Artificial intelligence (AI) is gradually transforming different aspects of healthcare, including patient adherence and engagement, administrative activities, and disease diagnosis and treatment recommendations. Moreover, the recent advancements in machine learning (ML), computing infrastructure, and digitized data acquisition have enabled the application of AI techniques in areas currently dominated by human experts. The article discusses the increasing importance and growing applications of AI in healthcare.
Image credit: raker/Shutterstock
Importance of AI in Healthcare
AI is increasingly becoming an important tool in the healthcare sector because it can perform different tasks more quickly, accurately, and efficiently at a lower cost. AI can also improve operational efficiency, connect fragmented healthcare data, and offer user-centric experiences.
Healthcare providers, pharmaceutical firms, and life science companies are already using multiple types of AI, including rule-based expert systems, natural language processing (NLP), and ML, for different applications.
These include automated experiments and data collection, biomarker discovery, drug target prioritization/repurposing/discovery, literature mining, molecular dynamics simulation, gene function and genetic variant annotation, chemical toxicity prediction, patient monitoring, patient risk stratification for prevention, patient genome interpretation, and treatment selection.
AI enables medical professionals to improve their understanding of the daily patterns and needs of the patients under their care. Thus, professionals can provide better guidance, support, and feedback to patients using AI to ensure positive outcomes. Several diseases, such as cancer, can be detected more accurately in their early stages using AI.
For instance, AI can significantly accelerate the review and translation of mammograms with extremely high accuracy, eliminating unnecessary biopsies. Similarly, AI, coupled with consumer wearables, can oversee heart diseases at early/treatable stages, enabling doctors/caregivers to better detect, monitor, and prevent potentially life-threatening events.
Big health data must be aligned with timely and appropriate decisions to improve patient care. Predictive analytics can support clinical actions and decision-making and prioritize administrative tasks. AI can also be used for pattern recognition to identify individuals at risk of developing/deteriorating a condition due to genomic, environmental, or lifestyle factors.
Clinicians can adopt a more comprehensive disease management approach, efficiently coordinate care plans, and assist patients to comply better and manage their long-term treatment programs using AI. Drug discovery and research is one of the recent applications of AI in healthcare. The latest advances in AI can significantly reduce costs and time-to-market for new drugs by streamlining the drug repurposing and drug discovery processes.
Moreover, AI can be used with advanced humanoid robots to significantly improve end-of-life care by reducing the need for hospitalization and enabling people to live independently. Surgical robots with embedded AI capabilities can improve the ability of surgeons to perform different surgical procedures, such as minimally invasive and precise incisions and stitching wounds. Surgical robotic surgery procedures include prostate, head and neck, and gynecologic surgery.
AI Techniques in Healthcare
Traditional ML is used extensively for precision medicine in which the most effective treatment protocols for a patient are predicted based on the treatment context and different patient attributes. Neural networks are utilized for categorization applications, such as determining the probability of a patient acquiring a specific disease.
Deep learning (DL) models/neural network models with several levels of variables/features that predict outcomes are used to identify potentially cancerous lesions in radiology images. The combination of DL and radiomics can lead to greater accuracy in diagnosis compared to previous-generation automated tools used for image analysis. Recently, the Food and Drug Administration (FDA) approved a DL system for cardiovascular disease diagnosis using magnetic resonance imaging (MRI) images.
In healthcare, NLP is primarily employed for creating, classifying, and understanding published research and clinical documentation. NLP systems can prepare reports, analyze unstructured clinical notes on patients, conduct conversational AI, and transcribe patient interactions.
Rule-based expert systems are used extensively for clinical decision support. However, they are gradually being replaced with ML algorithms and data-based approaches, as changing the rules with changes in the knowledge domain is extremely difficult and time-consuming.
Population health ML models are employed to predict populations at risk of accidents/specific diseases effectively or to predict hospital readmission. Chatbots have been used for patient interaction, telehealth, and mental health and wellness. These NLP-based applications can successfully perform simple transactions such as making appointments and refilling prescriptions. However, patients are concerned about the security of confidential data and poor usability while using chatbots. In claims and payment administration, ML can be utilized for probabilistic data matching across different databases.
Transfer learning is highly effective for the classification of medical images where the number of images is in the range of thousands to tens of thousands, as the technique can significantly reduce the number of training samples typically required to train a neural network with millions of parameters.
Convolutional neural networks are increasingly playing a crucial role in healthcare. For instance, convolutional neural networks trained using 129,450 clinical images attained dermatologist-level accuracy while diagnosing skin malignancy. Similarly, convolutional neural network models trained using 128,175 retinal images demonstrated a performance comparable to that of ophthalmologists while identifying diabetic macular edema and referable diabetic retinopathy (DR).
Moreover, the DL algorithm also revealed unrecognized associations between retinal image patterns and gender, age, smoking status, major adverse cardiac events, and the systolic blood pressure of an individual.
Deep convolutional neural networks can detect prostate cancer from biopsy specimens, identify breast cancer metastasis in lymph nodes, and detect mitosis in breast cancer. For instance, ML coupled with a live-cell biomarker imaging system, can allow the risk stratification of breast and prostate cancer patients.
ML methods can be used to identify molecular patterns linked with disease subtypes and disease status, obtain omics signatures for disease phenotype prediction, and facilitate high-level interactions among measurements.
Deep neural networks can identify non-coding deoxyribonucleic acid (DNA) functions and annotate pathogenic genetic variants more efficiently compared to conventional methods, such as logistic regression. Thus, these networks can effectively diagnose complex diseases with genetic components, such as cancer.
Challenges with AI in Healthcare
Technical Challenges: Embedding AI-based effective and accurate treatment and diagnosis recommendations in electronic health record (EHR) systems, and clinical workflows can be challenging, which hinders the extensive implementation of AI.
Additionally, data that correctly represents the target patient population must be compiled to ensure accurate results, as ML-based methods primarily rely on large amounts of high-quality data. Adequate data curation is crucial while handling heterogeneous data.
Interpreting results generated by several high-performing ML models, specifically deep neural network models trained using data excluding images, remains extremely difficult for unassisted humans. Although these models deliver better-than-human performance, the outcomes of these models cannot be explained in simple terms.
Social, Legal, and Economic Challenges: Although AI can enhance the quality of care by decreasing human errors and reducing physician fatigue due to routine clinical tasks, the technology will not significantly reduce physician workload as clinical guidelines can recommend frequent examinations for at-risk patients.
Large-scale use of AI applications can potentially eliminate healthcare jobs that involve routine tasks, which can lead to job losses, reshape the existing healthcare workforce, and change the reimbursement framework in healthcare. Implementing AI applications in clinical information systems can lead to alert fatigue, additional workload for clinicians, interpersonal communication style disruption, and specific hazards that require a higher vigilance to detect.
Legal challenges involving AI primarily include medical negligence caused by complex decision support systems. A legal system is required that can provide clear guidance about the entities that will be held liable for the malpractices due to medical AI applications.
Privacy Challenges: Implementing a computing environment to collect, store, and share EHRs and sensitive health data of patients by ensuring adequate data security and privacy remains a major challenge.
Privacy-preserving methods can be adopted to ensure secure data sharing through cloud services. However, interoperable applications must be developed to meet the clinical information representation standard to implement such methods extensively.
Additionally, data privacy impact assessments can be performed to assess the privacy risks and identify the privacy-preserving technologies that must be deployed to secure patient data before deploying AI algorithms in the healthcare sector.
To summarize, AI plays a crucial role in improving clinical diagnosis and decision-making performance in several healthcare applications. However, more research is required to effectively address the challenges with AI in healthcare to exploit the full potential of this technology.
References and Further Reading
Bartoletti, I. (2019). AI in Healthcare: Ethical and Privacy Challenges. Artificial Intelligence in Medicine, 7-10. https://doi.org/10.1007/978-3-030-21642-9_2
Davenport, T., Kalakota, R. (2019). The potential for artificial intelligence in healthcare. Future Healthcare Journal, 6(2), 94-98. https://doi.org/10.7861/futurehosp.6-2-94
Yu, K., Beam, A. L., Kohane, I. S. (2018). Artificial intelligence in healthcare. Nature Biomedical Engineering, 2(10), 719-731. https://doi.org/10.1038/s41551-018-0305-z
No longer science fiction, AI and robotics are transforming healthcare. [Online] Available at https://www.pwc.com/gx/en/industries/healthcare/publications/ai-robotics-new-health/transforming-healthcare.html (Accessed on 06 August 2023)
What is artificial intelligence in healthcare? [Online] Available at https://www.ibm.com/topics/artificial-intelligence-healthcare (Accessed on 06 August 2023)