Artificial Intelligence, or AI, is increasingly shaping the world around us. We now live in a digital age in which AI, far from being the domain of the tech giants, is such a regular feature of our everyday lives that it’s becoming difficult to find examples of areas in which it’s not used. From face recognition software on our mobile phones to informing internet search engines; social media algorithms; smart home devices; email spam filters; natural language processing (predictive text) and even banking (from fraud detection systems to verification of transactions). And at the end of a long day, most of us probably “Netflix and chill” to the recommendations of the streaming giant, courtesy of – you guessed it – their own AI algorithms. 

AI is everywhere, and it has the potential to shape – and drastically improve – the future. One of the fields which has already been significantly impacted by the introduction of AI and machine learning is the healthcare sector. Recent improvements in computing power (enabling the analysis of vast data sets), combined with the healthcare sector’s abundance of available digital data (in the form of electronic patient records, in addition to the digitisation of departments such as radiology) has led to an explosion in the use of AI-based technologies. Never before have the possibilities of ‘big data’ been quite so big.

The potential of AI to improve healthcare outcomes should not be underestimated: governments and healthcare systems throughout the world are already investing heavily in the development of AI, which is already being put to use in a variety of ways. Not only can AI mimic – and often outperform – human judgement when making medical diagnoses, it’s also capable of processing massive quantities of available healthcare data, opening the door to new opportunities in terms of medical image analysis, data mining, predictive modelling and a raft of other complex tasks.

Here are just a few of the ways in which AI is being used to improve healthcare services:

Radiology

AI is already making a significant contribution to radiology and image analysis. Machine-learning AI’s are being used to build and test extensive databases and image banks – comprising data from hundreds of thousands of patients – which are often outperforming humans in terms of diagnostic accuracy. AI can not only provide radiologists with valuable, evidence-based second opinions to support their professional diagnoses, but it can also capture information from existing data sources which the human eye may otherwise be unable to detect. 

Cardiology

Developments in AI are assisting cardiologists in numerous ways: supporting diagnosis and treatment planning; organising data into usable frameworks to facilitate wide-scale collaboration; identifying patterns in image analysis; and assisting in making reliable predictions of future health outcomes. Cardiologists now have access to AI-based diagnostic tools to support – and inform – clinical decision-making, relieving clinicians’ heavy workloads whilst significantly improving accuracy. 

Patient monitoring and disease prevention

The increased use of wearable devices (such as smartwatches) and fitness apps in recent years has not only yielded a massive increase in the amount of available health data, it has also provided patients with more autonomy over their own health outcomes. Remote monitoring has enabled healthcare providers to support patients in the community more effectively, by issuing medication alerts, providing effective monitoring of chronic diseases such as diabetes, and enabling patients to report symptoms outside of the healthcare setting. At the same time, AI-enabled remote monitoring can support physicians with disease prevention by providing an early warning system which can predict future health outcomes (e.g. blood pressure metrics), improve patient outcomes and inform long-term prognoses. Remote monitoring is also playing an important role in helping to promote health equity, by improving access to healthcare services in disadvantaged areas.

Managing demand and capacity

AI and machine learning models are already being used to help healthcare providers manage demand and capacity, forecast patient flow and optimise staff scheduling in primary care facilities as well as in the community. Interoperable AI systems also improve communication between stakeholders, enabling a more joined-up, collaborative approach across services and resulting in a more efficient, cost-effective and streamlined service which benefits both healthcare providers and patients.

Disease forecasting

AI has demonstrated an ability to accurately forecast the likelihood of an individual patient developing certain diseases – such as cardiovascular disease, renal failure, pneumonia, hypertension, liver cancer, diabetes, orthopaedic surgery and stroke – in the future, based on an analysis of risk factors, physical markers, presenting symptoms, clinical pathology, medical and family history. It would be almost impossible for an attending physician to read, comprehend and then act upon the wealth of medical information accumulated by an individual patient during a lifetime of medical care, however an AI is able to generate these clinical insights with relative ease, quickly, and with a high degree of accuracy.

Advances in research and treatment

Through the compilation of thousands – even millions – of electronic health records, AI has the potential to significantly advance medical research and treatment, optimising outcomes for patients and reducing costs for healthcare providers.

Considerations

As these examples demonstrate, AI has the potential to bring multiple benefits to healthcare systems throughout the world. However, there are also some potential drawbacks which need to be considered – and improved upon – in order to get the most out of AI and machine learning technology: 

Data quality

An AI algorithm is only as good as the data which is fed into it. Owing to the abundance of data generated by the healthcare sector, it’s inevitable that some of this data will be of poor quality, and any inaccuracies in data collection, input or processing will be reflected in inaccuracies in the resulting AI algorithms.

Bias

AI algorithms are often affected by intrinsic bias, which can be difficult to identify. For example, an AI designed to detect disease in a medical image may show bias towards the specific patient characteristics of its training sample (signs which may be present in patients from one region but not another). A study carried out by scientists at Harvard also discovered that AI’s also showed bias towards the make and model of imaging equipment, performing more accurately with formats from certain CT machine manufacturers. Put simply, used alone, AI’s aren’t able to reliably identify the wealth of racial, geographic or technological diversity that exists between patient groups. 

Decay

Because they are continuously learning and applying that knowledge, AI models decay over time – as their input moves further away from their training data set – which almost always results in a reduction in accuracy. 

Perpetuation of healthcare inequalities

Any inequalities in a data set – for example under-representation of certain groups, such as women or people of colour – translates into inaccuracies in an AI algorithm, which can lead to researchers neglecting differences in the way that diseases are manifested in those groups.

The future of AI in healthcare

There is no doubt that AI and machine learning will play an important role in the future of healthcare, providing access to large, accurate data sets and supporting the work of physicians and healthcare managers, whilst improving the patient experience.  

Deep learning algorithms can help us to access this wealth of clinical data, maximise performance through the use of computer-aided diagnostics, imaging and modelling tools and ultimately, improve experience and outcomes for both clinicians and patients.

However, in order to optimise this technology whilst maintaining transparency and equality, healthcare providers must also be aware of the potential drawbacks of AI, and take steps to address some of the issues outlined above.

It’s likely that the future healthcare landscape will see AI and machine learning supporting the work of trained, qualified professionals, to enhance the accuracy and validity of healthcare decision-making and provide the best outcomes for providers and patients. AI is a great tool in healthcare, but as a welcome addition to human expertise and professional experience, not a replacement for them.