Imaging experts have developed an artificial intelligence tool that can help predict delays in radiology turnaround times during nights and weekends, key info for quality improvement efforts.

University of California, San Francisco, researchers created the machine learning model utilizing more than 15,000 CT scans. Testing the tool out, they produced solid early results predicting delays greater than 245 minutes (area under the curve of 0.85) and interpretation setbacks of 57 minutes or longer (AUC 0.71).

“As delays in radiology are an important measure of patient safety and hospital efficiency, having the ability to predict such potential delays has important benefits,” Jae Ho Sohn, MD, a cardiothoracic radiology fellow at UCSF, and colleagues wrote June 27 in Academic Radiology. “Furthermore, prediction of delays in radiology can improve the referrer and radiologist relationship and help clinicians to prepare alternative options in case a delay is expected.”

For their study, San Francisco scientists gathered retrospective CT data from two hospitals within the same organization, logged between 2018 and 2019. The original set included nearly 30,000 inpatient and emergency cases, whittled down to about half that for their analysis. They tracked order and scan time, first communication by radiologist, free-text indications and more.

Sohn et al. used 85% of this data to train their ensemble machine learning model and the remaining 15% for testing. AI was tasked with predicting delays between when the exam was ordered to the first communication, along with delays between scan completion and interpretation.

The team discovered that CT study description, time of day and year in training were much more predictive features than body part imaged, inpatient status and hospital campus. In addition, some protocols were associated with delayed turnaround time because of the complexity of cases, including CT of the neck with contrast, were associated with delayed turnaround times

Future studies could potentially add additional variables, such as hospital and ED patient census, number of providers, transportation and average technology operating time. Sohn and colleagues see their work as an important starting point for quality improvement projects.

“Given the complexity of real-world radiology workflow, no algorithm can make perfect predictions on which cases will be delayed. However, attaining a reasonable prediction of such cases can be relevant,” the authors advised.

The post Artificial intelligence predicts delayed radiology turnaround times during nights and weekends appeared first on Artificial Intelligence.

Reston, VA-An advanced artificial intelligence technique known as deep learning can predict major adverse cardiac events more accurately than current standard imaging protocols, according to research presented at the Society of Nuclear Medicine and Molecular Imaging 2021 Annual Meeting. Utilizing data from a registry of more than 20,000 patients, researchers developed a novel deep learning network that has the potential to provide patients with an individualized prediction of their annualized risk for adverse events such as heart attack or death.

Deep learning is a subset of artificial intelligence that mimics the workings of the human brain to process data. Deep learning algorithms use multiple layers of “neurons,” or non-linear processing units, to learn representations and identify latent features relevant to a specific task, making sense of multiple types of data. It can be used for tasks such as predicting cardiovascular disease and segmenting lungs, among others.

The study utilized information from the largest available multicenter SPECT dataset, the “REgistry of Fast myocardial perfusion Imaging with NExt generation SPECT” (REFINE SPECT), with 20,401 patients. All patients in the registry underwent SPECT MPI, and a deep learning network was used to score them on how likely they were to experience a major adverse cardiac event during the follow-up period. Subjects were followed for an average of 4.7 years.

The deep learning network highlighted regions of the heart that were associated with risk of major adverse cardiac events and provided a risk score in less than one second during testing. Patients with the highest deep learning scores had an annual major adverse cardiac event rate of 9.7 percent, a 10.2-fold increased risk compared to patients with the lowest scores.

“These findings show that artificial intelligence could be incorporated in standard clinical workstations to assist physicians in accurate and fast risk assessment of patients undergoing SPECT MPI scans,” said Ananya Singh, MS, a research software engineer in the S­­­lomka Lab at Cedars-Sinai Medical Center in Los Angeles, California. “This work signifies the potential advantage of incorporating artificial intelligence techniques in standard imaging protocols to assist readers with risk stratification.”

Abstract 50. “­­­­Improved risk assessment of myocardial SPECT using deep learning: report from REFINE SPECT registry,” Ananya Singh, Yuka Otaki, Paul Kavanagh, Serge Van Kriekinge, Wei Chih-Chun, Tejas Parekh, Joanna Liang, Damini Dey, Daniel Berman and Piotr Slomka, Department of Imaging, Cedars-Sinai Medical Center, Los Angeles, California; Robert Miller, Department of Cardiac Sciences, University of Calgary, Calgary, Alberta, Canada, and Department of Imaging, Cedars-Sinai Medical Center, Los Angeles, California; Tali Sharir, Department of Nuclear Cardiology, Assuta Medical Centers, Tel Aviv, and Ben Gurion University of the Negev, Beer Sheba, Israel; Andrew Einstein, Division of Cardiology, Department of Medicine and Department of Radiology, Columbia University, Irving Medical Center and New York-Presbyterian Hospital, New York, New York; Mathews Fish, Oregon Heart and Vascular Institute, Sacred Heart Medical Center, Springfield, Oregon; Terrence Ruddy, Division of Cardiology, University of Ottawa Heart Institute, Ottawa, Ontario, Canada; Philipp Kaufmann, Department of Nuclear Medicine, Cardiac Imaging, University Hospital Zurich, Zurich, Switzerland; Albert Sinusas and Edward Miller, Section of Cardiovascular Medicine, Department of Internal Medicine, Yale University School of Medicine New Haven, Connecticut; Timothy Bateman, Department of Imaging, Cardiovascular Imaging Technologies LLC, Kansas City, Missouri; Sharmila Dorbala and Marcelo Di Carli, Department of Radiology, Division of Nuclear Medicine and Molecular Imaging, Brigham and Women’s Hospital, Boston, Massachusetts.

The post Deep Learning with SPECT Accurately Predicts Major Adverse Cardiac Events appeared first on Artificial Intelligence.

Claiming it has achieved a first in drug discovery, Insilico Medicine says it will begin preclinical trials with a novel compound discovered by artificial intelligence, aimed at a novel target discovered by artificial intelligence. The company isn’t revealing the compound or the target but says the small molecule compares favorably against an existing drug for treating idiopathic pulmonary fibrosis (IPF) in mice and in vitro.

The company also emphasizes the efficiency of their process, claiming that it took 18 months and about $2.6 million to identify the target and the candidate and reach the point of starting preclinical trials. It is marketing the AI-based software packages used to find the target and the molecule, a product several large pharmaceutical companies are now testing.

Insilico Medicine attracted attention and some skepticism in 2019, when it announced it had used AI to sift through databases and identify a molecule predicted to inhibit discoidin domain receptor 1 (DDR1), a kinase linked to fibrosis (Nat. Biotech. 2019, DOI: 10.1038/s41587-019-0224-x). Pulmonary fibrosis is scarring and thickening of lung tissue that makes breathing harder. At the time some experts praised Insilico Medicine for its speed (the effort took a claimed 46 days). Others questioned the value of the feat. Some pointed out it was hard to know if a human could have found the molecule as fast or faster. Ingo Hartung, director of medicinal chemistry at Merck KGaA—which announced in November it would start using Insilico Medicine’s compound discovery AI—described it as the algorithm doing an easy job well.

In a briefing for reporters on Monday, Insilico Medicine CEO Alex Zhavoronkov agreed that the DDR1 target was an easy one. The company said it is now the first to use AI to find a totally new drug target and to propose a previously unknown molecule effective against that target.

Zhavoronkov described using two separate software packages to do so. The first, called PandaOmics, used AI and informatics tools to identify biomarkers of lung fibrosis, as well as text-mining AI to search for other clues about the target and to try to establish its novelty. Insilico Medicine described some of the elements of the software in 2016 (Nat. Commun. 2016, DOI: 10.1038/ncomms13427). The software predicted 20 novel targets, from which company scientists and software selected one they thought was most promising.

Having chosen the target, the company used software it sells as Chemistry42 to predict new molecules that would be effective against that target. Zhavoronkov said the system uses AI models including recurrent neural networks, genetic algorithms, and generative adversarial networks—including some described in 2019—to look for binding pockets in crystal structures and propose molecules that would fit them. Chemistry42 suggested 50 compounds, and the company evaluated those plus 30 known molecules to select their candidate.

Zhavoronkov said at the briefing that experiments so far showed their candidate could be effective at treating fibrosis and that it passed other tests for potential pharmaceuticals like toxicity and metabolic availability studies. He showed reporters data comparing the compound to nintedanib, a tyrosine kinase inhibitor currently prescribed to treat IPF. He said Insilico Medicine’s candidate operates by a different mechanism of action than nintedanib and in lab tests was more effective at lower doses for blocking the cellular changes associated with fibrosis. Although this work is in lung fibrosis, Zhavoronkov says the biotech firm has data suggesting the candidate is also effective in addressing liver, kidney, and skin fibrosis.

Zhavoronkov also emphasized how quickly and cheaply the company reached the doorstep of preclinical trials. He pointed to a 2010 study that estimated a typical drug’s development takes 10 years and $2 billion (Nat. Rev. Drug Discovery 2010, DOI: 10.1038/nrd3078).

The comparison is far from perfect; that study did not include target identification but did include clinical trials, which, as medicinal chemist Derek Lowe points out, account for the largest share of drug development costs.

Lowe suspects Insilico’s drug candidate is another kinase inhibitor like nintedanib, which itself has limited efficacy. That approved drug only treats symptoms of fibrosis and does not improve survival in people with IPF. He wonders what advantages Insilico Medicine expects its candidate will show. Big pharmaceutical companies aren’t waiting to see how this candidate pans out before investing in Insilico Medicine’s products. Zhavoronkov said five companies, including Pfizer and Syngenta, are using its software. Hartung, who did not see the information available to reporters, says he thinks AI shows potential to make drug discovery more efficient, and he’s looking forward to Merck’s testing of the Chemistry42 software, as well as other companies’ products.

The post Artificial intelligence predicts a target for treating fibrosis and finds a compound to do it appeared first on Artificial Intelligence.

A team of scientists at the UK’s Liverpool University has used artificial intelligence (AI) to work out where the next novel coronavirus could emerge.

KEY HIGHLIGHTS

• The COVID-19 pandemic was the first such massive and natural calamity to strike mankind in almost a century.

• Mankind had just not provided for such an eventuality and was caught off guard on almost all counts of preparedness.

• With climate change being real and threat of pandemics looming large, it would certainly help to know if a disease is going to acquire pandemic proportions.

In a rapidly advancing globalisation that has turned the entire Earth into one huge village, speedy connectivity and communication also ensured a rapid advance of the COVID-19 pandemic that began with a strain of the novel coronavirus that first emerged in Wuhan, China in late 2019. Now, as per a science paper published in Nature Communications, “The spread of influenza can be modelled and forecast using a machine-learning-based analysis of anonymized mobile phone data. The mobility map, presented in Nature Communications this week, is shown to accurately forecast the spread of influenza in New York City and Australia.”

The year 2020 dawned with the world bracing to handle a possible crisis and by the end of the year, global deaths reached nearly 2 million.

To cut the long story short, mankind has now been through so much in terms of mental agony, pain, loss, death, long-lasting illnesses and economic downslide – all on account of this pandemic – despite rapid advances in science – that it has begun to dread the prediction by environmentalists and scientists that we have just entered a pandemic era and more such pandemics are likely to come.

Predicting the onset of a Pandemic:
According to a report in the BBC, a team of scientists has used artificial intelligence (AI) to work out where the next novel coronavirus could emerge.

The researchers are reportedly putting to use a combination of learnings from fundamental biology and tools pertaining to machine learning.

This is not mere conjecture and the scientists are taking ahead of what they have gained from similar experiments in the past. Their computer algorithm predicted many more potential hosts of new virus strains that have previously been detected. The findings have been published in the journal Nature Communications. 

According to this report in Nature Communications, the spread of viral diseases through a population is dependent on interactions between infected people and uninfected people. The Building-models that predict how the diseases will spread across a city or country currently make use of data that are sparse and imprecise, such as commuter surveys or internet search data.

Dr Marcus Blagrove, a virologist from the University of Liverpool, UK, who was involved in the study, emphasises the need to know where the next coronavirus might come from.

“One way they’re generated is through recombination between two existing coronaviruses – so two viruses infect the same cell and they recombine into a ‘daughter’ virus that would be an entirely new strain.”

Scientists say that to get the prediction algorithm right, the first step was to look for species that were able to harbour several viruses at once. Lead researcher Dr Maya Wardeh, who is also from the University of Liverpool, successfully deployed existing biological knowledge to teach the algorithm to search for patterns that made this more likely to happen.

This step concluded that many more mammals were potential hosts for new coronaviruses than previous surveillance work – screening animals for viruses – had shown.

The post Liverpool scientists deploy Artificial Intelligence to develop model that predicts the next pandemic appeared first on Artificial Intelligence.

June 12, 2019 – Using deep learning technology, researchers from Massachusetts Institute of Technology (MIT) and Massachusetts General Hospital (MGH) were able to predict women’s future risk of breast cancer development more accurately than when they used traditional methods, according to a study published in Radiology.

Current models use factors like genetics and family history to predict risk, but these tools often fall short. Breast density is an independent risk factor for predicting breast cancer risk, but it’s based on subjective assessment that can vary among radiologists.

Researchers developed a deep learning model that could standardize and automate breast density measurements.

“There’s much more information in a mammogram than just the four categories of breast density,” said study lead author Adam Yala, PhD candidate at MIT in Cambridge, Mass. “By using the deep learning model, we learn subtle cues that are indicative of future cancer.”

The team compared three different risk assessment models. The first used traditional risk factors, and the second used deep learning that evaluated the mammogram alone. The third was a hybrid method that used both the mammogram and traditional risk factors into the deep learning model.

Researchers trained and tested the models on nearly 90,000 screening mammograms from about 40,000 women and found that both deep learning models performed with greater accuracy than the traditional model.

When using the deep learning models to predict women’s risk based on breast density, the team found that patients with non-dense breasts and model-assessed high risk had 3.9 times the cancer incidence of patients with dense breasts and model-assessed low risk. These advantages held across different subgroups of women.

“Unlike traditional models, our deep learning model performs equally well across diverse races, ages and family histories,” said Regina Barzilay, PhD, an AI expert and professor at MIT. “Until now, African-American women were at a distinct disadvantage in having accurate risk assessment of future breast cancer. Our AI model has changed that.”

At MGH, clinicians are already using artificial intelligence to assist with breast density measurements. Researchers are tracking its performance in the clinic and working to refine how they communicate risk information to women and their primary care physicians.

“A missing element to support more effective, more personalized screening programs has been risk assessment tools that are easy to implement and that work across the full diversity of women whom we serve,” said Constance Lehman, MD, PhD, chief of breast imaging at MGH and professor of radiology at Harvard Medical School.

“We are thrilled with our results and eager to work closely with our health care systems, our providers and, most importantly, our patients to incorporate this discovery into improved outcomes for all women.”

Deep learning has proven itself to be a reliable support tool for cancer care. In 2018, a team at Google developed a deep learning tool that could detect metastasized breast cancer with 99 percent accuracy.

Researchers at Case Western Reserve University also built a model that achieved 100 percent accuracy when identifying invasive forms of breast cancer in pathology images.

“If the network can tell which patients have cancer and which do not, this technology can serve as triage for the pathologist, freeing their time to concentrate on the cancer patients,” Anant Madabushi, a biomedical engineering professor at Case Western Reserve and co-author of the study, said at the time.

“To put this in perspective, the machine could do the analysis during ‘off hours,’ possibly running the analysis during the night and providing the results ready for review by the pathologist when she/he were to come into the office in the morning.”

The research from MGH and MIT builds on these efforts, and further shows the potential for deep learning to transform cancer care and diagnosis.

“There’s a very large amount of information in a full-resolution mammogram that breast cancer risk models have not been able to use until recently,” Yala said. “Using deep learning, we can learn to leverage that information directly from the data and create models that are significantly more accurate across diverse populations.”

The post Deep Learning Predicts Women’s Future Risk of Breast Cancer appeared first on Artificial Intelligence.