ARTIFICIAL INTELLIGENCE DRIVEN DIAGNOSIS IN HEALTHCARE

AI-driven diagnosis has moved from research to regulated clinical products, with the strongest evidence in imaging, particularly screening for breast cancer and diabetic retinopathy. The documented benefits are faster reading, lower workload and consistent output. The documented limits are equally clear: biased data, thin external validation, opaque reasoning, privacy exposure and unsettled liability. The available evidence supports AI as a supervised aid and does not support autonomy beyond narrow, well-validated tasks. Three longer-term questions deserve separate mention. The first concerns evidence of patient benefit. Most studies report diagnostic accuracy, which is an intermediate measure; fewer report whether patients live longer, suffer less or avoid unnecessary procedures. Over-detection is a known risk in screening, since finding more small abnormalities does not always mean helping the patient, and the same caution applies to algorithms that are very sensitive [10], [17]. The second concerns generalist models. Large multimodal systems promise a single tool across many tasks, but their reliability on any one task is harder to bound than that of a narrow classifier, and their errors can be fluent and convincing [13]. The third concerns sustainability. Maintaining a model, auditing it and retraining it costs staff time and money, and these costs fall hardest on small hospitals [5], [24]. Several directions follow from the gaps found. Multimodal models that combine images, records and text need prospective evaluation. Datasets from low- and middle-income settings are under-represented and should be built. Methods for interpretable and calibrated output deserve more attention than post-hoc explanation. Legal frameworks for liability need to catch up with practice. Because this paper is a review without original experiments, its conclusions are limited by the studies it cites, and the 2025 regulatory figures should be rechecked against the sources before reuse.

[1]    E. J. Topol, Deep Medicine: How Artificial Intelligence Can Make Healthcare Human Again. New York, NY, USA: Basic Books, 2019. [2]    A. Esteva et al., "A guide to deep learning in healthcare," Nat. Med., vol. 25, pp. 24–29, 2019. [3]    P. Rajpurkar et al., "CheXNet: Radiologist-level pneumonia detection on chest X-rays with deep learning," arXiv:1711.05225, 2017. [4]    Y. Liu, P.-H. C. Chen, J. Krause, and L. Peng, "How to read articles that use machine learning: Users' guides to the medical literature," JAMA, vol. 322, no. 18, pp. 1806–1816, 2019. [5]    K.-H. Yu, A. L. Beam, and I. S. Kohane, "Artificial intelligence in healthcare," Nat. Biomed. Eng., vol. 2, pp. 719–731, 2018. [6]    H. Singh, A. N. D. Meyer, and E. J. Thomas, "The frequency of diagnostic errors in outpatient care: Estimations from three large observational studies involving US adult populations," BMJ Qual. Saf., vol. 23, no. 9, pp. 727–731, 2014. [7]    V. Gulshan et al., "Development and validation of a deep learning algorithm for detection of diabetic retinopathy in retinal fundus photographs," JAMA, vol. 316, no. 22, pp. 2402–2410, 2016. [8]    A. Esteva et al., "Dermatologist-level classification of skin cancer with deep neural networks," Nature, vol. 542, pp. 115–118, 2017. [9]    S. M. McKinney et al., "International evaluation of an AI system for breast cancer screening," Nature, vol. 577, pp. 89–94, 2020. [10]  K. Lång et al., "Artificial intelligence-supported screen reading versus standard double reading in the Mammography Screening with Artificial Intelligence trial (MASAI): A clinical safety analysis of a randomised, controlled, non-inferiority, single-blinded, screening accuracy study," Lancet Oncol., vol. 24, no. 8, pp. 936–944, 2023.