AI in Healthcare
Author
Artjom Gavryshev
Editor
Aya Samaha
Introduction
AI has revolutionised early disease detection, particularly in medical imaging and radiology, where its application proves highly useful. AI algorithms analyse radiological scans at a speed and accuracy unparalleled by human radiologists. Yet, to promote AI in healthcare accordingly we must ask what are the steps for making AI and humans compatible in this life-saving field? Is the challenge primarily about medical accountability, AI literacy among healthcare providers, or fostering public trust and social acceptance of AI-driven decisions in clinical settings?
AI in Disease Detection
A striking example of the transformative potential of AI in the medical field is its application in mammogram analysis, which can process images up to 30 times faster than conventional methods while achieving an accuracy rate of 99%. Embedding AI detection significantly shrinks false positives and unnecessary biopsies, alleviating psychological distress for patients and optimizing clinical workflow.
AI is also helping address declining R&D productivity in drug discovery. Currently, it costs $2.6 billion and takes between 10 to 15 years on average to introduce a new drug, and only 12% of new molecular entities in clinical trials receive FDA approval. AI can accelerate the process by predicting how different molecules are going to interact with one another, pushing the research ahead. At an early stage of drug discovery, a ‘target’ must be identified–a protein or a gene associated with the disease. Subsequently, researchers search for a molecule with the potential to constrain or enhance the activity of the identified target. The search occurs within an immense ‘chemical space’, which refers to the vast number of possible molecules that could be synthesized and tested for therapeutic potential–approximately 10^60 to 10^100 organic molecules with a molecular weight under 500 Daltons–a number way beyond the number of stars in the observable universe. AI models, combined with high-performance computing (HPC), can screen up to billions of molecules within hours or days. In a study by Insilico Medicine, AI designed a novel drug molecule in 21 days, a process that traditionally takes years. Likewise, DeepMind’s AlphaFold database mapped 200 million proteins, reducing research timelines from years to weeks. In drug discovery, AI can act as a compass capable of steering research teams in the right direction.
Reducing Time
Medical errors, which result in life-threatening consequences such as delays in treatment, are the third-leading cause of death in the U.S., accounting for approximately 250,000 deaths annually. The World Health Organization estimates that sepsis—a condition that can result from delayed treatment of infections—affects 30 million people worldwide annually, potentially leading to 6 million deaths. By reducing the time it takes for patients to receive treatment, AI can potentially save numerous lives, and shrink this number in the future.
According to Cancer Research UK, the waiting time from obtaining an urgent referral for a cancer biopsy to the beginning of the treatment is 62 days. A South Korean study demonstrates AI models exhibiting a 91% sensitivity in breast cancer detection compared to 74% for radiologists. With the help of AI-detection programs, like iCAD’s ProFound, for mammography approved by the U.S. Food and Drug Administration, hospitals can detect cancer earlier. Beyond oncology, Brainomix’s e-Stroke software, developed at Oxford University, has significantly reduced the time from hospital admission to stroke treatment by over an hour, thereby tripling the proportion of patients who achieve functional independence—from 16% to 48%.
AI as a Money Saver
Drug development costs have increased exponentially. Consequently, each breakthrough demands significantly greater financial effort. AI reduces redundant trial phases, accelerates data processing and facilitates drug repurposing—identifying new applications for existing compounds—which is more cost-effective than developing novel drugs from scratch.
Exorbitant drug development prices have reinforced the dominance of a few pharmaceutical conglomerates. These titans dominate global drug markets and control intellectual property (IP) rights, clinical trial funding, and market access, making drug prices high and limiting accessibility, particularly for life-saving treatments such as insulin and cancer drugs. AI could be a solution to such artificially elevated market barriers. The success of AlphaFold in predicting protein structures shows a promising future for a more equal playing field in pharma. AI-driven computational biology tools allow smaller research teams to conduct high-throughput screening of molecular interactions, a process that once required the financial backing of pharmaceutical giants. Cloud-based AI platforms, such as Atomwise, BenevolentAI, and Insilico Medicine, now provide computational power and predictive modelling capabilities to startups and academic institutions at a fraction of traditional costs, leveling the playing field in the sector and challenging the pharma oligopoly.
Similar to drug discovery, AI-driven diagnostic tools democratize healthcare by increasing efficiency and compensating for workforce and knowledge gaps. For instance, utilizing digital in-line holographic microscopy (DIHM) has improved malaria detection by enabling high-throughput screening and detection rate of malaria red blood cells.
Challenges
Despite its transformative potential, AI in healthcare raises complex ethical, and regulatory challenges. Greater attention must be allocated towards the ‘cleanness’ and representativeness of datasets AI is trained on, as well as tighter accountability frameworks, and more transparency in the sourcing of data–the skeleton of every AI model.
Privacy and data concerns are another key challenge. AI systems rely on vast amounts of patient data, including medical records, imaging scans, and genetic information. Much of this data is sensitive and must be stored and accessed appropriately to omit data leaks. AI systems are dynamic and continuously learning, raising questions about how often they should be re-evaluated, re-certified, and updated. Regulatory bodies like the FDA and EMA (European Medicines Agency) have taken steps to introduce adaptive approval mechanisms, yet no global consensus exists on how AI-driven medical tools should be governed.
Another critical issue is accountability and liability–a grey zone of the current AI landscape. Simply put, if the AI system is given a degree of decision-making autonomy, who is to blame for erroneous diagnoses or treatment recommendations that lead to inferior health conditions of the patient, or in extreme scenarios, death? Unlike traditional medical errors, AI-generated decisions often lack transparency due to the ‘black box’ nature of deep learning models.This ‘Black Box Problem’ means that the decision-making of AI is not easily interpretable by humans. Clinicians must be able to trust AI-driven recommendations, which for now, they are hesitant to do.
Steps Forward: Trust accompanied by Proper Regulation
Tackling the ‘cleanliness of datasets’ requires a diligent data curation and validation process. The bias of AI based on underrepresented datasets can be tackled by the development of synthetic datasets, helping to fill gaps where real-world data is limited or underrepresented.
Privacy is a more pressing concern worldwide. One promising solution lies in the utilisation of Blockchain technology, which provides more secure and immutable records for AI-driven medical applications. Projects like BurstIQ enable patients to control access to their medical data, while MedRec, promotes decentralised record management and leverages blockchain to enhance security in managing patient consent, medical history, and AI-generated diagnostics.
Solving the ‘Black Box Problem’ of non-transparency of AI is perhaps the biggest effort to consider when elevating the trust of professionals in AI and increasing their usage. One notable effort in this domain is the Defense Advanced Research Project Agency’s (DARPA) XAI program, which aims to make AI systems more transparent and understandable, allowing users to better understand the reasoning behind AI decisions and predictions.
Conclusion: A Brick-by-Brick Development
The integration of AI into healthcare is not an eventuality but an inevitability, and its application in healthcare will expand substantially over the next decade. These technologies can provide a ‘second pair of eyes’ in diagnosis establishment, save money in drug discovery, and speed up productivity. AI in healthcare is not going to reduce the need for humans in the loop, but make the loop more productive–and its success will depend on how well transparency, regulation, and human oversight evolve alongside it.