Artificial Intelligence and Healthcare: Transforming Medical Practice

Healthcare systems worldwide are grappling with immense pressures to meet the ‘quadruple aim’: enhancing population health, improving patient experience, boosting caregiver satisfaction, and curbing escalating costs. Factors like aging populations, the rising prevalence of chronic diseases, and overall healthcare expenses challenge governments, payers, regulators, and providers to innovate healthcare delivery models. The recent global pandemic further underscored these issues, highlighting workforce shortages and care access inequities, demanding systems that can both perform effectively and transform concurrently by leveraging real-world data. Technology, particularly Artificial Intelligence And Healthcare applications, offers significant potential to address these supply-and-demand challenges. The growing availability of diverse data types (genomic, economic, demographic, clinical, phenotypic) combined with advancements in mobile tech, IoT, computing power, and data security creates a pivotal moment for integrating AI into healthcare, fundamentally reshaping delivery models through AI-augmented systems.

Cloud computing plays a crucial role, offering the capacity to analyze vast datasets faster and more cost-effectively than traditional on-premises infrastructure. Consequently, technology providers are increasingly partnering with healthcare organizations, driving AI-led medical innovation enabled by cloud platforms. Leaders in the tech industry recognize AI’s transformative power, viewing healthcare as a prime application area and significant business opportunity, aiming to empower clinicians with new assistive technologies. This review summarizes key breakthroughs in AI’s healthcare applications, outlines a roadmap for developing effective AI systems, and explores the future trajectory of AI-enhanced healthcare.

Understanding Artificial Intelligence in a Medical Context

At its core, artificial intelligence (AI) involves creating intelligent machines using algorithms that enable them to mimic human cognitive functions like learning and problem-solving. AI systems can anticipate issues or address them proactively, operating intentionally, intelligently, and adaptively. AI excels at identifying patterns and relationships within large, complex datasets. For instance, an AI system might synthesize a patient’s complete medical history into a single score indicating a probable diagnosis. Furthermore, these systems are dynamic, learning and adapting as new data becomes available.

AI encompasses various subfields, including machine learning (ML) and deep learning (DL), which add intelligence to applications. ML focuses on algorithms that allow computer programs to learn from experience automatically. ML subtypes include supervised learning (learning from labeled data), unsupervised learning (finding patterns in unlabeled data), and reinforcement learning (learning through trial and error with rewards/penalties). Ongoing research also explores semi-supervised, self-supervised, and multi-instance ML.

Building Effective and Trusted AI-Augmented Healthcare Systems

Despite over a decade of focus, the adoption of AI in clinical practice remains limited, with many AI healthcare products still in development. Often, attempts are made to fit AI solutions to healthcare problems without adequately considering the local context, including clinical workflows, user needs, trust, safety, and ethical implications.

The perspective here is that AI should augment, not replace, human intelligence. Building AI for healthcare must preserve the crucial human element in medicine, focusing instead on enhancing the efficiency and effectiveness of human interactions. True AI innovation in healthcare stems from a deep, human-centered understanding of patient journeys and care pathways. A problem-driven, human-centred approach is essential for creating effective and reliable AI-augmented healthcare systems.

Design and Develop

The initial stage involves designing and developing AI solutions for pertinent problems using a human-centred approach and experimentation, engaging all relevant stakeholders, particularly end-users in healthcare.

Stakeholder Engagement and Co-creation

Assemble a multidisciplinary team comprising computer and social scientists, operational and research leaders, clinical stakeholders (physicians, caregivers, patients), and subject matter experts (e.g., biomedical scientists). This team should include authorizers, motivators, financiers, conveners, connectors, implementers, and champions. Such diverse expertise is vital for defining problems, goals, success metrics, and milestones effectively.

Human-centred AI

This approach integrates an ethnographic understanding of health systems with AI capabilities. Employ user-centred research (e.g., qualitative studies) to first grasp the core problems: What is the problem? Why is it significant? Who does it affect? Why hasn’t it been solved? Why isn’t it receiving attention? Understand the needs, constraints, and workflows within healthcare settings, as well as the barriers and facilitators to AI integration. Once problems are defined, determine which are suitable for AI solutions and if appropriate datasets are available for building and evaluating the AI. Contextualizing algorithms within existing workflows ensures they align with established norms and practices, enhancing adoption by providing relevant solutions.

Experimentation

Focus on piloting stepwise experiments to build AI tools. Utilize tight feedback loops with stakeholders to enable rapid learning and incremental adjustments. Experimentation allows for testing multiple ideas simultaneously, identifying what works, what doesn’t, and understanding the reasons why. This process helps clarify the AI system’s purpose, intended uses, target users, and potential harms or ethical concerns (like data privacy, security, equity, and safety).

Evaluate and Validate

The next crucial step is iteratively evaluating and validating the AI tool’s predictions to assess its performance. Evaluation spans three key dimensions: statistical validity, clinical utility, and economic utility.

• Statistical validity: Assesses the AI’s performance based on metrics like accuracy, reliability, robustness, stability, and calibration. High performance in retrospective, simulated (in silico) settings alone does not guarantee clinical usefulness.
• Clinical utility: Requires evaluating the algorithm in a real-time clinical environment using hold-out and temporal validation datasets (e.g., longitudinal data, data from different geographic locations) to demonstrate effectiveness and generalizability across diverse settings.
• Economic utility: Measures the net financial benefit derived from the investment in the AI system relative to its costs.

Scale and Diffuse

Many AI systems are initially developed to address a problem within a specific healthcare system, tailored to its unique patient population and context. Scaling up these systems demands careful consideration of deployment methods, processes for model updates, regulatory compliance requirements, variations between different healthcare systems, and the reimbursement landscape.

Monitor and Maintain

Even after clinical deployment, AI systems require continuous monitoring and maintenance. This involves tracking performance, identifying potential risks, and monitoring for adverse events through effective post-market surveillance. Collaboration between healthcare organizations, regulatory bodies, and AI developers is essential for collecting and analyzing data related to AI performance, clinical effectiveness, safety risks, and adverse outcomes.

Current and Future Applications of Artificial Intelligence in Healthcare

AI holds the potential to help healthcare systems achieve the ‘quadruple aim’ by fostering a future characterized by connected and AI-augmented care, precision diagnostics, precision therapeutics, and ultimately, precision medicine. Research into AI applications in healthcare is rapidly advancing, demonstrating potential uses across the entire sector, including physical and mental health, drug discovery, virtual consultations, disease diagnosis and prognosis, medication management, and health monitoring. The following outlines potential AI capabilities in the near, medium, and long term.

AI Today (Near Term: 0-5 Years)

Current AI systems primarily function as pattern recognizers or signal translators rather than reasoning engines capable of human-like clinical intuition derived from experience. They excel at translating complex data patterns. Healthcare organizations are starting to adopt these systems to automate time-consuming, high-volume, repetitive tasks. Significant progress is evident in precision diagnostics, particularly in areas like diabetic retinopathy screening and optimizing radiotherapy planning.

AI in the Medium Term (5-10 Years)

The medium term is expected to see substantial advancements in developing powerful algorithms that are more data-efficient, can utilize unlabeled data, and effectively integrate diverse structured and unstructured data sources (imaging, electronic health records, multi-omics, behavioral, pharmacological). Healthcare organizations will likely transition from merely adopting AI platforms to becoming active co-innovators with technology partners, developing novel AI systems specifically for precision therapeutics, such as in synthetic biology and immunomics. Large-scale adoption of precision imaging is also anticipated.

AI in the Long Term (>10 Years)

In the longer term, AI systems are projected to become significantly more intelligent. This evolution will enable healthcare systems to achieve a true state of precision medicine through deeply integrated AI-augmented healthcare and fully connected care networks. The paradigm will shift from a reactive, one-size-fits-all approach to a proactive, preventative, personalized, and data-driven disease management model. This transformation aims to deliver improved patient outcomes and enhanced clinical experiences within a more cost-effective delivery framework, potentially featuring autonomous virtual health assistants and networked care organizations operating on a unified digital infrastructure.

Connected/Augmented Care

AI can significantly reduce healthcare inefficiencies, optimize patient flow and experience, and improve caregiver satisfaction and patient safety across the care continuum. For example, AI applied to remote patient monitoring via wearables and sensors can facilitate the early identification of patients at risk of deterioration, enabling timely intervention. Long-term visions include a fully interconnected digital infrastructure linking clinics, hospitals, social care services, patients, and caregivers, possibly leveraging passive sensors and ambient intelligence for seamless monitoring and support.

Virtual Assistants and AI Chatbots

AI-powered chatbots, such as those developed by Babylon Health and Ada, are increasingly used by patients for symptom checking and guidance on subsequent actions, particularly in community and primary care settings. When integrated with wearable devices like smartwatches, these AI assistants can provide valuable insights to both patients and caregivers, helping improve behaviours related to sleep, activity, and overall wellness.

Ambient and Intelligent Care

The emergence of ambient sensing technologies, capable of monitoring individuals without requiring wearable peripherals, represents another frontier in connected care. These systems could passively collect health-relevant data within a person’s environment, contributing to more continuous and unobtrusive health monitoring and potentially enabling more predictive and anticipatory care models.

Precision Diagnostics

Diagnostic Imaging

Automated medical image analysis is currently the most prominent application of Artificial Intelligence And Healthcare. A review of AI/ML-based medical devices approved between 2015-2020 in the USA and Europe revealed that radiology was the predominant field, accounting for over half of the approved devices (58% in the USA, 53% in Europe). Numerous studies have shown AI systems matching or surpassing human expert performance in image-based diagnosis across various specialties:

• Radiology: Detecting pneumonia on chest X-rays.
• Dermatology: Classifying skin lesions from clinical images.
• Pathology: Identifying lymph node metastases in breast cancer from whole-slide images.
• Cardiology: Diagnosing heart attacks from electrocardiograms with accuracy comparable to cardiologists.

Exemplars exist within systems like the NHS (e.g., University of Leeds Virtual Pathology Project), suggesting wider adoption of AI-based diagnostic imaging is likely in the medium term.

Diabetic Retinopathy Screening

Screening for diabetic retinopathy is crucial for preventing diabetes-related vision loss. Given the large number of diabetic patients and limited ophthalmology resources globally, screening presents a significant cost and logistical challenge. Automated AI algorithms developed and tested in the USA, Singapore, Thailand, and India have demonstrated strong diagnostic performance and cost-effectiveness. Notably, the FDA-approved AI algorithm ‘IDx-DR’, which showed 87% sensitivity and 90% specificity for detecting moderate diabetic retinopathy, received Medicare reimbursement approval, marking a significant step towards clinical integration.

Improving Radiotherapy Planning Precision

AI offers substantial assistance to clinicians in the preparation and planning stages of radiotherapy for cancer treatment. Manual segmentation of medical images to delineate tumors and organs at risk is a time-consuming and labor-intensive task. AI-based tools, like the open-source InnerEye technology developed by Microsoft Research, can drastically reduce this preparation time – by up to 90% for head and neck and prostate cancers. This acceleration can significantly shorten waiting times for patients needing potentially life-saving radiotherapy.

Medical imaging scans showing AI-powered tumor segmentation using the InnerEye deep learning toolkit for precise radiotherapy planning in head and neck cancer.

Precision Therapeutics

Achieving true precision therapeutics requires a vastly improved understanding of disease mechanisms at cellular and molecular levels. Researchers worldwide are collecting multimodal datasets (genomic, proteomic, clinical, imaging) to identify digital and biological biomarkers for diagnosis, severity assessment, and progression prediction. Key future AI applications lie in immunomics, synthetic biology, and drug discovery.

Immunomics and Synthetic Biology

By applying advanced AI tools to complex multimodal datasets in the future, researchers aim to gain deeper insights into the cellular basis of diseases. This understanding could enable better disease clustering and identification of patient subgroups, leading to more targeted preventive strategies and personalized treatment options, particularly leveraging immunomics. Such advances could revolutionize care standards, especially in oncology, neurology, and rare diseases, tailoring the care experience to the individual.

AI-Driven Drug Discovery

AI is poised to significantly enhance clinical trial design, optimize drug manufacturing processes, and improve various combinatorial optimization tasks within healthcare. Early examples, like DeepMind’s AlphaFold predicting protein structures with remarkable accuracy, demonstrate AI’s potential to accelerate the understanding of disease processes and facilitate the development of more targeted therapeutics for both common and rare conditions. This capability could fundamentally change the economics and timeline of drug development.

Precision Medicine

New Curative Therapies

Synthetic biology has yielded breakthroughs like CRISPR gene editing and some personalized cancer treatments over the last decade. However, developing these advanced therapies remains highly inefficient and costly. In the future, enhanced access to comprehensive data (genomic, proteomic, glycomic, metabolomic, bioinformatic) combined with AI’s ability to manage systemic complexity could transform how we understand, discover, and manipulate biology. AI could improve drug discovery efficiency by better predicting drug efficacy early on and anticipating adverse effects, thereby reducing late-stage failures and potentially lowering the cost of novel therapies, making them more accessible.

AI Empowered Healthcare Professionals

In the longer term, healthcare professionals will increasingly leverage AI as a tool to augment their capabilities. This synergy will enable them to deliver safer, more standardized, and highly effective care, operating at the peak of their professional license. For example, clinicians might utilize an ‘AI digital consult’ involving ‘digital twin’ models of their patients. These detailed digital replicas would allow ‘testing’ the effectiveness and safety of interventions (like specific cancer drugs) in a virtual environment before administering them to the actual patient, ushering in an era of truly predictive and personalized treatment planning.

Overcoming Challenges in AI Healthcare Adoption

Despite the immense potential, the widespread adoption and deployment of artificial intelligence and healthcare systems face significant hurdles. Key challenges include ensuring data quality and secure access, developing robust technical infrastructure, building organizational capacity and digital literacy within the workforce, and navigating complex ethical considerations. Establishing responsible AI practices, ensuring patient safety, and developing clear regulatory frameworks are also critical factors that need careful attention and resolution. Addressing these multifaceted issues is paramount for realizing the full benefits of AI in transforming healthcare delivery.

Conclusion and Key Recommendations

Breakthroughs in artificial intelligence and healthcare promise to reshape medicine, paving the way for a future that is more personalized, precise, predictive, and portable. Whether the adoption of these technologies will be incremental or revolutionary remains uncertain, but health systems must prepare for the digital transformation AI will inevitably bring. For systems like the NHS, AI offers the potential to free up healthcare professionals’ time, allowing them to focus more on patient-centred care. Furthermore, leveraging global data assets could enable clinicians to operate at the forefront of scientific knowledge, delivering consistently high standards of care universally. Globally, AI could serve as a powerful instrument for advancing health equity.

The past decade focused heavily on digitizing health records for efficiency and administration. The upcoming decade will likely centre on extracting valuable insights from these digital assets using AI to drive tangible improvements in clinical outcomes and generate novel data tools. We stand at a critical juncture where medicine and technology converge. While opportunities abound, formidable real-world implementation challenges must be overcome. Expanding translational research specifically focused on healthcare applications of AI is crucial. Equally important is investing in upskilling the current and future healthcare workforce and leadership to be digitally adept, enabling them to understand and embrace, rather than fear, the potential of an AI-augmented healthcare system.

Healthcare leaders planning to leverage AI should prioritize:

• Ethical Data Access: Establishing clear processes for responsible and ethical access to sensitive, often inconsistent and siloed healthcare data, ensuring it is suitable for ML development and validation.
• Domain Expertise: Ensuring access to clinical and domain expertise to interpret data correctly and define the rules needed for generating meaningful insights.
• Computing Power: Securing sufficient computational resources, increasingly available via cloud computing, to enable real-time decision support.
• Implementation Research: Critically investigating the practical challenges that arise when deploying algorithms in real-world clinical settings, focusing on building trust and integrating AI seamlessly into workflows.