Artificial intelligence: Role in healthcare

Introduction and Evolution of Artificial Intelligence

In today’s increasingly digital world, technology permeates every facet of our lives, from everyday tasks to complex healthcare decisions. Among the most transformative technologies is artificial intelligence (AI), which holds immense promise for revolutionizing healthcare. As with any breakthrough, however, AI’s potential benefits must be balanced by the challenges of its implementation.

The evolution of AI has been nothing short of remarkable since the term was first coined by John McCarthy in 1956.[1] AI refers to systems that enable computers and machines to simulate human intelligence—encompassing learning, comprehension, problem-solving, decision-making, creativity, and autonomy. It is important to also acknowledge Alan Turing, whose pioneering work on the Enigma code-breaking machine during World War II laid the groundwork for modern computing.[2] Turing’s quest to develop a machine capable of “imitating sentient behaviour”—the famous Turing Test—posed the question: If a machine can convincingly simulate human behaviour, does it indicate sentience?[3]

From early rule-based systems to today’s sophisticated machine learning models and generative AI, the field has advanced significantly. AI is no longer a theoretical concept; it is rapidly becoming integral to numerous industries. In this article, we attempt to provide an overview of AI being used in different sectors of the healthcare front.

AI in Healthcare

Diagnostics

This volume presents a meta-analysis of various AI technologies and their accuracy in disease detection. The findings underscore AI’s significant role in improving diagnostic precision and decision-making. The authors highlight the success of AI-based tools such as AI-based object detection system for malaria diagnosis (AIDMAN) for malaria detection, which demonstrated diagnostic accuracy of 95%, with an area under the curve of 0.96. Similarly, predictive models for chronic kidney disease (93% accuracy) and diabetes (91% accuracy) show tremendous promise. AI-powered TB screening via cough analysis also achieved an impressive 86% accuracy. These findings suggest that AI has the potential to significantly enhance diagnostic processes, leading to improved healthcare outcomes and cost reductions.[4] The current global scenario highlights how AI can “plug-in” to almost every aspect of healthcare and patient care.

Early applications of AI were for the “low hanging fruits,” where most of the data is readily available in computer-readable formats like radiology, to detect clinical conditions through analysis of X-rays, CT scans, and MRIs.[5] AI algorithms can interpret medical images with high sensitivity and specificity, and this is even rivalling the performance of healthcare professionals.[6] In terms of screening, AI systems can generate “heat maps” by analyzing vast numbers of histopathology slides, allowing early detection of cancers before symptoms appear.

AI in follow-up

Beyond diagnostics, AI-powered chatbots and virtual assistants can help guide patients throughout their “patient journey,” helping them understand symptoms, explain the medical advice given by doctor in detail, and even schedule follow-up appointments.[5] Gong et al.[7] created a chatbot for Type 2 diabetes management and supported both voice and text interactions. Over 12 months, it significantly improved patients’ quality of life (difference: 0.04, P = 0.04) with high engagement—98.9% of patients used it at least once. Scientists are developing chatbots based on patients reported outcomes to manage diseases like irritable bowel syndrome. After a long wait, this may be our real solution for providing personalized care in an overburdened healthcare system. It can help us tailor therapies to individual patient profiles and patient-specific data. However, issues with automation can be dealt with using “Human-In-the-Loop models”, which uses human reasoning, as seen in the finance sector. On similar grounds, Eachempati et al.,[8] have proposed a “Healthcare In-the-Loop” framework. This model is based on real-time learning rather than periodically training the software to learn new updates. This plays a critical role as rapid advances are manifesting in the diagnosis and treatment of various diseases.

Personalised real time updates

For those with chronic conditions or recovering from illness, patient monitoring becomes crucial. Wearable devices are already widespread and can continuously track vital signs. Powerful AI-based algorithms can now support complex pattern recognition in these signals. They can identify deviations from baseline levels personalized for that individual and alert the patient to seek care or even alert their assigned healthcare professional. This real-time monitoring can allow for swift intervention, preventing complications before they escalate. This is a step closer to “secondary prevention” or, for that matter, even “primary prevention” when it is used to track health data in healthy individuals. However, here we must consider the data privacy as a challenge, as the data is available and vulnerable to being used for purposes other than those agreed upon. Another aspect of chronic disease management is the patient’s engagement and compliance with the physician’s advice and medications. AI-powered conversationalists can provide interactive personalized education, diet recommendations, and therapy reminders.

AI to aid rehabilitation

But AI’s role does not stop there. In rehabilitation, AI assists patients in their recovery journey. Sandal et al.[9] developed the selfBACK app using case-based reasoning AI that provided personalized self-management recommendations for lower back pain. The system learned from successful previous cases to suggest suitable plans and resulted in significantly reduced back pain disability at 3 months compared to usual care (P = 0.03). Burns et al.[10] created a smartwatch system using convolutional neural networks to detect and monitor physiotherapy exercises with 90%–95% accuracy. The system helped establish a clear dose-response relationship between physiotherapy participation and pain outcomes in patients with rotator cuff disorders. From monitoring compliance with prescribed exercises to analyzing gait patterns, AI can ensure that patients stay on track and maximize their chances of a full recovery.

Medical records

Behind the scenes, administrative applications streamline the workload for healthcare professionals. AI automates tasks like managing electronic health records, data entry, and reviewing lab results, freeing up time for doctors and nurses to focus on what they do best—caring for patients. Public health at large scale can be managed by correctly integrating these technologies. Dalakoti et al.[11] utilized two main AI systems at Singapore’s National University Health System: CardioSight, a real-time dashboard using Endeavour AI for geographic cardiovascular risk monitoring, and chronic disease management program (CHAMP), which employs large language models and clinical decision support systems to automate patient communications and provide treatment recommendations through WhatsApp chatbots. The system successfully identifies high-risk cardiovascular patients and delivers targeted preventive care interventions without increasing healthcare worker burden.

Medical Research and Drug Discovery

Biotechnology

Meanwhile, in the realm of Medical Research and Drug Discovery, AI is accelerating innovation. By analyzing vast datasets, AI identifies potential drug candidates and optimizes clinical trials, bringing life-saving treatments to patients faster than ever before. DeepVS by Pereira et al.[12] developed a docking system for 40 receptors and 2950 ligands, efficiently screening 95,000 decoys. This AI-driven approach accelerated virtual screening and reduced time and cost in identifying potential drug candidates. A very well-known example, AlphaFold by DeepMind, predicts 3D protein structures with high accuracy and aids in structure-based drug design. This breakthrough has enabled faster identification of drug targets and reduced reliance on experimental methods.[13]

Research and publication

AI has revolutionized various aspects of scientific publishing. AI-powered tools are being used for literature searches and information retrieval, allowing researchers to quickly find relevant studies and data. In manuscript preparation, AI assists with grammar checks, citation management, and even generating abstracts or summaries. For publishers and editors, AI algorithms help detect plagiarism, assess the quality of statistical analyses, and identify potential conflicts of interest. Machine learning models are also being employed to streamline the peer review process by matching manuscripts with appropriate reviewers and providing preliminary assessments of methodological soundness.[14]

However, the integration of AI in scientific publishing also raises ethical concerns and challenges. There are ongoing debates about the appropriate disclosure of AI use in manuscript preparation and the potential for AI to generate fabricated or manipulated content. Publishers and ethics committees are grappling with issues such as authorship attribution for AI-generated text and the detection of AI-produced fraudulent papers or “paper mills.” As AI technologies continue to advance, there is a growing need for clear guidelines and ethical frameworks for the responsible use of AI in scientific publishing. The scientific community must strike a balance between harnessing the benefits of AI and preserving human oversight and accountability in the publication process.

Having said that, it comes with its own set of challenges during implementation.

Inappropriate Use of AI in Medicine

A major concern with the use of AI in medicine is algorithmic bias, where AI systems may perpetuate or amplify existing biases and health disparities. For example, an AI system used to predict healthcare needs was found to exhibit significant racial bias, underestimating the healthcare needs of Black patients compared to White patients with similar conditions. This can potentially lead to certain groups receiving inadequate care or resources. AI systems trained on datasets that lack diversity may perform poorly for underrepresented populations. There are also concerns about AI systems being deployed prematurely without sufficient testing and validation across diverse populations.[15]

The “black box” nature of many AI algorithms in medicine raises ethical issues around transparency and explainability. When AI systems make clinical recommendations or predictions, it can be difficult for clinicians to understand the reasoning behind them or verify their accuracy.[16] This lack of interpretability makes it challenging to detect errors or biases. There are also concerns about over-reliance on AI systems, where clinicians may defer to AI recommendations without critically evaluating them. This automation bias could lead to errors if the AI system makes mistakes. The use of AI may negatively impact the doctor-patient relationship if it reduces face-to-face interactions.

Ethical Issues

Privacy and data security are major concerns with the increased use of AI in healthcare. AI systems require access to large amounts of sensitive patient data, raising the risk of data breaches or misuse. There are also issues around patient consent and autonomy—patients may not be fully aware of how their data is being used to train AI systems.[17] The commercialization of AI in healthcare by large technology companies has raised concerns about the commodification of patient data. In fact, the use of AI for purposes like predictive risk scoring could potentially be used in discriminatory ways, such as denying care or insurance coverage. Careful governance and oversight are needed to ensure AI is deployed ethically in healthcare settings.

Challenges and Opportunities

While AI could help bridge the urban-rural healthcare divide, the country’s infrastructure limitations and inconsistent medical record-keeping practices present substantial hurdles.[18] Most critically, the lack of standardized electronic medical records (EMRs) across healthcare systems in India means that AI models trained on data from Western countries may not be fully applicable to the Indian population and may lead to under- or over-diagnosis, incorrect dosing, and many such incongruities. Thus, there is a pressing need to develop AI models using Indian-specific healthcare data, tailoring them to the local context.

The integration of AI into the Indian healthcare system will require addressing issues such as data privacy, security, and accountability. Nonetheless, with targeted investment and collaboration between developers, healthcare providers, and policymakers, AI could become a powerful tool for improving the delivery and accessibility of healthcare in India.

AI will—no doubt—transform the landscape, and still, there will be areas where its impact must be carefully managed. One of the most pressing concern is accuracy—especially with generative models hallucinating the generated content, a particularly sensitive concern in research.[19] However, these models can greatly assist with literature review, data analysis, and manuscript preparation.

Accountability is another concern—especially where the application of AI is in decision-making processes.[20] The use of AI will require clear guidelines to prevent misuse, including over-reliance on AI that may narrow the scope of inquiry or lead to biased outcomes.

In some use cases, AI is already surpassing human experts—e.g. in diagnosis based on symptoms.[21] This can lead to work-related anxiety in professionals. We may need to rethink the role of humans, especially in activities that can now be automated to an ever-increasing extent. Ten ethical principles to guide the use of AI in healthcare have already been issued by the ICMR, which will be very important for clinical research in India.[22]

The Way Forward

For someone who is well-versed with the use of state-of-the-art technology, it is fairly easy to reach the limits of what AI can do. Until the development of artificial general intelligence (AGI)—which is still a debatable topic—we are sure to run into these edge cases where there are diminishing returns from the use of AI. Such boundaries are where human interventions and human expertise may be best placed.

The path to integrating AI into healthcare and research workflows must be approached with caution. Pilot projects should be prioritized in areas where AI’s impact can be immediately realized, followed by gradual scaling. There is a need to establish clear guidelines for AI use in healthcare. This, along with the rollout of digital health frameworks such as the Ayushman Bharat Digital Mission (ABDM), may propel India toward faster and safer adoption of these technologies.

AI should not be seen as a replacement for human expertise but rather as a complement to it, augmenting clinical decision-making and research efficiency.

To conclude, the future of healthcare in India lies in the careful, thoughtful, and contextually tailored implementation of AI technologies. With the right frameworks in place, AI could become an invaluable tool in advancing the nation’s healthcare infrastructure, improving health outcomes, and addressing long-standing challenges. Coming back to the “Turing Test,” if we are expecting the machines to be “sentient,” an important question still remains to be answered: “Will machines have consciousness that is higher than being sentient?”

Conflicts of interest

The views expressed in this article are solely those of the authors and do not represent the views or positions of Abbott.

Funding Statement

Nil.