Artificial Intelligence in Healthcare: Clinical Applications, Benefits, and Ethical Challenges
Master's Thesis · ~92 pages · English
Abstract
This thesis examines the adoption of artificial intelligence in clinical healthcare, assessing where machine learning systems demonstrably improve care and where they introduce new risks. Drawing on evidence from medical imaging, electronic health record prediction, and clinical decision support, the analysis evaluates diagnostic accuracy, workflow integration, and patient outcomes alongside the governance challenges of algorithmic bias, accountability, and regulatory approval. The research applies a socio-technical framework to argue that AI value in medicine depends less on raw model performance than on validation against real clinical populations, clinician trust, and the institutional safeguards surrounding deployment. Findings indicate strong, replicated performance in narrow image-classification tasks, promising but less mature results in EHR-based prediction, and a recurring failure mode in which models that perform well in development degrade or encode bias when applied to populations unlike their training data.
1. Introduction
Artificial intelligence has moved from research prototypes to deployed clinical tools, with regulatory bodies now clearing hundreds of machine-learning-based medical devices. Proponents argue that AI can extend scarce specialist capacity, reduce diagnostic error, and surface risk earlier; critics caution that opaque models deployed without rigorous validation can entrench bias and shift accountability away from clinicians.
This thesis asks a narrower and more answerable question than whether AI will transform medicine: under what conditions does a clinical AI system actually improve patient outcomes, and what institutional and technical safeguards separate the systems that help from those that harm? It synthesizes evidence across imaging, prediction, and decision support to identify the recurring determinants of clinical value.
2. Clinical Applications and Evidence
The analysis organizes current evidence into three application classes:
Medical imaging - Deep neural networks achieve dermatologist-level and radiologist-level performance on well-scoped classification tasks (skin lesion triage, diabetic retinopathy, certain radiographs), with the strongest and most replicated results where labels are objective and training data are large.
Electronic health record prediction - Models trained on longitudinal EHR data forecast deterioration, readmission, and mortality, but performance is sensitive to site-specific coding practices and frequently drops on external validation.
Clinical decision support - Systems that embed predictions into clinician workflow show benefit only when alerts are calibrated, interpretable, and integrated without adding alarm fatigue; poorly integrated tools are ignored or override clinical judgment.
3. Key Findings, Bias, and Governance
The synthesis yields several consistent findings:
• Diagnostic performance in narrow imaging tasks is real and reproducible, but generalization to new hospitals, devices, and demographics is the dominant failure mode • Algorithmic bias is frequently a function of the training label and data, not only the model: a widely used population-health algorithm was shown to under-refer Black patients because it used healthcare cost as a proxy for need • Trust and workflow fit, not accuracy alone, determine whether clinicians act on model output • Regulatory frameworks are adapting to continuously-learning systems, raising unresolved questions about post-market monitoring and liability
The thesis recommends a governance architecture combining mandatory external validation on representative populations, transparent reporting of subgroup performance, human-in-the-loop accountability, and continuous post-deployment monitoring for performance drift.
References
- [1]Topol, E. J. (2019). High-performance medicine: The convergence of human and artificial intelligence. Nature Medicine, 25(1), 44-56.
- [2]Esteva, A., Kuprel, B., Novoa, R. A., Ko, J., Swetter, S. M., Blau, H. M., & Thrun, S. (2017). Dermatologist-level classification of skin cancer with deep neural networks. Nature, 542(7639), 115-118.
- [3]Rajkomar, A., Oren, E., Chen, K., Dai, A. M., Hajaj, N., Hardt, M., ... & Dean, J. (2018). Scalable and accurate deep learning with electronic health records. npj Digital Medicine, 1, 18.
- [4]Obermeyer, Z., Powers, B., Vogeli, C., & Mullainathan, S. (2019). Dissecting racial bias in an algorithm used to manage the health of populations. Science, 366(6464), 447-453.
- [5]Jiang, F., Jiang, Y., Zhi, H., Dong, Y., Li, H., Ma, S., ... & Wang, Y. (2017). Artificial intelligence in healthcare: Past, present and future. Stroke and Vascular Neurology, 2(4), 230-243.
This is a sample excerpt. Full papers include complete chapters, a reference list, and downloadable formats.
More Example Papers
See how OpenPaper works for your use case: OpenPaper for Researchers
Free to try · No credit card required · Free to start, 3 credits