For years, artificial intelligence (AI) may have merely seemed to be a pipe dream; only seen in movies and our imagination. But as we delve deeper into an age of Siri and Alexa, AI is becoming increasingly commonplace in our day-to-day lives. With big data forming the building blocks for AI technologies, coupled with the widespread adoption of big data meaning that we are generating an estimated 2.5 quintillion bytes of data every day,1 could we be in a position for AI applications to forever change the healthcare industry as we know it?

Potential of AI in healthcare

There exist many varieties of AI technologies, several of which may be applied to healthcare.2 These applications have the potential to change the industry. Time, costand efficiency could all be improved by their introduction.3 For example, the predicted future deficit of 12.9 million healthcare professionals (HCPs) globally by 2035, could be supported or minimised by the application of AI.4 Could these technologies serve to streamline processes, reduce administrative time and support HCP decision-making so as to reduce the impact of this deficit?4

The use of AI in healthcare has been of great interest since the 1970s. At this time, the predictive technology ‘MYCIN’ was developed to provide therapy recommendations for treating bacterial infections.2 Despite never being introduced into clinical practice, MYCIN demonstrated a 69% success rate in choosing effective therapy and set a foundation for the future of AI in healthcare.2

Machine learning (ML) is one of the most commonly used forms of AI. It is especially useful in healthcare to confirm and predict disease through analyses of structured data sets, such as imaging and genetic data.5 One example of ML is neural networks, a technology based on the way neurons fire in the brain. Neural networks can be used to determine binary queries, for instance a ‘yes or no’ answer to a clinical question.2 As with any technology, ML is constantly progressing, and new branches of ML are regularly being developed. Deep learning is one such branch which uses neural networks with many layers and has the potential to benefit the complex world of healthcare.6 The first applications of deep learning were in image analysis and were used to detect Alzheimer’s disease from magnetic resonance images (MRIs).6 Convolutional neural networks (CNNs) have also demonstrated diagnostic capabilities on par with 21 board-certified dermatologists in classifying images of different types of skin cancer from over 130 thousand images.6 Application of deep learning in medical research almost doubled in 2016 alone, showing great promise for the future.5

AI in diagnostics

As of today, AI applications are mainly being used in image analysis.5 AIs have the ability to detect abnormalities that human eyes can’t and can assist physicians in their decision-making processes.5 AI algorithms can be developed to learn clinical features from a large volume of existing healthcare data and use this to aid clinical diagnoses, whilst also learning and self-correcting with repeated use .5 For example, deep learning has been applied to detect malignant lung nodules from a dataset of more than 42,000 computerised tomography (CT) images.7

AI algorithms can be extremely beneficial in diagnosis, often proving to be more accurate than human diagnosis. A recent study reported that when interpreting Pulmonary Function Tests (PFTs), an AI system (ARTIQ.PFT) could correctly interpret results every time, while pulmonologists could only do this 74.4% of the time. The same PFTs also revealed that AIs made a correct diagnosis 82% of the time, whereas pulmonologists made an accurate diagnosis less than half of the time (44.6%).8 In addition to diagnosing patients, AI solutions may also be used to predict disease onset; for instance, 4-layer CNNs have been able to predict chronic obstructive pulmonary disease (COPD) onset in patients.6

Delivery of medication may also be enhanced with AI. AI-supported inhalers or ‘smart inhalers’ are able to track when patients are using them and how effective their inhaler technique is.9 This information can then be fed back to the patient’s physician to help ensure that patients are receiving the optimal benefit from using the inhaler.9 These inhalers can also include more patient-driven features such as medication reminders.9

Despite the many benefits afforded by the use of AI in healthcare, the complexity of diseases and their innate variability makes their effective use more challenging than in other sectors. Thus, the pace of progression and change in healthcare may prove a limiting factor in how AI is implemented.6 The approach taken to deep learning with medical AIs must consider how we should handle continuously changing healthcare data.6 

AI in drug discovery

Artificial neural networks (ANNs) have already been used to aid in drug discovery, for example, in predicting the sensitivity of cancer cells to drug compounds through analysis of the genomic profile of the tumor cell line.6 AI can be beneficial to drug discovery and can be used in a host of different ways, namely to predict drug interactions, targets, and side-effects when testing for new drugs.6

Platforms exist in which data from research papers, clinical trials, patents and patient records are used to train AI models to produce ‘knowledge graphs’ of conditions and their associated genes and/or compounds affecting them.10 AI can put these data into context, identify useful information for drug-discovery and can be used to find existing drugs that may already be in use.10 One such system, suggested over 100 existing compounds that might have the potential to treat motor neuron disease, which could otherwise have been missed.10

Additionally, AI software can be used in clinical trial recruitment to identify patients who would make the best candidates for a particular trial, thus streamlining the selection process.6 Although these assistances may seem quite small, their implementation can improve the efficiency of the drug discovery process, saving both time and money for all involved.3

AI has the potential to be used in various aspects of healthcare; its ability to predict disease risk, personalise prescriptions and streamline clinical trial recruitment are seemingly just the tip of the iceberg. 6 However, there is still more work to be done to truly realise the role of AI in medicine and we must be wary not to over-inflate our expectations of AI.9

Overall, it seems that the use of big data and AI solutions in healthcare is reaching the forefront in this technological age, but we are yet to see the well-anticipated development; from aiding diagnosis to curing disease.