Article Text

PDF

Highlights from this issue
  1. Philip E M Smith1,
  2. Geraint N Fuller2
  1. 1 University Hospital of Wales, Cardiff, UK
  2. 2 Department of Neurology, Gloucester Royal Hospital, Gloucester, UK
  1. Correspondence to Dr Geraint N Fuller, Gloucester Royal Hospital, Gloucester GL1 3NN, UK; geraint.fuller{at}nhs.net

Statistics from Altmetric.com

Artificial intelligence promises to provide innovations for many aspects of life in the 21st century, from unbeatable chess players and self-driving cars to smart houses. We can anticipate it will improve healthcare. The exponential growth of computing power makes machine reporting of MRI imminent and attractive—surely a better way to evaluate scans in a patient with multiple sclerosis than laboriously comparing adjacent images slice by slice? EEG and histopathology reporting will surely follow. But how might artificial intelligence help in clinical diagnosis?

As the possibility of implementing some of these ideas comes closer, artificial intelligence itself comes under greater scrutiny. Purists distinguish machines that can learn from their experience—deep learning—from those that are driven by sophisticated algorithms. Deep learning might be the ultimate objective, but the algorithmic approach is much more attainable; most current examples—such as self-driving cars—are just algorithms on wheels, and only as good as the algorithms that run them.

In this edition of Practical Neurology, we have algorithms, deep learning and quite a lot between. Guidelines are algorithms in paper form and we appreciate those that are distilled into a single-page infogram. Two great examples in this edition are the Association of British Neurologists’ guidelines on periprocedural antithrombotic management for lumbar puncture (page 436)—tackling the knotty problem of cerebrospinal fluid sampling in people taking anticoagulants or antiplatelet agents—and an infogram summary of the recent consensus guidelines on idiopathic intracranial hypertension management from Alex Sinclair’s group (pages 485). Two to cut out and keep and pin on the wall in both the neurology and the acute medical unit.

For algorithm-run machines, the work must mainly go into algorithm design; once written and programmed, the machine just runs. But to get doctors or medical students to adopt, learn and use the algorithms is more challenging. Simulation training is one way to do this, and Clare Galtrey and colleagues (page 477) share their experiences in simulation training for acute neurology. David Nicholl, who reviewed this paper for us, has also made available some of his acute neurology simulation cases which can be found attached to the digital version of the paper.

Despite technological advances, much of what we do in neurology remains low tech and clinical: taking a history, examining and discussing illnesses and diseases with patients. Every so often patients bring new ideas into the consultation, especially topics discussed in social media. The media noise is often ahead of the evidence base, making discussion challenging. Two such issues are the use of cannabis for epilepsy and stem cells for treating multiple sclerosis. Rhys Thomas and Mark Cunningham (page 465) provide a practical overview of what we know and do not know about cannabis and epilepsy. Clare Rice and colleagues (page 472) provide neurologists and their patients with the background on the range of stem cell treatments and multiple sclerosis, and the questions that patients should ask of any potential treatment provider.

We have more conventional reviews: Amy Ross-Russell and colleagues provide a practical overview of the diagnosis and management of Lyme disease (page 455), not exactly an algorithm but all the essential background information that we need about this condition. Cingulate gyrus onset epilepsy may not be an entity on many neurologists’ radar, but Rob Powell and colleagues’ review (page 447) should make it easier for us to recognise this presentation in clinic.

We have an assortment of clinical case reports that allow us to learn from the experience of others—an opportunity for deep learning—with an explicit discussion of one colleague’s thought processes too in our clinicopathological case (page 505).

Our occasional reports of ‘Me and my neurological disease’ feature a very particular type of clinical case in which a neurologist is both a historian and a story-teller, detailing clinical phenomenology through personal experience rather than through the usual second-hand report. Eric Nieman, a retired and distinguished neurologist from St Mary’s Hospital in London, describes his distressing experience of Charles Bonnet syndrome (page 518), and Chris Kennard provides some background and a summary of the mechanism of this unusual and debilitating condition (page 434).

We hope you will enjoy Carphology, but before you do please consider your carbon footprint—Sui Wong allows us to calculate the size of our carbon neurological footsteps depending on the machines we use. Our machines may be rapidly changing, and more and more algorithms are shaping our clinical practice, but deep learning still seems an approach limited to humans. The human elements of clinical neurology—active and empathic listening, emotional intelligence, truly shared decision making—remain way beyond any man-made machine. Artificial intelligence and the machines may be coming, but our jobs seem safe for the moment….

View Abstract

Footnotes

  • Competing interests None declared.

  • Patient consent Not required.

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.