Ophthalmology and Machine Learning AI

The capacity of humanity has been seemingly endless throughout time, with ideas that reveal glimpses into the perfect world that could be achieved by the work of humans. The chemistry of life, the chemical building blocks for life, rarely make mistakes; with purine aligning with pyrimidine, with codons binding with their respective complementary anti-codon, and each atom cohering with the basic laws of chemistry. Our constituents are perfect, but oddly, we are not. Few, if any, can go throughout their lives without proliferated mistakes at regular intervals throughout life, mistakes are the result of an innumerable set of variables that can sway towards the side of success or failure, you have no control of them, but they will define your greatest and worst moments throughout your professional career. No career feels the extent of this effect in the same way as Medicine. Those who find their way into Medicine, and onto a healthcare profession, are nothing but academic elites. On top of lengthy, regular volunteering commitments, and diverse extra-curricular activities, they have to provide the highest grades to be within a chance of fulfilling their dreams. How they can such an excellent cohort then be so prone to mistakes? In fact, the 1991 New England Journal of Medicine, known for its Harvard Medical Practical study, revealed that 1% of admissions in healthcare suffered complications directly due to treatment, although it seems like a small figure, upwards of forty thousand patients die due to circumstances such as these, and it is no surprise that the NHS pays around £1.1 billion per year to pay for errors in healthcare – even earlier this month, headlines revealed that a woman sued a doctor for operating on the wrong eye, these ‘blunders’ in medicine been occurring at a fairly rate, but why is this? Looking at the 2016 NHS contact, which brought the most change in terms of working hours for NHS doctors, maximums of 72 hours per week (formerly 91), 13 hours for a single shift (formerly 14), 5 consecutive long (>10 hours) day shifts and 4 consecutive long night shifts under legislation, both of which were previously capped at 7. These are clearly depriving conditions for most doctors to be under, and the additional pressure and technicality of their profession severely undermines their efficacy in their jobs, particularly in the field of ophthalmology that we will consult today, where a great deal of care must be undertaken in the precise deciphering of difficult and intricate scans. Note, this cannot be taken as a criticism of these workers, but rather felicitations for their immense dedication for healthcare. However, what can be done to help these overburdened clinicians?

In this article, we will be exploring one solution. Machine Learning Artificial Intelligence. As a small measure of clarification, machine learning is a mean of data analysis that can infer patterns and trends within data, and can independently make observations within new data entries, with what it has learnt from the former data sets, in effect, a microcosm of the human mind. The current issues in Medicine stem from over-working current staff (or on the other hand, under-staffing). Given that according to the NHS Health Careers, it can take up to 14 years to become a trained surgeon, the element of time is severely hindering to the NHS.  You might think initially that I am proposing some concept where robots take over surgeries, but frankly that is not only ludicrous, but also expensive for the NHS, which is already considered underfunded by the general population. In addition to this, the human element in healthcare would be hindered, and would clearly propose a step backwards in healthcare as it has been clinically proven that the standard of human contact that a patient receives is critical towards their recovery. Hence. I put forward to you the concept of assisted healthcare; man and machine working side by side, in this case as hinted towards beforehand, we will look in detail at what artificial intelligence can achieve for ophthalmology, the science that deals with the diagnosis and treatment of eye conditions – with the implications for other healthcare fields to be explored further in coming articles.

At Moorfields Hospital, the leading provider of eye health services in the United Kingdom, consultant ophthalmologist Pearce Keane comments on the great numbers of false-positive referrals, which not only compromises the healthcare received by those who indeed suffer from some eye disorder, but also burdens those who don’t require such healthcare with unnecessary stress and wasted time – or even worse, irreparable damages that were simply negligent.  Roughly three years ago, Moorfields Eye Hospital NHS Foundation Trust and DeepMind Technologies, a large British artificial intelligence researching company, entered a collaborative five-year partnership to assess the potential impact of artificial healthcare in ophthalmology. According to the Moorfields NHS Trust site, the Machine Learning AI system is able to ‘recommend the correct referral decision for over 50 eye diseases with 94% accuracy, matching world-leading eye experts’. The progression of eye diseases, such as glaucoma and macular degeneration, can often lead to irreversible eye damage, with both contributing to the loss of eyesight (particularly in the elderly), and hence the implementation of artificial intelligence can be seen as a much-needed time-reducing measure of elephantine value for a vulnerable patient, as this will ultimately lead to reduced periods between initial scanning and diagnosis/treatment. With technology such as this, professionals could easily identify with patients require more immediate attention and hence realistically grant more severe patients a greater chance in being effectively treated, whilst not compromising the treatment of those with milder conditions and ensure that emergency cases towards much needed avenues such as LASIK, LASEK, and PRK methods. This technology has been accelerated by the advanced deep learning architecture behind the trans-formative AI system; since the Moorsfields NHS Trust has publicly released a huge bank of OCT (Optical Coherence Tomography) scan samples that spans well over a million unique data entries, it is a perfect data set provider for deep learning systems, and as such the neural network performance and cognition has been far more superior given the vast data sets it has access to, which is a major component that drives data-set optimized technologies such as deep learning ( deep learning systems have also seen improvements due to significant advances in computing, larger models, and  renowned algorithms produced by DeepMind Technology over the current 3 year span of the project). OCT scans are 3D images provide detailed images that take cross-sectional images of the retina (showing distinctive layers), and map the of your eye, they are complex and require expert analysis to draw valid conclusions that form the basis of diagnosis and treatment. According to DeepMind’s Co-Founder Mustafa Suleyman, the time taken to analyze these scans, combined with the sheer number of scans that are taken each day (over a thousand per day at Moorfields alone) can lead to great delays between scanning, diagnosis and treatment, regardless of the progression and seriousness of the eye condition, and this crucial time delay could be the difference in saving sight, or not. The image below shows an example of an OCT scan, this hopefully demonstrates the painstaking interpretative processes involved within clinical ophthalmology.

An OCT scan example, showing the distinct layers of retinal tissue

An OCT scan example, showing the distinct layers of retinal tissue

 The research project, over the last three years, has revolutionised the hospital’s workings, with the algorithm boasting world-leading statistics in the diagnostic accuracy of retinal disease. The DeepMind machine learning artificial intelligence system comprises two neural networks, the first analysing the optical coherence tomography scan to map the different eyes tissues and any initial features of disease it may detect, such as lesions or unusual proportions of fluid. The second network then analyses the mapping and observations of the first, and provides a diagnosis and referral recommendation Most notably, any observations are given with a percentage, allowing a clinician to measure the confidence of the system in its analysis, and hence the clinician has a far less burdening load of work, and is less likely to make errors. In this way, we have AI-assisted healthcare, the AI system doesn’t make the diagnosis, it simply outputs suggestions that can be critically time-saving, preventative of misdiagnosis and hence more effective. Artificial intelligence is still in early phases for healthcare, but research carriers on with greater sets of data, models, and computing, the healthcare industry could be transformed, not only for the patients, but also the staff that are currently working tirelessly in our current NHS.

 

As for the next phase, DeepMind and Moorfields are preparing the system for practical usage, with a series of clinical trials and regulatory appraisal in its path. Once this has been cleared, this is intended to be a free service for the 30 hospitals and clinics associated with the Moorfields NHS Trust for an initial period of five years, which could impact over 1.5 million patients – hence the immeasurable value in the increased speed and accuracy of clinical assessments. DeepMind also intends to revolutionise eye research globally, with a significant investment in publicising the system database for the world to use in the pursuit of optimal eye care. This has also found its way into nine separate studies by hospital researchers into a multitude of conditions and is currently owned by Moorfields so that they can use it in their future endeavours in clinical research. Collaborative efforts by researchers and technicians will hopefully improve the quality healthcare, its costs, and the strain on our NHS workers – and most importantly grant humanity a fighting chance in caring for a population that is becoming ever more strenuous to support.

 

A very big thank you towards Moorfields NHS Trust and DeepMind Technology for their pioneering efforts and publicity of their works, especially for the honourable persons mentioned in the article, namely Pearce Keane and Mustafa Suleyman, and indeed all who work to strengthen the case of humanity in survival.