By Dr. Kashif Rajpoot
AI is bringing astonishing and pervasive advances through applications deployed around us in our daily lives, for example on our smartphones in the form of language translation and face recognition, and in financial services for loan decision and credit fraud prevention. The current success of AI is duly credited to deep neural networks that simulate human brain to develop the powerful ability in computers for learning insights, complex patterns, and trends from data.
There have been rapid advances in deep networks, owing to fascinating new ideas in the design of neural networks, availability of huge amounts of digital data, and cheaper hardware like GPUs. The AI’s potent learning ability enables it with the decision-making power to analyse vast amounts of digital datasets in healthcare which otherwise would be impossible
to analyse in a systematic way. With AI’s intelligent decision-making powers in its armoury, the stage is set to witness the next wave of technological disruption in healthcare. Health AI possesses the potential to help clinicians in diagnostic,
prognostic, patient care, and treatment planning.
According to an analysis by Accenture , AI is considered an engine for growth in healthcare by becoming its nervous system. Accenture research shows that key health AI applications can potentially create $150 billion in annual savings for the US
healthcare economy by 2026. The health AI market size has grown vastly from $600 million in 2014 to $6.6 billion in 2021. According to a recent report by World Economic Forum , AI expenditure in India is expected to reach $11.78 billion by 2025 and is expected to add $1 trillion to India’s economy by 2035.
Health AI Examples and Applications
AI for Cancer Analysis: Globally, cancer is responsible for an estimated 9.6 million deaths each year. In computational pathology, deep learning algorithms are automating the analysis of cancer tissue slides. The AI algorithms are being used to
identify and classify cells, detect cancerous cells, and identify patterns in tissue slides that can help with diagnosis and treatment. For example, a recent research study from Harvard has shown that an AI system termed TOAD can successfully
predict origin for cancers of unknown primary from whole slide images which is a considerable challenge. In another research study from the University of Warwick, AI automated cancer image analysis for profiling tumour microenvironments through
analysing cellular community. The tissue microenvironment analysis is very tedious and extremely timeconsuming to do manually. The AI system can free the pathologist to focus on more high-level intellectual tasks, thus improving patient care.
AI for Radiology: AI is being used in radiology to automate and improve processes such as image analysis-based diagnosis. Recently, an AI system termed CheXNet has beaten radiologists for x-ray image analysis to detect 14 different types of pneumonia from chest x-rays with an accuracy of 92%. The system was trained on 50,000 images by Stanford University researchers. Radiology at times has been listed most at risk in medical practice by AI. However, any models trained are for
very specific tasks whereas radiologists perform a set of comprehensive tasks which machine has no comparison to take care of.
AI for Drug Discovery: DeepMind, a subsidiary of Google’s parent Alphabet, presented AlphaFold as a deep machine learning system that can accurately predict the 3D structure of proteins. The prediction of 3D protein structure is a very complex problem and the AlphaFold development has created huge excitement in research community. It has the potential to revolutionize drug discovery, as it can enable researchers to quickly and accurately identify potential drug targets and design drugs that are more effective and have fewer side effects.
AI for Cardiovascular Disease Analysis: Cardiovascular disease is the number one killer disease globally, responsible for over 17.9 million deaths each year. AI systems are helping in understanding cardiovascular disease at clinical and research
level. Google Health has developed an AI system to detect heart disease from retinal fundus scans by training it on data from 284,335 patients. In a recent research study by the University of Birmingham, an AI based software ElectroMap was developed and openly shared to measure cardiac electrophysiology parameters for understanding how arrythmia develops in the heart.
AI for Dermatology: Recently, a research team from Germany, the United States and France trained an AI system on 100,000 images to distinguish dangerous skin lesions from benign ones. Comparing the AI system with 58 dermatologists from 17 countries, the research found that most dermatologists were outperformed by the convolutional neural network-based AI system.
AI for Preventive Healthcare: AI possesses enormous promise to analyse heaps of digital data supporting preventive healthcare. For example, Google Health developed ARDA as an AI powered solution to help address screenings for diabetic retinopathy. The system has screened over 100,000 patients to date. Similarly, Google Natural Language Processing (NLP) APIs are being used to extract insights from electronic medical records. Usually, patients’ history in clinics is well-documented in medical records but is typically buried as unstructured clinical notes. These notes provide a rich space to extract structured information from a natural workflow. The automatic extraction of insights, trends, and associations from clinical notes using Google NLP APIs is helping to improve patient care by understanding clinical protocols, pathways, and outcomes.
Research studies have reported success in the development of smartphone-based AI systems for early screening and diagnosis of various diseases to enable timely treatment and avoiding disease progression. Modern day smartphones are equipped with capable sensors that can enable inexpensive continuous monitoring and screening by relying on AI based analysis of the data from these sensors. The AI based telehealth systems connected with smartphones and remote care providers have the potential to provide healthcare at home. Such systems can offer health assistance, condition managing and medication management at home. A recent report by McKinsey predicts that up to $265 billion worth of healthcare services could shift to the home by 2025 while healthcare at home may deliver more value and higher-quality care.
Health AI Future
In summary, AI will undoubtedly transform healthcare future by automating and improving processes in diagnosis, treatment, and patient care. AI in healthcare is inevitable, where it will pave the way for efficient healthcare provision. The
improvement in efficiency through AI-powered technology will allow for more effective and enhanced healthcare accessibility for patients in developed and developing countries.
It is worth noting that health AI is aimed to support healthcare practice, but it is not designed or capable to replace human knowledge and experience. It will become more of an ‘assistant’ to a clinician to provide AI-augmented healthcare. AI can work with doctors for healthcare by providing them with data-driven insights to help them make better decisions. AI can automate mundane tasks such as data entry andscheduling, freeing up doctors to focus on patient care oranalyse large amounts of data to identify patterns and trends that can help improve patient outcomes. Medical practice relies on inherent human empathy that AI can never possess. A research analysis by IBM’s Institute for Business Value reports that AI’s real value lies in human augmentation, not in replacement.
This analysis highlights that AI will be integrated into the industry’s processes, systems, and interactions. Therefore, it is
extremely important for organisations and employees alike to continuously seek and adapt AI opportunities to reap the rewards of AI revolution.
Barriers to Health AI
While health AI possesses tremendous promise, it has crucial barriers to cross. These barriers include lack of access to data, lack of trust in AI systems, lack of understanding of AI technology, scaling up challenges, reproducibility issues, data
bias, and ethical concerns. Additionally, there are also regulatory and legal issues that need to be addressed before AI can be widely adopted in healthcare. Importantly, the clinicians need upskilling for AI to understand the fundamentals of AI, learn how to use AI tools, learn how to interpret the data generated by AI systems, and understand the ethical implications of using AI in healthcare.
(The author is an Associate Professor and Programme Director (Computer Science), University of Birmingham Dubai. Views expressed are personal and do not reflect the official position or policy of the FinancialExpress.com.)