AI In Healthcare: Prevention, Diagnostics and Treatment
Artificial Intelligence Is Changing Healthcare For The Better. AI, the impact of which is simultaneously over-hyped and under-rated in most sectors, is catalyzing revolutionary use cases in healthcare. Deep learning has come a long way from identifying cats and dogs and can now perform independent image-based diagnosis with comparable or better accuracy than (human) doctors. X-Ray or scan-based diagnosis of tumours, fractures, strokes, electrodiagnosis can be entirely done by algorithms. While several of these options have received FDA approval and starting a clinical trial or beginning to be used in production, researchers are attempting selfie diagnosis to detect ~50 diseases from abnormalities in eye colour (Nature,2018) and diagnosis from (molecules emitted from) body odour.
Using data from wearables or behavioural observations such as changes in gait, driving patterns, mouse usage etc., machine learning algorithms can predict the onset of physiological (particularly neurological or cardiovascular) or psychiatric disorders, assist management of chronic conditions such as diabetes or epilepsy, and even raise real-time warnings. Unsupervised or semi-supervised techniques such as clustering can help uncover obscure patterns in the individual as well as population health leading to early diagnosis or prevention. Prognosis evaluation and alerting for patients have achieved up to 2X better results with supervised learning algorithms than existing methods using standard regression in most cases.
Machine learning is also largely responsible for the advances in precision medicine. Personalized treatment, along with genome and molecular sequencing is changing cancer treatment. One of the latest milestones being Deepmind’s Alphafold system which could help uncover the mysteries of proteins and the conditions caused by defects in their structure such as Parkinson’s, Huntington’s, cystic fibrosis etc.
Identifying rare diseases and gathering up-to-date medical information is now just one voice query away. Chatbots assisting in personal health management or answering primary questions are finding popularity. The clinical trial design is better informed and several times more efficient — identifying patients who benefit the most, characteristics that make drugs work, finding different applications for existing drugs, all powered by different branches of AI. We are not that far from fully robotic surgeries especially in standard and routine cases coming far from assisting to improve surgical precision by following hand movements.
Automated diagnosis and prescriptions could help better utilize the heavily constrained healthcare resources enabling early and faster detection often resulting in a more efficient and effective course of actions. In developing countries, AI could provide access to healthcare for millions. Hospital administrations are benefiting from the predictive capabilities of data by automating parts of operations such as staffing, scheduling, logistics, documentation, incorporating patient feedback etc. Monitoring prognosis of both in and outpatients has proven a task AI can perform to a reliable degree saving lives of critical care patients and helping manage cases.
Medical dramas becoming less exciting aside, a take-over by algorithms will be dependent on the data they are trained on which explains the progress in image-based diagnosis as the relationship between input data and predicted outcomes are relatively straightforward. When the connection is less explicit or when there is more than one statistically significant answer, things get murky and we need to rely on doctors. IBM’s Watson, after the initial hype, has come under scrutiny for a few misses recently, most of which were due to data inadequacy and misguided expectations. There is still a long way to go starting from data acquisition, Healthcare data had been unorganized and unstructured for a long time. As that changes with natural language processing to extract information from medical records, journals, notes etc and standardize dictations and transcripts, and data collection becomes straight-forward, AI will become an indispensable ally.
About the Author
Aparna is an Enterprise Data and Analytics Consultant based in London. She has several years of full cycle digital transformation project experience designing and developing Data strategy and architecture. She holds an MBA from Imperial College and a Bachelor’s degree in Electronics Engineering. As part of the MBA she did a consulting project for eHarmony about the impact of AI in dating, was part of Best Practice AI team and did her thesis on developing a guide for Enterprises to develop AI strategy.