Healthcare has always been a data-rich area, but with new technologies for processing and structuring, and new ways of collecting data, such as using sensors, like many other industries, the available data is growing exponentially. Artificial Intelligence (AI) makes it possible to analyse all this data in real-time by combing Machine Learning (ML) and Natural Language Processing (NLP), in order to gain valuable insights.
A survey by OpenText found that 100% of Life Science companies were considering AI in the next 12 months, but many didn’t know where to start. In this article, we will go over some of the ways that AI is being used in healthcare, as well as developing use cases and some potential applications in the future.
Starting with the basics, AI and automation could have a huge impact on the administrative side of healthcare. The industry relies on records, checks and organisation, and in this area alone there is a pool of opportunity to optimise processes and increase efficiency, from finding outpatient eligibility to moving data and even interfacing with patients online in the form of chatbots that can talk through symptoms and schedule appointments. These optimisations will save time, increase efficiency and cut costs, relieving pressure on health organisations like the NHS. Service operations are already leading the way in AI adoption in healthcare, but there is plenty of room for growth in the remaining areas.
Deep learning for early detection and diagnosis
Using vast amount of medical imaging data, deep learning models for diagnosis have been making the news as they prove themselves to be as accurate as a medical specialist at predicting and diagnosing various cancers and other diseases. The accuracy will only increase as time goes on, supporting healthcare professionals who are already pushed to the edge with workload, particularly with various ageing populations across the globe.
AI will be able to support doctors, helping them diagnose faster, and prescribe more accurately and suggest correct dosages, improving the treatment of patients. And with a holistic view of patient data, platforms can be built to analyse health records and pick out other patterns that humans wouldn’t even notice. This could contribute to the easier detection and diagnosis of certain diseases, saving lives in the cases that earlier detection increases chances of survival.
Applying AI to drug development
Traditionally, developing drugs has been a lengthy and costly process of rigorous testing and analysis, taking an average of 12 years and funds running into the billions to get to market. For every drug that makes it, thousands will fall at the various stages of testing. With so many trials and so much information, comes a whole lot of data to which AI can be applied to streamline the development process and significantly cut costs.
In 2018, research was published showing how AI was used to identify a pathway of harmful toxin formation in an oral anti-fungal medicine. Not only was the machine learning algorithm able to solve the problem that had been eluding researchers for 22 years, but it revealed that such algorithms could be used to figure out other possible metabolic pathways, how a drug will respond in chemical environments, reveal patterns and make predictions in drug development. It could help identify unsuccessful lines of investigation to drop sooner so less time and money is wasted, and to help successful drugs get to market faster.
AI can be applied in used to analyse scientific papers, extracting text to find patterns and make links that could take humans years to stumble across if they don’t read the right papers. It is also being used to identify candidates for drug trials who may be more at risk of bad reactions. Additionally, it could be used to generate ideas. One company is aiming to replicate the decision processes of medicinal chemists with machine learning, meaning machines can generate the ideas and allow the chemists to focus on the higher details. As of January 2020, 186 startups were using AI in drug discovery in different ways, and this is only the beginning.
Machine learning for designing personalised treatments
With all the data available, even at an individual level, healthcare could shift from reactively treating illnesses, to being more predominantly preventative and generating personalised outcomes. Patients are closer than ever to their own health since the mainstream use of health apps and wearable IoT devices like FitBit. This self-generated data combined with electronic medical records, processed with AI analytics, could be used to deliver real-time insights and lifestyle advice, reducing the need for medical interventions.
In fact, there is a lot of potential for wearable IoT devices, which already tracks things like heart rate, indicating stress and giving activity prompts. But they could be developed to monitor chronic diseases and conditions, such as diabetes. Symptoms like blood sugar levels could be tracked in real-time, giving patients more time to respond. To get even more personal Google X are currently developing smart magnetic nanoparticles intended to live in your bloodstream to collect data at the molecular level and predict medical conditions from cancer to heart attacks.
Furthermore, using machine learning to analyse patient data, personalised treatment plans can be created that take into account individualities. By cross-referencing similar patients and constantly improving its algorithm, AI can predict how a patient might respond to certain treatments and recommend the best course of action. Such models are already being used in cancer treatments.
Robotics in healthcare
The use of robots in healthcare already ranges from transporting supplies and dispensing medicines, to analysing samples and even conducting surgery. Several hospitals are already using robots such as Aethon’s TUG and other AVGs that use sensors to navigate their way through hospital corridors, even calling lifts to get to where they need to go.
Many hospitals also use robotic surgical systems for intricate operations, microsurgeries, to enhance the performance of surgeons and to reduce the invasiveness of surgery. AI can then be used to analyse these surgeries and determine patterns and best practices, with capabilities moving towards real-time analysis. AI-assisted surgery still has a long way to come, so we can expect to see a lot of growth in this space.
AI for education and training
AI also has some relevant application in education that can be used in medical training. Using Knowledge Space Theory, training systems can be built that can gauge a student’s knowledge and adjust to meet their requirements, drawing on vast cloud databases for up to date knowledge. When you roll this into a smart-app, you create an accessible training system that can be used anywhere, at any time to refresh trainees and professionals alike.
Virtual Reality (VR) can also play a role in medical training, allowing students and doctors to practice operations that they don’t come across often, simulate emergencies and develop surgical skills without any risk to human subjects. It can also be used to train general practitioners by simulating scenarios, rather than the traditional way of building experience from real patients. VR is transforming medical training because aside from reducing risk, it helps trainees retain more information, and learning can be analysed so students will receive insights and feedback.
Trust and ethics of AI
With all the potential that AI holds in healthcare, we must be careful to build a system that we can trust and that excludes bias. With so much data in healthcare coming from white and male subjects, algorithms built on such data will be prone to racial, cultural, gender and minority biases. We discuss this issue in our article on removing AI bias, and suggest ways to avoid implicating these biases, i.e. working collaboratively to help spot these issues, and ensuring that data sets are large enough to be inclusive of sufficient examples.
Several facial recognition systems have been found to perform poorly on female or darker-skinned individuals. An algorithm designed to identify repeat offenders has shown racial bias. Advertising algorithms have shown gender bias, showing executive positions to less woman, resulting in fewer applicants. When bias comes through in AI it can be quite harmful, and even more so in healthcare when wellbeing and lives are in the mix. By raising awareness of the potential issues, fairer models that are transparent, compliant and robust can be built by basing them on reliable and inclusive data.
It is also important for professionals that work with such systems to understand their floors in order to be able to call them out. As it stands, all forms of AI have a long way to come before reaching perfection, but in the meantime, AI can complement many professions and relieve some of the pressure on workers. With the right guidance and policies, AI innovations can keep adding to the progress of society, rather than taking away.