Integrating AI into Healthcare By Kish Adoni, Research and Membership Intern, AMRC Published: 26 February 2020 Five years ago, Artificial Intelligence (AI) was incapable of identifying a cat. Today, even well disguised skin cancer cells can’t hide from the almost sentient sophistication of cutting-edge AI. It’s small wonder that in 2016, Healthcare AI projects attracted more investment than AI projects within any other sector of the global economy. This technology has the potential to facilitate the widespread implementation of rapid and accurate diagnoses, personalised treatment, improved prediction of disease progression, and a complete reworking of the drug discovery pipeline… and that’s just a smattering of examples. To include everything would see this blog way past its word limit! So, what is AI and how does it relate to healthcare? AI is the development of a computer system that can perform tasks normally requiring human intelligence. Machine learning is the most lauded sub-category of AI in the medical community. It is a statistical technique that uses large datasets to generate an output. Such is the power of Machine Learning that Google DeepMind launched a device which can scan an Optical Coherence Tomography image for 50 different eye diseases with a 94.5% accuracy, on par with world leading ophthalmologists. Google, Enlitic and various other Start-Ups are developing AI derived image interpretation algorithms. Thus, AI can analyse unstructured medical data - radiology images, blood tests, EKG’s, genomics and patient medical history - to aid diagnoses. The technology is trained to detect the most minute of changes in images, identifying diseases long before a doctor could! These findings can then be linked to the patient’s medical records to assess their risk potentially saving thousands of lives. Machine Learning algorithms also feature in wearable watches that can monitor 24/7 patient health and, coupled with medical records, spot aberrant data. This puts the patient in control of their own health, providing immediate access to any potential health risks. AMRC’s members are at the heart of AI research Our daily dependence on smartphones provides an ideal opportunity for the storage of thousands of continuous datapoints; and AMRC members have pinpointed the huge potential in harnessing this for health benefits. Alzheimer’s UK have partnered with various organisations to develop Machine Learning based physical assessment tools. These programs can detect cognitive decline up to 15 years before formal diagnosis. One such example is ‘Sea Hero Quest’- a 7-minute smartphone test that can predict whether the user has the APOE4 gene, which predisposes patients to Alzheimer’s. Parkinson’s UK, together with Oxford University, are developing an app that measures voice, balance, walking, reaction-time, finger tapping, rest tremor and posture to provide personalised predictions on disease progression. Cancer Research UK are developing various AI based technologies, one of which combines patient history and lifestyle with 300 genetic indicators to predict women’s predisposition to breast cancer. They are also investing £56 million towards the study of AI in radiography. The Medical Research Council and The British Heart Foundation (BHF) have invested £6.7 million for a global human cell atlas initiative that can map every cell in the human body. The aim is to create a data-foundation from which various AI projects could spawn. BHF are also using AI to study pulmonary hypertension. Cystic Fibrosis Trust have collected over 2 million data-points from 9,695 patients over one year. These data will provide the feedstock for next-generation algorithms that could help to deliver more accurate treatment methodologies. Why haven’t we realised the full potential of AI yet? With all these shiny new toys available for the health sector to play with, you’d be forgiven for thinking we should be seeing widespread implementation in the medical industry. However, there are still a plethora of barriers to optimising the tech at our disposal. A key issue is integrating the tech into industry alongside the workforce. For example, many image analysis machines only search for specific datapoints, whereas a human radiographer would analyse thousands of different detection tasks as well as engaging in diagnosis, treatment, patient interaction, palliative care and much more. For automated patient notes and Machine Learning algorithms, access to large amounts of sensitive data would be required, causing huge legal and moral challenges in acquiring data ownership. Another legal issue with Machine Learning algorithms is in accountability, as it is often impossible to understand how a machine has gone from data analysis to diagnostic output. In the case of misdiagnosis, who is responsible? In fact, is it acceptable for a doctor to make a diagnosis despite not knowing how AI has generated this output? Perhaps the most challenging moral issue is the potential for AI to exasperate health inequalities. Since algorithms are trained using datasets that may involve certain groups and not others, the technology could reflect this inherent bias. For example, some BAME communities tend to have less access to world-class healthcare, consequently their data is not part of training datasets and as such AI may not work as effectively when using their data. Another example is in wearable watches, likely trialled on digitally literate demographics; thus, it may not be as practical for less tech-savvy patients. For AI to deliver on its full potential, patient involvement is vital Lack of information, especially in our social media age, is quickly filled with misinformation and public trust can be lost - think GM crops. Ultimately, the technology of Artificial Intelligence is not holding it back, but rather the policy that must facilitate its implementation.