A Re-think of medical imaging: how can AI help backlog in the NHS post COVID

Artificial Intelligence (AI) has seen many advances in recent years. Many of those advances – some deployed more widely than others – found direct applications to the battle against COVID. This essay will analyze the applications of recent technological advances from the realm of Artificial Intelligence to the fight against the grievously impactful pandemic by suggesting their use in medical imaging modalities.

Personally, I first identified the value of Artificial Intelligence when I worked on a research project at the Weizmann AI Center in Israel. While my project dealt with simulating the variable resolution of human vision, AI can indeed be applied much more broadly and in the context of COVID-related medical imaging.

Unquestionably, images produced by X-rays, gamma cameras, ultrasound scanners, Magnetic Resonance Imaging, PET, thermography, and other devices yet to be invented are scanned by the skilled eye for diagnostic evidence resulting in treatment, prognosis and monitoring progress. All of these images will necessarily have information that is ignored by the trained eye as being irrelevant to what is being sought. Unquestionably, that unnecessary data slows down identification of those major diagnostic features that are being looked for. If Artificial Intelligence can be used to identify what is unnecessary, then diagnosis by all imaging modalities can be speeded up with an increase in throughput and earlier commencement of any treatment necessary.

Within the UK in particular, the NHS AI Lab – a team of AI specialists at the NHS – was swift to initiate the nation’s largest database of COVID-related medical images – covering X-Ray, CT and MRI images totaling about 40,000 as it stands.

This gave research universities across the nation the unique opportunity of training highly complex machine learning models for doing just that – saving medical specialists precious time and speeding up diagnosis. The University of Cambridge, for example, set out to build a medical tool for both detecting and tracking the progression of COVID from CT scans of the lungs. Many other research universities across the UK are working on similar COVID imaging projects, utilizing AI.

Such tools have demonstrated the ability, for example, to detect Ground-Glass Opacity (GGO) in the lungs within seconds – this is one of the primary signs signaling the development of potentially deadly COVID-relating breathing complications. This same task can take a trained radiologist many minutes.

In practice, the reliability of such tools has been demonstrated outside the UK – in China. Reportedly, such COVID imaging software can reach accuracies in diagnosing COVID ranging from 88% to 99%.

Over 34 hospitals around Wuhan, China have deployed such COVID-screening software into their clinical practice. The tool – developed by the Chinese startup InferVision – was used to screen more than 32,000 patients!

This is of clear benefit to overworked medical staff: both reducing the rate of transmission by minimizing the need for collecting PCR-test samples and reducing the average time taken to serve patients.

As it stands in the UK, such software is deemed not yet ready for clinical use. This is mainly due to machine learning models developed thus far having poor documentation and large biases.

The discrepancy between China’s widespread use of AI and UK’s more conservative one is likely due to differences in privacy policies in the two countries. In China, privacy laws are much more relaxed – making it easier to deploy new software rapidly and collect medical image data on a broader scale.

However, the NHS has launched an initiative that strongly suggests the institution’s desire to also intermingle AI and COVID imaging in the future. The NHS’s Moorfield Eye Hospital in London is already using AI to diagnose eye disorders with a staggering 94% accuracy – putting it on the same level as professionals with 20+ years of experience! Addenbrooke Hospital in Cambridge is also using AI to diagnose and segment prostate cancer from medical images! Both of those applications are speeding up the clearing of extensive patient queues formed in the past months due to the pandemic; thus, helping tightly stretched and overworked medical specialists. It is a matter of time before the UK’s safer and stricter requirements are met by the ever-evolving COVID imaging software.

AI-powered robotic assistants are another tool that can positively impact the imaging and tracking of COVID patients.

It is estimated that more than 50% of those infected with COVID develop a fever. It is of crucial importance for public safety to limit social contact with individuals suffering of this symptom and a commonly implemented methodology to achieve that goal is thermal imaging – usually done by staff at building entrances.

Thermal cameras and sensors are most commonly operated by humans, but this is quickly changing. There have been several attempts to produce AI-powered robots equipped with thermal cameras, having a capacity of as much as 100 patient screenings per minute!

There is no public data of such robots being used in the UK, but their efficacy has resulted in adoption of the technology by other countries. Hamad International Airport (Qatar), for example, has equipped staff with helmets displaying the estimated body temperature of passengers.

The use of such AI-powered hardware tools can be of great utility at public venues; firstly, reducing the number of human-to-human interactions and, secondly, allowing for more precise tracing of potential disease sufferers at a distance.

Both of those factors reduce the rate of disease transmission and help institutions with tightly stretched staff cope with the pandemic using more efficient tools.

Having said all that, the great utility of AI in medical imaging can only be achieved by aggregating large amounts of data – needed to train the AI models for autonomous diagnosis, tracking and thermal imaging of patients. Understandably, this raises a point of great concern to the public – privacy. In a world where user data is collected and sold to corporations, it is difficult to maintain faith in public institutions: that our even more sensitive medical data will be used with care, dignity and clear purpose.

However, with privacy measures in place, such as the ones utilized by the NHS in its AI efforts, innovation can clearly light the way to a better tomorrow – in medical imaging and beyond!