AI Improves Heart Disease Diagnosis

Researchers from the University of Oxford are using artificial intelligence (AI) to improve diagnostic accuracy for heart disease. The team hope to roll out the system across the NHS later this year, helping to improve patient outcomes and saving millions is misdiagnoses. The research, led by Prof Paul Leeson and RDM DPhil student Ross Upton (Cardiovascular Clinical Research Facility), took place in the Oxford University Hospitals Foundation Trust and is the basis of spin-out company Ultromics.


Thousands of people every year have an echocardiogram – a type of heart scan – after visiting hospital suffering with chest pain. Clinicians currently assess these scans by eye, taking into account many features that could indicate whether someone has heart disease and if they are likely to go on to have a heart attack. But even the most well trained cardiologist can misdiagnose patients. Currently, 1 in 5 scans are misdiagnosed each year – the equivalent to 12,000 patients. This means that people are either not being treated to prevent a heart attack, or they are undergoing unnecessary operations to stave off a heart attack they won’t have.

The new system uses machine learning – a form of artificial intelligence – to tap into the rich information provided in an echocardiogram. Using the new system, AI can detect 80,000 subtle changes inviable to the naked eye, improving the accuracy of diagnosis to 90%. The machine learning system was trained using scans from previous patients, alongside data about whether they went on to have a heart attack. The team hope that the improved diagnostic accuracy will not only improve patient care and outcomes, but save the NHS £300million a year in avoidable operations and treatment.  So far the system has been trialled in six cardiology units in the UK. Further implementation of the technology is now being led by Ultromics – a spin-out company co-founded by Ross Upton and Paul Leeson (Cardiovascular Clinical Research Facility). The software will be made available for free throughout the NHS later this year.


A Single Drop Of Blood To Test Agressive Prostate Cancer

A new diagnostic developed by Alberta scientists will allow men to bypass painful biopsies to test for aggressive prostate cancer. The test incorporates a unique nanotechnology platform to make the diagnostic using only a single drop of blood, and is significantly more accurate than current screening methods.

The Extracellular Vesicle Fingerprint Predictive Score (EV-FPS) test uses machine learning to combine information from millions of cancer cell nanoparticles in the blood to recognize the unique fingerprint of aggressive cancer. The diagnostic, developed by members of the Alberta Prostate Cancer Research Initiative (APCaRI), was evaluated in a group of 377 Albertan men who were referred to their urologist with suspected prostate cancer. It was found that EV-FPS correctly identified men with aggressive prostate cancer 40 percent more accurately than the most common test—Prostate-Specific Antigen (PSA) blood test—in wide use today.

Higher sensitivity means that our test will miss fewer aggressive cancers,” said John Lewis, the Alberta Cancer Foundation‘s Frank and Carla Sojonky Chair of Prostate Cancer Research at the University of Alberta. “For this kind of test you want the sensitivity to be as high as possible because you don’t want to miss a single cancer that should be treated.”

According to the team, current tests such as the PSA and digital rectal exam (DRE) often lead to unneeded biopsies. Lewis says more than 50 per cent of men who undergo biopsy do not have prostate cancer, yet suffer the pain and side effects of the procedure such as infection or sepsis. Less than 20 per cent of men who receive a are diagnosed with the aggressive form of prostate cancer that could most benefit from treatment.

It’s estimated that successful implementation of the EV-FPS test could eventually eliminate up to 600-thousand unnecessary biopsies, 24-thousand hospitalizations and up to 50 per cent of unnecessary treatments for prostate each year in North America alone. Beyond cost savings to the health care system, the researchers say the diagnostic test will have a dramatic impact on the health care experience and quality of life for men and their families.

Compared to elevated total PSA alone, the EV-FPS test can more accurately predict the result of prostate biopsy in previously unscreened men,” said Adrian Fairey, urologist at the Northern Alberta Urology Centre and member of APCaRI. “This information can be used by clinicians to determine which men should be advised to undergo immediate prostate biopsy and which men should be advised to defer and continue screening.”


How To Fine-Tune NanoFabrication

Daniel Packwood, Junior Associate Professor at Kyoto University’s Institute for Integrated Cell-Material Sciences (iCeMS), is improving methods for constructing tiny “nanomaterials” using a “bottom-up” approach called “molecular self-assembly”. Using this method, molecules are chosen according to their ability to spontaneously interact and combine to form shapes with specific functions. In the future, this method may be used to produce tiny wires with diameters 1/100,000th that of a piece of hair, or tiny electrical circuits that can fit on the tip of a needle.


Molecular self-assembly is a spontaneous process that cannot be controlled directly by laboratory equipment, so it must be controlled indirectly. This is done by carefully choosing the direction of the intermolecular interactions, known as “chemical control”, and carefully choosing the temperature at which these interactions happen, known as “entropic control”. Researchers know that when entropic control is very weak, for example, molecules are under chemical control and assemble in the direction of the free sites available for molecule-to-molecule interaction. On the other hand, self-assembly does not occur when entropic control is much stronger than the chemical control, and the molecules remain randomly dispersed.

Packwood teamed up with colleagues in Japan and the U.S. to develop a computational method that allows them to simulate molecular self-assembly on metal surfaces while separating the effects of chemical and entropic controls. This new computational method makes use of artificial intelligence to simulate how molecules behave when placed on a metal surface. Specifically, a “machine learning” technique is used to analyse a database of intermolecular interactions. This machine learning technique builds a model that encodes the information contained in the database, and in turn this model can predict the outcome of the molecular self-assembly process with high accuracy.