Artificial Intelligence was a term most often found within the genre of Science Fiction. Now, in hindsight, it seems almost prophetic. Artificial Intelligence (AI) could be defined simply as the ability of a computer to perform its own reasoning, usually associated with human intelligence. AI technology is rapidly growing more sophisticated, efficient, accurate, faster, and cheaper. AI is being applied to a wide range of fields in business and society at large, but what potential does AI have in the healthcare space?
The reality is, AI has already seen action in healthcare, while it is not the norm or at a large scale (yet) the introduction of AI into this world brings with it many advantages. Most significantly, AI could analyse large collections of data significantly faster than a human brain and can interpret and offer solutions to complex medical problems.
While it is still in its infancy, the potential for AI to assist in diagnosis is undeniably exciting. Around the world there have been developments that make robotic diagnosis much closer to a reality. Stanford University recently conducted a study testing an AI algorithm in its ability to detect skin cancers and found it to perform at a similar level to human dermatologists. In Denmark, AI software company Corti conducted an experiment to determine the potential for early detection of cardiac arrest. They did this by having a computer listen in while human dispatchers received emergency calls. They then used a deep-learning program to analyse what the caller is saying, their tone of voice, and the background noise to detect cardiac arrest. This particular experiment resulted in a success rate of 93%. AI has not yet reached its full potential in healthcare, and increasingly in the future AI will aid clinicians in their diagnosis and clinical judgment. One area that could see great benefit from AI is in Clinical Decision support, that is in enhancing care by offering relevant clinical data, patient information or other health information. Improving care requires timely intervention, by recognising patterns in health data AI can help to provide clinicians with early warnings of deteriorating conditions. This would allow them to act proactively and avert possible negative events even when such warnings would otherwise require extensive analysis of substantial amounts of data.
Like diagnosis, the integration of AI in image analysis is in the early stages. The potential value it has, however, is undeniable. When performed by a human, analysing radiology images can be very time-consuming, potentially taking two hours or longer. This compares to a machine learning algorithm developed by an MIT research team that analysed 3D scans up to 1000 times faster than a regular clinician, providing near real-time assessment of scans. This could be a considerable help during surgery, allowing surgeons to assess the success of a surgery significantly sooner, theoretically while still in the operating room.
The use of AI in assisting and automating administrative workflows has freed up clinicians to prioritise urgent issues and save time on non-clinical tasks. In fact, an estimated 40% of tasks performed in administration can be automated. One area that has significant potential for streamlining through AI is in the billing process. The inclusion of Robotic Process Automation (RPA) to enter, process, and adjust claims decreases the potential of human error and simplifies the process. Artificial intelligence also offers benefits in similar areas such as voice-to-text transcriptions, prescription of medicines, and ordering tests.
However, for AI to be at its most effective, it needs access to sensitive patient data which presents one of the major challenges facing the use of Artificial Intelligence. For companies developing AI solutions and the healthcare organisations working with them, they must be careful not to violate any of the privacy laws in place to protect patient data. It is critical that a balance is found between providing data for AI learning and protecting the privacy of patients. Which leads into the second big challenge in this area; Without a sizeable amount of data, the results of an AI application’s work could lack validity and accuracy, and while there is a significant amount of high-quality clinical data out there it has strict controls and is limited primarily to exchange between hospitals.
These challenges can be mitigated, however. Firstly, by ensuring the dataset used to teach the AI is clinical but also by ensuring it is sufficiently well-balanced to avoid biases in the AI and ensure its reviewed to avoid errors and duplication, which are possible even in clinical datasets. Another simple solution is to use a pre-trained and proven AI as a starting point.
Artificial Intelligence is not yet perfect. It is undeniable that the existing use cases demonstrated across the globe combined with the associated leaps forward, are exciting. Regardless of how it is used, AI will reduce the chance of human error while increasing the speed of resolution and quality of clinical outcomes. By embracing AI as a part of healthcare’s technological tool belt it can benefit researchers, clinicians, and most importantly, patients.
By Matt Green who has been leading digital transformations centralised around automation for the last 15 years. Working around the globe and across industries, he has helped many household names setup, scale and realise true value with automation. He’s an expert in all things AI and automation and loves to chat all about it.