined their APT metric, useful predictive patterns began to materialize.
“I think the most important one that we focus on is the diversity of interest from across the fundamental to clinical research axis. When people across that axis — from fundamental scientists often in the same field as the work that’s being published, all the way to people in the clinic — show an interest in the form of citations in those papers, then the likelihood of eventual citation by a clinical trial or guideline is quite high.”
According to Santangelo, the selected features show genuine promise in predicting the translation from research paper to a clinical method. Data on a publication collected over at least two years from the date of publication often give accurate predictions about a paper’s eventual citation in a clinical article.
Santangelo explained that thanks to the new metric and machine learning algorithms the researchers can have more complete knowledge of what is going on in the literature and that this allows better insight into the research areas which are more likely to appeal to clinical scientists.
Santangelo also explained that their algorithms integration into the iCite tool is intended to leverage the free, open nature of the NIH’s Open Citation Collection database.
The NIH Open Citation Collection database is currently comprised of over 420 million citation links and growing. The Santangelo team’s algorithm will be presenting the APT values for these citations when iCite 2.0 launches in the future.
Many databases are restrictive and propriety, and according to Santangelo, these barriers inhibit collaborative research. Santangelo opines that there isn’t a fantastic justification for keeping the data behind a paywall and that because their algorithm is supposed to let others see the calculated APT values, it wouldn’t be beneficial to use proprietary data sources.
Team Develops Blood-Sampling Robot
A blood-sampling robot, able to perform as well or better than humans, has been developed by a team at Rutgers University. It was tested during the first human clinical trial of an automated blood drawing and testing device.
Because the device can deliver quicker results, healthcare professionals would not have to spend so much time sampling blood. It would allow them to focus more on the treatment of patients within hospitals and other settings.
The results were published in the journal Technology, and they were comparable to or exceeded clinical standards. The overall success rate for the 31 participants who had their blood drawn was 87%. 25 people had veins that were easier to access, and that success rate was 97%.
Within the device is an ultrasound image-guided robot that draws blood from veins. One of the possible developments is a fully integrated device that includes a module to handle samples and a centrifuge-based blood analyzer. This could be used in ambulances, emergency rooms, clinics, doctors’ offices, hospitals, and bedsides.
The most common clinical procedure, numbered at more than 1.4 billion performed daily in the United States, is Venipuncture. This is a process that involves inserting a needle into a vein to get a blood sample or perform IV therapy. However, previous studies have shown that clinicians fail in 27% of patients without visible veins, 40% of patients without palpable veins and 60% of emaciated patients.
With the repeated failure to start an IV line boost comes an increased risk of phlebitis, thrombosis, and infections. It could also require the targeting of large veins in the body or arteries, and this is riskier and more costly. Because of this, venipuncture is one of the leading causes of injury to patients and clinicians. Other problems associated with difficulty accessing veins are that it can increase procedure time by up to an hour, it requires more staff and the estimated costs are more than $4 billion a year in the United States.
Josh Leipheimer is a biomedical engineering doctoral student in the Yarmush lab in the biomedical engineering department in the School of Engineering at Rutgers University-New Brunswick.
“A device like ours could help clinicians get blood samples quickly, safely and reliably, preventing unnecessary complications and pain in patients from multiple needle insertion attempts,” Leipheimer said.
The team hopes that the device can eventually be used in procedures such as IV catheterization, central venous access, dialysis and placing arterial lines. They will now work to refine the device and increase success rates in those patients with difficult veins to access.
In order to improve its performance, data will be taken form this study and used to enhance artificial intelligence in the robot.
The Rutgers co-authors include Max L. Balter and Alvin I. Chen, both graduates with doctorates; Enrique J. Pantin at Rutgers Robert Wood Johnson Medical School; Professor Kristen S. Labazzo; and principal investigator Martin L. Yarmush, the Paul and Mary Monroe Endowed Chair and Distinguished Professor in the Department of Biomedical Engineering. The study also had contributions from a researcher at Icahn School of Medicine at Mount Sinai Hospital.
The newly developed device by the team at Rutgers is just another example of how robotics and artificial intelligence are overtaking the healthcare industry. These devices will greatly assist those working within healthcare, and they will make procedures and other forms of care much more successful.
How AI Is Being Used In The Fight Against The Wuhan Coronavirus
Artificial intelligence is being leveraged in the fight against the Wuhan Coronavirus. Artificial intelligence as being employed by researchers track the spread of the disease and to research potential treatments for the virus.
The Wuhan Coronavirus manifested in China in December, and in the two months since then it has spread across China and to other parts of the globe. It’s still unknown just how contagious the virus is and how quickly the virus could spread, although there are currently more than 40,000 confirmed cases within China. In order to get a better understanding of how the virus might spread and how fast the virus can spread, researchers are employing machine learning algorithms focused on data pulled from social media sites and other parts of the web.
Over the course of the past week, the rate of infection seems to have decreased somewhat, but it’s unclear if the disease is falling under control or if new cases are becoming harder to detect. While other countries around the world have only seen a few cases of coronavirus, in comparison to China, the world health community remains concerned about the virus’s ability to spread. Researchers are trying to get ahead of the viruses’ spread by using machine learning and big data collected from the internet.
As reported by Wired, an international team of researchers have extracted data from various parts of the internet, including posts from doctors and medical groups, public health channels, social media posts, and news reports, compiling a database of text that might relate to the coronavirus. The researchers then analyze the data for signs that the virus could be spreading outside of China’s borders, making use of machine learning techniques in order to find relevant patterns in the data that could hint at how the virus is behaving.
The researchers sift through social media posts looking for potential symptoms of coronavirus, centering their search on regions where doctors think cases may manifest. The social media posts are processed using natural language processing techniques, techniques which can distinguish between posts where a person mentions their own symptoms versus someone saying symptom-related words in another context (such as discussing news about the coronavirus).
According to Alessandro Vespignani, as Wired reported, Northeastern University professor and expert contagion analyst, argues that even with advanced machine learning techniques it’s often difficult to track the spread of the virus because the characteristics of the virus are still somewhat unknown, and most social media posts are from media companies and currently about the outbreak in China. However, Vesignani believes that if the virus ever did take hold in the US it would become easier to monitor thanks to more posts concerning the virus.
Despite the challenge in gaining relevant information about the potential behavior of the coronavirus, the model created by the researchers does seem to be reasonably effective at finding clues within a large sea of social media posts. The model used by the researchers was able to find evidence of a viral outbreak on December 30th, although it took time to determine just how serious the situation would become. Crowdsourced information could improve the effectiveness of disease tracking models even further, as it enables the more efficient collection of relevant data regarding the virus. As an example, an analysis of data crowdsourced by Chinese physicians suggests that people younger than 15 years of age are more resilient to the virus.
Artificial intelligence can also be combined with data collected from mobile devices to build models that can potentially predict the direction a virus is spreading as well as the rate of a spread. For instance, Researchers from University of Southampton used mobile data to determine the path that the virus may have taken as it moved out of Wuhan in the days following its manifestation. Other researchers analyzed data collected by Tencent, a Chinese mobile app developer, and found that the restrictions imposed by the Chinese government potentially reduce the virus’ spread, buying vital time to develop a plan of attack.
As Fortune reported, the startup Insilico Medicine has made use of artificial intelligence to identify molecules that could potentially treat the coronavirus. Insilico’s AI identified thousands of possible drug molecules over the course of four days. Insilico explained that the 100 most promising candidates will be synthesized and all of their research on molecular structures will be published for other researchers to take advantage of. Medical researchers and companies are fast-tracking the development and testing of treatments, with the US-based biotech company Gilead planning to start the immediate testing of a new antiviral drug within the Wuhan region.
After Insilico decided to begin researching treatments, it focused its research on an enzyme called 3C-like protease. The coronavirus relies on this enzyme to reproduce and spread. According to Insilico, it decided on this specific enzyme because it’s quite similar to other viral proteases whose structures have already been documented, and because Shanghai Tech University had developed a model of the 2019-nCoV 3C-like protease. In the span of four days Insilico was able to generate hundreds of thousands of candidate molecules and choose only the hundred or so that were most likely to be useful. The results of the research were recently published in the repository bioRxiv and on Insilico’s website.
AI Being Used to Analyze Retinal Images
In a newly developed approach, artificial intelligence (AI) is being used to analyze retinal images. The system could be used by doctors in order to select the best treatment for patients suffering from vision loss from diabetic macular edema, a diabetes complication. That problem often leads to vision loss among working-age adults.
One of the first types of therapy that is often used as a line of defense against diabetic macular edema is anti-vascular endothelial growth factor (VEGF). The problem with VEGF agents is that they do not work for everyone. Those who could benefit from the therapy need to be identified first since it requires multiple injections. Those injections cost a lot, and they are burdensome for both patients and physicians.
The leader of the research team is Sina Farsiu from Duke University.
“We developed an algorithm that can be used to automatically analyze optical coherence tomography (OCT) images of the retina to predict whether a patient is likely to respond to anti-VEGF treatments,” she said. “This research represents a step toward precision medicine, in which such predictions help clinicians better select first-line therapies for patients based on specific disease conditions.”
The work was published in The Optical Society (OSA) journal Biomedical Optics Express. In the journal, Farsiu and her team demonstrated how the new algorithm is capable of accurately predicting whether a patient is likely to respond to anti-VEGF therapy, after just one volumetric scan.
“Our approach could potentially be used in eye clinics to prevent unnecessary and costly trial-and-error treatments and thus alleviate a substantial treatment burden for patients,” Farsiu said. “The algorithm could also be adapted to predict therapy response for many other eye diseases, including neovascular age-related macular degeneration.”
The newly developed algorithm is based on a novel convolutional neural network (CNN) architecture. A CNN is a type of artificial intelligence, and it assigns importance to various aspects or objects in order to analyze images. The algorithm was used by the researchers to examine images acquired with OCT, which is a noninvasive technology. OCT produces high-resolution cross-sectional retinal images, and it is considered the standard of care for the assessment and treatment of various eye conditions.
“Unlike previously developed approaches, our algorithm requires OCT images from only a single pretreatment timepoint,” said Reza Rasti, first author of the paper and a postdoctoral scholar in Farsiu’s laboratory. “There’s no need for time-series OCT images, patient records or other metadata to predict therapy response.”
The new algorithm works by highlighting global structures in the OCT. At the same time, it also enhances local features from diseased regions. It searches for CNN-encoded features that can be correlated with anti-VEGF response.
The algorithm was tested with OCT images from 127 patients who had undergone treatment for diabetic macular edema with three consecutive injections of anti-VEGF agents. The algorithm then analyzed OCT images that were taken prior to the anti-VEGF injections, and the algorithm’s predictions were compared to OCT images taken after anti-VEGF therapy. This told researchers whether or not the therapy resulted in an improvement of the condition.
The algorithm was found to have an 87 percent accuracy rate for predicting those who would respond to treatment. It had an average precision and specificity of 85 percent and a sensitivity of 80 percent.
The researchers now want to confirm the findings and undertake a larger observational trial of patients who have yet to go through treatment.
- Researchers Improve Robotic Arm Used in Surgery
- DeepMind and Google Brain Aim Create Methods to Improve Efficiency of Reinforcement Learning
- Deep Learning Used to Find Disease-Related Genes
- AI “Maths Robot” Helps Manage Microclimates and Increase Berry Yield Predictions
- Computer Scientists Tackle Bias in AI