In part three of our applications of AI in healthcare series, we take a deeper look into how AI has exponentially enhanced the area of ophthalmology through university research and private health technology suppliers.
Also featuring is this edition, is the use of AI in skin disease, in visual neuroscience, as well as treatment for lower limb wounds and sepsis.
If you missed part one and two of the series, you can catch up on them here:
- From AI to support colonoscopies to strokes: the diverse applications of AI in healthcare
- From disease detection to drug creation: the diverse applications of AI in healthcare
Here we cover even more ways that AI is being adopted across health and care:
Melanoma UK & SkinVision
Melanoma UK and SkinVision have partnered to support patient skin checks and early detection of skin cancers.
Its mobile application is certified by the British Standards Institute and uses AI to compare a user’s skin spots with millions of images of known skin cancers to provide a risk score. The app has demonstrated a sensitivity rate of 95.1% and specificity rate of 78.3% in the most recent peer reviewed clinical trial.
Using a phone camera to take a photo the image runs through a risk assessment algorithm, and is then processed via a Convolutional Neural Network (CNN) – a class of neural networks that are particularly suitable for classification tasks on image data.
The app provides a risk assessment to label the level of risk for the specific skin spot. Based on the risk class – low, low (with symptoms reported by the user) and high risk – the algorithm provides a recommendation on the next steps to take.
University of Dundee & NHS Tayside – Deep Learning for skin disease triage
A £150,000 grant was awarded to the collaboration between the University of Dundee and NHS Tayside back in September 2020 by Secretary of State for Health and Social Care, Matt Hancock.
The grant was to fund the ‘deep learning for effective triaging of skin disease in the NHS’ project which will allow researchers to develop an AI system to accurately distinguish between benign lesions and cancers.
Project leader Professor Stephen McKenna of the university’s computing department said: “Success in this area will be gradual, starting with goals such as clinical decision support for the most common benign lesions.
“Skin disease naturally lends itself to automated image analysis. Lesions can be photographed easily and then analysed with the help of deep learning technology.”
The technology is said to deliver ‘many benefits’ including patient reassurance about benign lesions, immediate education for GPs from the system itself, and fewer patients needing to attend hospital appointments.
University of Oxford – OxNNet Toolkit for foetal growth restriction (FGR)
The OxNNet Toolkit has received funding from NIHR to develop an AI ultrasound tool to screen for foetal growth restriction in the first trimester.
The system uses AI to automatically recognise the placenta in a 3D ultrasound scan and then measure it to determine if it is small or has a poor blood supply. If either of these conditions are detected, it can be determined that the baby has an increased risk of FGR.
Discovering FGR early in pregnancy means the mother can be monitored more intensively.
Professor Sally Collins from Oxford University’s Nuffield Department of Women’s and Reproductive Health said: “Fetal growth restriction is the leading cause of stillbirth which devastates eight families a day in the UK.
“This grant will enable us to further develop our OxNNet Toolkit, which aims to predict which babies will not grow well, a condition called foetal growth restriction, as early as 11-13 weeks into the pregnancy.
“If our OxNNet Toolkit proves to be a reliable early screening test, the holy grail of a treatment to prevent foetal growth restriction from developing becomes a real possibility.”
Newcastle University & Newcastle NHS Foundation Trust – Octahedron Project
Because Alzheimer’s disease and Parkinson’s disease affects the retina, the Octahedron Project analyses commonly conducted Optical Coherence Tomography (OCT) scans (performed at high street opticians), to capture signs of neurological disease.
Anya Hurlbert, Professor of Visual Neuroscience at Newcastle University who is leading the project explains: “The aim of the project is to use NHS data to teach computers how to detect early signs of neurological disease via retinal imaging.
“Ultimately, the project will help to catch those at risk earlier, before other symptoms develop.”
Like with other AI systems, Octahedron is trained on vast quantities of previously scanned images in order to develop specific detection algorithms.
Like with the University of Dundee and NHS Tayside, and the University of Oxford OxNNet Project, the Octahedron Project is also one of the recipients of government funding.
University of Liverpool & the Royal Liverpool University Hospital – AI & OCTA
Wet age-related macular degeneration, or wAMD, is where abnormal new blood vessels form in the retina of the eye, causing bleeding and scarring.
Optical coherence tomography angiography, or OCTA, has been developed as a rapid, non-invasive imaging modality which provides detailed images of the retinal vasculature removing the risks associated with current diagnosis methods.
OCTA is now increasingly being used in the diagnosis for wAMD, where OCTA and AI will eventually combine to potentially diagnose the main subtypes of wAMD.
Researchers at the University of Liverpool are about to begin training AI algorithms such as convolutional neural networks to classify wAMD into its subtypes.
Researchers aim to integrate machine learning with computational simulations, known as ‘in-silico simulations’, looking at blood flow and drug transport in the retina. It is said that most research currently uses either ML to learn from data OR in-silico simulation models to simulate physiology; very little work has looked at the amalgamation of both methods.
The research is to be undertaken in collaboration with the Royal Liverpool University Hospital’s ophthalmic team.
Macusoft Ltd – MacuSense
MacuSense is an AI clinical decision support system for the management of sight-threatening macular disease. It integrates with existing clinical systems to improve this specific patient pathway using the MacuSense product to analyse OCT scans (see Octahedron above) from high street opticians.
The cloud-based service is said to be ‘scalable’ and can be ‘delivered across sites for real-time results at the point-of-care’.
Dem Dx – AI clinical reasoning platform
Dem Dx have developed an AI clinical reasoning platform to help frontline healthcare professionals diagnose, order investigations, and make care recommendations at the first point of contact prior to a patient seeing a clinician.
The AI clinical reasoning platform has trained on a database developed by over 200 world-leading doctors and is educated by clinicians in the field.
Dem Dx state that their platform is 85% accurate at first diagnosis, ‘22% better than Harvard’s junior doctors’.
The digital platform covers 14 speciality pathways from gastroenterology to dermatology, and over 1,400 diagnoses. The clinician using the platform can click through a multitude of patient symptoms to arrive at final diagnosis.
The database has media built in for a more visual view of a condition and national guidelines are readily available also.
Nine Health Global Ltd – Woubot
Nine Health Global are developing an AI system to treat chronic lower limb wounds.
‘Woubot’ aims to be a precognitive tool for community and wound clinics, where it will first focus on leg ulcers and provide recommendations from ‘several thousand’ possible treatment combinations.
According to Nine Health Global, 40% of chronic leg wounds will not fully heal often leading to amputation and even death.
Woubot therefore uses machine learning algorithms to sift through millions of data items already processed onto NHS cloud servers to create bespoke treatment plans for patients where treatment is often administered in the community and clinics.
Nine Health Global claim that currently there are no other automated predictive or preventative software solutions to manage treatment of chronic lower limb wounds.
Imperial College London – AI for Healthcare (sepsis treatment)
Imperial College researchers have developed an AI system to help treat patients with sepsis.
The AI system is able to analyses patient data such as blood pressure and heart rate and subsequently decide the most effective treatment strategy. Researchers found that when a clinician’s treatment decision matched the AI system’s treatment recommendation, patients had a better chance of survival.
Professor Anthony Gordon, senior author from the Department of Surgery & Cancer at Imperial explained: “We know that most patients with sepsis need fluid drips and in more severe cases also need vasopressors to maintain blood pressure and blood flow.
“There is still much debate amongst clinicians about how much fluid to give and when to start vasopressors.
“There are clinical guidelines but they provide general advice. The AI Clinician is able to learn what is the best option for each individual patient at that moment in time.”
