Artificial Intelligence and Machine Learning in Healthcare

Innovative scientific and technological developments have ushered in a exceptional transformation in medication that continues to influence just about all stakeholders from sufferers to suppliers to Healthcare Organizations (HCOs) and the group in common.1,2 Increasingly integrated into scientific apply over the previous few a long time, these improvements embody widespread use of Electronic Health Records (EHR), telemedicine, robotics, and resolution help for surgical procedures. Ingestible microchips enable healthcare suppliers to observe affected person compliance with prescribed pharmacotherapies and their therapeutic efficacy by means of massive knowledge evaluation,1–5 in addition to streamlining drug design, screening, and discovery.6 Adoption of novel medical applied sciences has allowed US healthcare to keep up its vanguard place in choose domains of scientific care comparable to enhancing entry by lowering wait occasions, enriching patient-provider communication, enhancing diagnostic accuracy, enhancing affected person satisfaction, augmenting end result prediction, reducing mortality, and extending life expectancy.3–5,7
Yet regardless of the theoretical benefits of those progressive medical applied sciences, many points stay requiring cautious consideration as we combine these novel applied sciences into our armamentarium. This descriptive literature-based article explicates on the benefits, future potential, challenges, and caveats with the predictable and impending importation of AI and ML into all aspects of healthcare.
Innovative Technological Advances and Applications
By far probably the most revolutionary of those novel applied sciences is Artificial Intelligence (AI), a department of pc science that makes an attempt to assemble clever entities by way of Machine Learning (ML), which is the “capacity of computer systems to study with out being explicitly programed”.8 ML makes use of algorithms to determine patterns, and its subspecialty Deep Learning (DL) employs synthetic neural networks with intervening frameworks to determine patterns and knowledge.1,8 Although ML was first conceived by pc scientist Arthur Samuel way back to 1956, functions of AI have solely just lately begun to pervade our each day life with computer systems simulating human cognition—eg, visible notion, speech recognition, decision-making, and language translation.8 Everyday examples of AI embody good telephones, autonomous autos, digital assistants (eg, Siri, Alexa), chatbots and auto-correcting software program, on-line banking, facial recognition, and transportation (eg, Uber, air visitors management operations, and so on.). The iterative nature of ML permits the machine to adapt its methods and outputs following publicity to new knowledge with supervised studying—ie, using coaching algorithms to foretell future occasions from historic knowledge inputs—or unsupervised studying, whereby the machine explores the information and makes an attempt to develop patterns or constructions de novo. The latter methodology is usually used to find out and distinguish outliers. Neural networks in AI make the most of an adaptive system comprised of an interconnected group of synthetic neurons and mathematical or computational modeling for processing data from enter and output knowledge by way of sample recognition.9 Through predictive analytics, ML has demonstrated its effectiveness in the realm of finance (eg, figuring out bank card fraud) and in the retail business to anticipate buyer habits.1,10,11
Extrapolation of AI to medication and healthcare is predicted to extend exponentially in the three principal domains of analysis, instructing, and scientific care. With improved computational efficiencies, widespread functions of ML in healthcare will embody enhanced diagnostic modalities, improved therapeutic interventions, augmenting and refining workflow by processing massive quantities of hospital and nationwide EHR knowledge, extra correct scientific course and prediction by means of precision and personalised medication, and genome interpretation. ML can present fundamental scientific triage in geographical areas inaccessible to specialty care. It can even detect treatable psychiatric circumstances by way of evaluation of affective and nervousness problems utilizing speech patterns and facial expressions (eg, bipolar dysfunction, main despair, nervousness spectrum and psychotic problems, consideration deficit hyperactivity dysfunction, dependancy problems, Tourette’s Syndrome, and so on.)12,13 (Figure 1). Deep studying algorithms are extremely efficient in comparison with human interpretation in medical subspecialties the place sample recognition performs a dominant function, comparable to dermatology, hematology, oncology, histopathology, ophthalmology, radiology (eg, programmed picture analyses), and neurology (eg, evaluation for seizures using electroencephalography). Artificial neural networks are being developed and employed for diagnostic accuracy, well timed interventions, outcomes and prognostication of neurosurgical circumstances, comparable to spinal stenosis, traumatic mind damage, mind tumors, and cerebral vasospasm following aneurysmal subarachnoid hemorrhage.14 Theoretically, ML can enhance triage by directing sufferers to correct remedies at decrease price and by retaining these with power circumstances out of pricey and time-intensive emergency care facilities. In scientific apply, ~5% of all sufferers account for 50% of healthcare prices, and these with power medical circumstances comprise 85% of whole US healthcare prices.3
Figure 1 Potential Applications of Machine Learning.

Patients can profit from ML in different methods. For follow-up visits, not having to rearrange transportation or take day without work work for face-to-face interplay with healthcare suppliers could also be a lovely various to sufferers and to the group, much more so in restricted circumstances just like the current COVID-19 pandemic-associated lockdowns and social distancing.
Ongoing ML-related analysis and its functions are sturdy. Companies growing automation, topological knowledge evaluation, genetic mapping, and communications methods embody Pathway Genomics, Digital Reasoning Systems, Ayandi, Apixio, Butterfly Network, Benevolent AI, Flatiron Health, and a number of others.1,10
Caveats and Challenges
Despite the numerous theoretical benefits and potential advantages of ML in healthcare, a number of challenges (Figure 2) have to be met15 earlier than it could possibly obtain broader acceptance and utility.
Figure 2 Caveats and Challenges with use of Machine Learning.

Data Acquisition and Validation
Frequent software program updates will likely be obligatory to make sure continued enchancment in ML-assisted fashions over time. Encouraging using such software program, the Food and Drug Administration has advisable a “pre-certified strategy” for agility.1,2 To be of pragmatic scientific import, high-quality input-data is paramount for validating and refining diagnostic and therapeutic procedures. At current, nevertheless, there’s a dearth of strong comparative knowledge that may be validated towards the generally accepted gold normal, comprised of blinded, placebo-controlled randomized scientific trials versus the ML-output knowledge that’s usually an “area-under-the-curve” evaluation.1,7 Clinical knowledge generated from ML-assisted calculations and extra rigorous multi-variate evaluation will entail integration with different related affected person demographic data (eg, socio-economic standing, together with values, social and cultural norms, religion and perception methods, social help constructions in-situ, and so on.).16
Paradigm Shift in Patient Care
All stakeholders in the healthcare supply system (HCOs, suppliers, sufferers, and the group) must regulate to the paradigm shift away from conventional in-person interactions. Healthcare suppliers must surmount precise or perceived added workload to keep away from “burnout” particularly in the course of the preliminary adaptive section. They will even have to deal with elevated ML-generated false-positive and -negative alerts. The conventional apply of scientific medication is deeply entrenched in the framework of formulating a scientific speculation by way of rigorous history-taking and bodily examination adopted by sequential affirmation by means of even handed ancillary and diagnostic testing. Such conventional in-person interactions have underscored the significance of an empathetic strategy to the provider-patient relationship. This conventional view has been characterised as “archaic”, significantly by these with a “futuristic” mindset, who envision an evolutionary change resulting in “entire physique scans” that ship a extra correct evaluation of well being and prognosis of illness. However, incidental findings not attributable to signs might result in extreme ancillary exams underscoring the adage “testing begets extra testing”.17
Cost-Benefit and Value Proposition
Healthcare is likely one of the quickest rising segments of the world economic system and is presently at a crossroads of unprecedented transformation. As an instance, US healthcare expenditure has accelerated dramatically over the previous a number of a long time (~19% of Gross National Product; exceeding $4.1 trillion, or $12,500 per individual per 12 months)18 with widespread ramifications for all stakeholders together with sufferers and their households, healthcare suppliers, authorities, group, and the US economic system.1,3–5 A paradigm shift from “volume-based” to “performance-based” reimbursements from third-party payers warrants deal with a number of the most pressing points in healthcare together with price containment, entry, and offering “low-cost, high-value” healthcare commensurate with the proposed six-domain framework (“secure, efficient, patient-centered, well timed, environment friendly, and equitable”) articulated by the Institute of Medicine in 2001.3–5,19 Of word, uncontrolled use of pricy expertise and extreme ancillary testing account for ~25–30% of whole healthcare prices.17 While applied sciences will most likely by no means fully substitute the operate of healthcare suppliers, they are going to undoubtedly remodel healthcare, benefiting each suppliers and sufferers. However, there’s a paucity of price–profit knowledge and evaluation of using these progressive rising medical applied sciences. All stakeholders ought to stay cost-conscious because the newer technological diagnostic approaches might additional drive up the already rising prices of healthcare. Educating and coaching the subsequent era of healthcare suppliers in the context of AI will even require transformation with simulation approaches and inter-professional schooling. Therefore, the worth proposition of novel applied sciences have to be critically appraised by way of longitudinal and steady valuations and affected person outcomes in phrases of its influence on well being and illness administration.13 To mitigate healthcare prices, we should management the “technological crucial”—the overuse of expertise due to straightforward availability with out due consideration to illness course or outcomes and no matter price–profit ratio.3
Data Ownership and Integrity
Issues surrounding shopper privateness and proprietorship of colossal portions of healthcare knowledge underneath an AI regime are legit issues. Malicious or unintentional breaches might end result in monetary or different hurt. Akin to the challenges encountered with EHR, easy accessibility to knowledge and interoperability with broader compatibility of interfaces by healthcare suppliers unfold throughout area and time will current distinctive challenges. Databases will seemingly be owned by massive profit-oriented expertise corporations who might resolve to dispense knowledge to 3rd events. Additional prices are predictable as effectively, significantly in the course of the early phases of improvement of ML algorithms, which is more likely to be extra bearable to massive HCOs. Delay in using such processes is anticipated by smaller organizations with ensuing potential for mergers and acquisitions and even failure of smaller hospitals and clinics. Concerns relating to possession, duty, and accountability of ML algorithms might come up owing to the likelihood of detrimental outcomes, which ideally ought to be apportioned between developer, interpreter, healthcare supplier, and affected person.1 Simulation strategies will be preemptively utilized for ML coaching for scientific situations; apply runs might require formal certification programs and workshops. Regulations have to be developed by policymakers and legislative our bodies to delineate the function of third-party payers in ML-assisted healthcare financing. Finally, schooling and coaching by way of media retailers, web, and social media will likely be obligatory to deal with public opinion, misperceptions, and naïve expectations about ML-assisted algorithms.7
Conclusions and Future Directions
For centuries, the apply of drugs has been deeply embedded in a practice of meticulous history-taking, bodily examination, and considerate ancillary investigations to verify scientific hypotheses and diagnoses. The nice doctor, Sir William Osler (1849–1919)14,20 encapsulated the specified apply of excellent medication together with his well-known quotes, “Listen to your affected person – he’s telling you the prognosis”, “The good doctor treats the illness; the nice doctor treats the affected person who has the illness”, and “Medicine is a science of uncertainty and an artwork of likelihood”. With speedy technological advances, we’re on the crossroads of training medication that will be distinctly completely different from the normal strategy and apply(s), a change which may be characterised as “evolutionary”.
AI and ML have huge potential to rework healthcare and the apply of drugs, though these modalities won’t ever substitute an astute and empathetic bedside clinician. Furthermore, a number of points stay as as to if their worth proposition and cost-benefit are complementary to the overarching deal with offering “low-cost, high-value” healthcare to the group at massive. While progressive technological advances play a essential function in the speedy prognosis and administration of illness, the phenomenon of the technological imperative3–5,17 deserves particular consideration amongst each public and suppliers for the long run use of AI and ML in delivering healthcare.

Disclosure
The creator studies no conflicts of curiosity in this work.
References
1. Bhardwaj R, Nambiar AR, Dutta D A Study of Machine Learning in Healthcare. 2017 IEEE forty first Annual Computer Software and Applications Conference. 236–241. Available from: https://ieeexplore.ieee.org/stamp/stamp.jsp?arnumber=8029924. Accessed March 30, 2022.
2. Deo RC. Machine Learning in Medicine. Circulation. 2015;132:1920–1930. doi:10.1161/CIRCULATIONAHA.115.001593
3. Shi L, Singh DA. Delivering Health Care in America: A Systems Approach. seventh ed. Burlington, MA: Jones & Bartlett Learning; 2019.
4. Barr DA. Introduction to US Health Policy. The Organization, Financing, and Delivery of Health Care in America. 4th ed. Baltimore, MD: John Hopkins University Press; 2016.
5. Wilensky SE, Teitelbaum JB. Essentials of Health Policy and Law. Fourth ed. Burlington, MA: Jones & Bartlett Learning; 2020.
6. Gupta R, Srivastava D, Sahu M, Tiwan S, Ambasta RK, Kumar P. Artificial intelligence to deep studying; machine intelligence strategy for drug discovery. Mol Divers. 2021;25:1315–1360. doi:10.1007/s11030-021-10217-3
7. Dabi A, Taylor AJ. Machine Learning, Ethics and Brain Death Concepts and Framework. Arch Neurol Neurol Disord. 2020;3:1–9.
8. Handelman GS, Kok HK, Chandra RV, Razavi AH, Lee MJ, Asadi H. eDoctor: machine studying and the way forward for medication. J Int Med. 2018;284:603–619. doi:10.1111/joim.12822
9. Hopfield JJ. Neural networks and bodily methods with emergent collective computational talents. Proc Natl Acad Sci U S A. 1982;79:2554–2558. doi:10.1073/pnas.79.8.2554
10. Ghassemi M, Naumann T, Schulam P, Beam AL, Ranganath R Opportunities in Machine Learning for Healthcare. 2018. Available from: https://pdfs.semanticscholar.org/1e0b/f0543d2f3def3e34c51bd40abb22a05937bc.pdf. Accessed March 30, 2022.
11. Jnr YA Artificial Intelligence and Healthcare: a Qualitative Review of Recent Advances and Predictions for the Future. Available from: https://pimr.org.in/2019-vol7-issue-3/YawAnsongJnr_v3.pdf. Accessed March 30, 2022.
12. Chandler C, Foltz PW, Elvevag B. Using machine studying in Psychiatry; the necessity to set up a Framework that nurtures trustworthiness. Schizophr Bull. 2019;46:11–14.
13. Ray A, Bhardwaj A, Malik YK, Singh S, Gupta R. Artificial intelligence and Psychiatry: an outline. Asian J Psychiatr. 2022;70:103021. doi:10.1016/j.ajp.2022.103021
14. Ganapathy Ok Artificial intelligence in neurosciences-are we actually there? Available from: https://www.sciencedirect.com/science/article/pii/B9780323900379000084. Accessed June 10, 2022.
15. Sunarti S, Rahman FF, Naufal M, Risky M, Febriyanto Ok, Mashina R. Artificial intelligence in healthcare: alternatives and threat for future. Gac Sinat. 2012;35(S1):S67–S70. doi:10.1016/j.gaceta.2020.12.019.
16. Yu B, Beam A, Kohane I. Artificial Intelligence in Healthcare. Nature Biomed Eng. 2018;2:719–731. doi:10.1038/s41551-018-0305-z
17. Bhardwaj A. Excessive Ancillary Testing by Healthcare Providers: causes and Proposed Solutions. J Hospital Med Management. 2019;5(1):1–6.
18. Fact Sheet NHE. Centers for Medicare and Medicaid Services. Available from: https://www.cms.gov/Research-Statistics-Data-and-Systems/Statistics-Trends-and-Reports/NationalHealthExpendData/NHE-Fact-Sheet. Accessed April 14, 2022.
19. Institute of Medicine (IOM). Crossing the Quality Chasm: A New Health System for the twenty first Century. Washington, D.C: National Academy Press; 2001.
20. Bliss M. William Osler: A Life in Medicine. New York, NY: Oxford University Press; 1999.

https://www.dovepress.com/promise-and-provisos-of-artificial-intelligence-and-machine-learning-i-peer-reviewed-fulltext-article-JHL

Recommended For You