Artificial intelligence (AI) can be broadly applied to transform and improve patient care. Advancements in AI in healthcare settings are in part due to improvements in learning algorithms and the availability of big data, in addition to new insights in the biological understanding of diseases. AI can play a substantial role in improving patient outcomes and reducing treatment costs, for instance by contributing to quicker and more accurate diagnoses, and a reduction in the battery of testing that patients often endure to reach a diagnosis. Thus, the use of AI to assist healthcare professionals in finding an accurate and timely diagnosis is of great interest.
AI has the capacity to improve several aspects of medical diagnosis. First, by shortening the time to diagnosis. Rare diseases remain challenging for healthcare teams to identify and address; patients wait on average six years from symptomatic onset to diagnosis. Given the association between diagnostic delay and decreased quality-of-life and a worsened symptom burden, decreasing time to diagnosis offers room to make gains from a clinical perspective. For example, learning algorithms using automatic image evaluation of lung CT scans have been used to predict the risk of lung cancer in patients prior to symptomatic display. Earlier detection and reducing the tests that are requisite for a diagnosis will help to quicken the process and reduce the financial burden to patients and the healthcare system.
A second promising use of AI in the diagnostic process is in providing a more specific diagnosis for patients. In oncology, where there are thousands of cancer subtypes, accurate identification of a pathological specimen can help inform decisions related to disease prognosis and treatment plans. Training of AI models have allowed for successful detection of skin cancer in patients at rates similar to dermatologists without the use of an invasive biopsy. In instances of diseases such as melanoma, accurate diagnosis of malignancy is pivotal to improving patient prognosis. This can support the decision-making process for both patients and clinicians.
Finally, AI can improve the diagnosis process by acting as a second opinion for physicians. Rates of physician burnout have been on the rise and amplified by the COVID-19 pandemic. AI offers an opportunity to establish complex patient diagnoses and assist with time-consuming administrative tasks.
While AI is an excellent supplement to the diagnostic process, its limitations must be acknowledged. Clinical and algorithmic bias have been reported where differences in error rates for race and gender, amongst other factors, may be introduced to a model. Moreover, the quality of the algorithm and predictions made are only as strong, unbiased, and population-representative as the dataset used for training and validation. Issues of data reproducibility have highlighted a need for transparency related to the use and publication of AI-related data in healthcare. Therefore, AI is ultimately dependent on the availability of high-quality data and the ability of those applying it to be meticulous in its use.
The advantages of the use of AI in medicine and medical diagnosis are evident, yet accompanied by many limitations and must be examined critically. The process of medical diagnosis can be complex and emotional; while pathological details can be improved by various algorithms and training sets, the emotional component is provided by the healthcare team. Certainly faster and more specific diagnosis have been implicated in improved quality-of-life, however significant associations between the patient-physician relationship and improved quality-of-life have been identified and highlight the importance of patient-centered care.
AI is becoming an indispensable tool in healthcare and can be applied to various aspects of patient care. On one hand, AI can improve the process of diagnosis by contributing to fast and accurate diagnoses for patients. On the other, AI and the application of learning algorithms can show substantial biases, require high-quality data, and cannot replace the emotional component of medical diagnosis provided by the healthcare team. Overall, AI is an invaluable tool in the clinic; it can help to provide patients with the highest-quality care when used correctly, and it is the duty of the physician to use it responsibly.
- Blöß S, Klemann C, Rother AK, et al. 2017. Diagnostic needs for rare diseases and shared prediagnostic phenomena: Results of a German-wide expert Delphi survey. PLoS ONE 12(2):e0172532.
- Robinson KM, Christensen KB, Ottesen B, Krasnik A. 2012. Diagnostic delay, quality of life and patient satisfaction among women diagnosed with endometrial or ovarian cancer: a nationwide Danish study. Qual Life Res Nov;21(9):1519-25.
- Ardila D, Kiraly AP, Bharadwaj S. et al. 2019. End-to-end lung cancer screening with three-dimensional deep learning on low-dose chest computed tomography. Nat Med 25, 954–961.
- Esteva A, Kuprel B, Novoa R et al. 2017. Dermatologist-level classification of skin cancer with deep neural networks. Nature 542, 115–118.
- Sondermann W, Zimmer L, Schadendorf D, et al. 2016. Initial misdiagnosis of melanoma located on the foot is associated with poorer prognosis. Medicine95(29):e4332.
- Khan N, Palepu A, Dodek P, et al. 2021. Cross-sectional survey on physician burnout during the COVID-19 pandemic in Vancouver, Canada: the role of gender, ethnicity and sexual orientation. BMJ Open 10;11(5):e050380.
- Chen IY, Szolovitz P and Ghassemi M. 2019. Can AI Help Reduce Disparities in General Medical and Mental Health Care? AMA J Ethics21(2):E167-179.
- McDermott MBA, Wang S, Marinsek N et al. 2021. Reproducibility in Machine Learning for Health Research: Still a Ways to Go. Sci Transl Med 13(586):eabb1655.
- Hulme K, Chilcot J, Smith MA. 2018. Doctor-patient relationship and quality of life in Irritable Bowel Syndrome: an exploratory study of the potential mediating role of illness perceptions and acceptance. Psychol Health Med. 23(6):674-684.