In December 2020, HTN featured artificial intelligence tools supporting everything from colonoscopies to strokes, and the diverse applications of AI in healthcare.
Part two of our ‘applications of AI in healthcare’ series looks specifically at AI in radiology; AI in psychiatry; and in the creation of new drugs and the interaction with existing drugs.
The latter topic being the most pertinent, particularly over the past year, where AI can be accredited with the rapid development and identification of existing drugs in the fight against COVID-19.
Starting with radiology, here are some of the critical ways in which AI is transforming and revolutionising healthcare in the NHS and beyond.
AI in Radiology
Creating a new, common digital radiology system
EMRAD, or ‘the East Midlands Radiology Consortium’, is a partnership of seven NHS trusts. The partnership is spread over 11 hospitals, which oversee the treatment of five million patients.
When launched in 2013, EMRAD set out to ‘create a new, common digital radiology system’ with the aim of implementing interoperable systems, and improving data sharing between multiple NHS sites.
According to NHSX, the EMRAD system has ‘set the national benchmark for a new model of clinical collaboration within radiology services.’
EMRAD formed a partnership with Faculty (more below), and Kheiron (featured in part one with their ‘Mia’ system) – to develop, test, and deploy AI systems to support the East Midlands’ breast cancer screening programme.
Where Kheiron focuses on the clinical applications of AI with Mia, Faculty focuses on the non-clinical operational applications of AI through the ‘Test Bed’ project which seeks to ‘apply AI tools and techniques to the operational and administrative aspects of the breast screening programme.’
This project is testing the ways AI can help to improve efficiencies and the effectiveness of the breast cancer service.
Faculty Platform
The specific role Faculty has been assigned in the project is to test the potential application of process optimisation tools and techniques, with the overall aim to make the best possible use of resources, notably radiologists’ time and expensive machinery, and also to reduce stress on the clinical and administrative workforce.
Faculty’s AI planning and scheduling tool provides the radiologist with an interactive dashboard, which is said to offer insights from historical data; there is also a demand and capacity forecasting tool to optimise service delivery and staff scheduling. The tool also has scheduling capability and a simulation tool to help understand the long-term impact of variations in capacity and demand.
The programme is also exploring the potentially of the AI to be able to provide early warning of increases in demand for assessment clinics, where service managers would be able to schedule extra clinics prior to a spike.
According to NHSX, the future evolution of the breast screening programme could open up the use of AI tools to allow clients ‘to be called for screening at a location that optimised their travel time and the available service capacity.’
Mirada – DLCExpert
Mirada’s DLCExpert, an auto-contouring tool, works as a ‘plug-in’ to any RT vendor systems and is said to be a ‘zero click’ solution. The tool uses AI Deep Learning, automated contouring and structure sets to save time in the targeting of radiation to cancer inflicted organs.
The tool is pre-packaged with RT AI trained structures for the clinician to choose from, such as head and neck, thorax, and breast and prostate.
Deep Learning essentially trains by mimicking human behaviour on hundreds of training examples; in this case, hundreds of manually contoured images of organs.
Mirada claims that their validation results for the DLCExpert system ‘delivers expert acceptance of OAR contours at a similar level as clinically drawn contours.’
AI in Psychiatry & Mental Health
University of Colorado BD
Since 2019, a team of researchers from the University of Colorado Boulder have been working on applying machine learning AI in psychiatry.
The team have developed a speech-based mobile app that can identify the status of a patient’s mental health.
Research Professor Peter Foltz, from the Institute of Cognitive Science, stated when talking to Health Europa that: “We are not in any way trying to replace clinicians, but we do believe we can create tools that will allow them to better monitor their patients.
“Language is a critical pathway to detecting patient mental states. Using mobile devices and AI, we are able to track patients daily and monitor these subtle changes.”
The App
The unnamed app is said to not yet be commercially available, although it has been presented with a working demonstration through a ‘patient’ interacting with it.
In the field of psychiatry, AI applications are still in the proof-of-concept phase, although evidence is rapidly mounting in favour of the use of chatbots which imitate human behaviour and conversation, and have been studied for use in the treatment of anxiety and depression.
The app is presented on a smartphone and begins with the chatbot asking the ‘patient’ “what time did you wake up this morning?”, with the patient responding vocally.
Other prompts see the ‘patient’ moving sliders to indicate mood: for example, a ‘sad’ slider is moved anywhere between ‘not sad’ and ‘very sad’, while some sliders, such as that of ‘sleep’, require a vocal follow-up response by the ‘patient’, who explains to the app the reasons why they didn’t sleep well.
The app also shows pictures to the ‘patient’ and asks the ‘patient’ to verbalise “what is happening in this picture?”. Other tasks ask the ‘patient’ to match a steady tapping noise by tapping their finger on the phone screen to the beat.
Further tasks include tapping a moving target, remembering numbers and other ordering activities, and fascinatingly asking the ‘patient’ to firstly listen to a story and then retell it verbally to the app.
AI applications within psychiatry have primarily been proposed by private corporations thus far, such as Facebook back in 2017, in an attempt to address the link between social media use and depression or anxiety leading to suicide rates in young people.
Novoic
Similar to the team from the University of Colorado, Novoic have developed a speech analytics platform in order to monitor brain health.
The platform uses machine learning and natural language processing to support clinicians in identifying and monitoring subtle changes in speech patterns, which occur when neurological disease is present.
According to Novoic, neurological diseases ‘affect the words we speak and how we speak them – sometimes decades before clinical symptoms present.’
Novoic has applied research in natural language processing and built ‘better’ pattern extractors. Scientists at Novoic analyse the prosodic, acoustic, and affective elements of raw speech signals, subsequently learning how these signals change when the brain’s health deteriorates.
Machine learning makes this learning possible by finding patterns in large datasets of speech ‘paired with biomarkers’ and ‘other health data.’
University of Huddersfield ADHD AI Diagnosis Project
Professor Grigoris Antoniou from the University of Huddersfield has used AI to speed up the process of ADHD diagnosis in adults.
He said: “There are long and growing waiting lists, as people wait to be diagnosed and treated, and this can result in adverse effects on their work, their social life and their family life.
“So, we set out to use AI to provide help with decisions. The idea is that the AI technology will be able to identify the clear-cut cases.
“In many cases, the data itself more or less tells us whether it is a ‘yes’ or a ‘no’ for further treatment.
“The technology is fully embedded in a clinical pathway which ensures there will always be a clinician who can over-ride what the AI says.”
Data that is collected routinely before ADHD diagnosis is analysed by an AI algorithm where there is a possibility of three results: ‘yes’ or ‘no’ for further treatment, or ‘unclear’ where the patient will then undergo further assessment.
Two types of AI technologies were harnessed for the project: machine learning and knowledge-based:
“One is machine learning-based. We took data from previous cases and trained a prediction model.
“The second method is knowledge-based. We worked with clinical experts and asked what their diagnosis would be if they are faced with this data. We then encoded this knowledge.”
Grow MedTech has provided backing for the project in order to explore the viability of commercialisation.
AI in Drug Creation & Interaction
AI has dramatically decreased the creation time of new drugs from what was traditionally several years of development, to under a single year. It was reported back in 2019 that Insilico Medicine created six novel inhibitors of the DDR1 gene (human gene) ‘implicated in fibrosis and other diseases’, in just 21 days. The system used by Insilico Medicine is known as ‘Generative Tensorial Reinforcement Learning’ (GENTRL) and tested a lead candidate whilst also showing positive results in mice, proving that using AI for drug creation is not only efficient, but also just as effective as traditional pathways.
Exscientia
British start-up Exscientia, in partnership with Sumitomo Dainippon Pharma (Japanese pharmaceutical firm) invented a molecule of the drug DSP-1181 used for treatment of OCD.
Using AI, the drug was developed in a single year and DSP-1181 was accepted for human trial.
Exscientia state that they ‘are the first to apply AI to small molecule drug discovery’ through drug target selection; AI is applied to a breadth of already published literature to acquire and analyse new drug knowledge.
Exscientia’s AI system ‘Centaur Chemist’ using the slogan ‘design, make, test’, analyses selected data based on drug target selection.
The AI process identifies and selects compounds to synthesise and test, cutting the process down from five to 1.5 years. The process is said to be exponentially more accurate in terms of compound identification, where 2,500 compounds would be identified over five years, cut to 500 viable compounds over 1.5 years.
