How Can AI Help us Enhance Human Health?
How Can AI Help us Enhance Human Health?
Nan Mu1
1. Army Medical University
1. Introduction
Thanks to the fast advances of technologies particularly in the fields of the computer science and informatics, Artificial intelligence (AI) is reforming medicine. AI, including machine learning(ML), natural language processing(NLP) and deep learning(DL), can assist doctors and professionals in identifying health-related needs faster and more accurate by data-driven patterns. Currently, the deployment of AI in the medical field has reached a notable level of maturity, allowing for its widespread use across a diverse range of domains and scenarios. This signifies that medical AI harbors immense and promising possibilities, which are accompanied by formidable challenges such as intricate ethical considerations, the need for a human-centric approach, and the interpretability of highly sophisticated models.
In this paper, we begin by outlining the contemporary applications of AI in the medical field and their promising outlook. Following this, we delve into the challenges they present. Lastly, we explore the potential ecological impact of AI technologies and discuss how they could serve to enhance human health.
2. Opportunities and Challenges with AI in Medicine
2.1 Opportunities
As AI-based tools gradually integrate into the medical and healthcare system, here is a growing anticipation for a revolution in medical procedures, particularly in precision medicine, rehabilitation treatment, and medical decision-making. With the substantial investment and governmental support on the development for AI development, the processes of related research and commercialization transformation are constantly accelerating. This indicates that using data-driven AI models to identify healthcare needs and medical solutions will gradually become a tangible reality.
The practice of AI in healthcare can be roughly categorized into seven types (refer to Fig.1), encompassing a wide range of medical requirements, ranging from patient-centric applications to administrative tasks.
Virtual Assistant
In recent years, hospitals have witnessed the integration of at least three types of AI systems, each offering unique benefits. Electronic healthcare records have gained widespread usage across counties[1] , kindly improved the consultation efficiency and alleviating the workload of doctors. Additionally, intelligent speech technology (IST) has been implemented to provide doctors with more time for patient communication[2]. This kind of software is already available and can significantly reduces the turnaround times[3]. Furthermore, the Intelligent Guidance Robot has been introduced in countries like China, where patients may not require a referral from a family doctor or face challenges in determining the appropriate specialty to visit[4]. The guidance robot can provide efficient location consultation, medical procedures guidance, specialty recommendations and health educations, leading to an improved patients’ experience and reduced administrative pressure. Digital triage is another online guidance system that provides suggestions of self-management, triage suggestions and specialty to visit. Its implementation not only enhances accessibility to medical advice but also reduces the workload on primary healthcare providers [5].
Medical Image
Deep learning methods utilizing neural networks have achieved remarkable strides in image recognition and classification, showcasing the potential of AI in image-based diagnosis of disease categories such as radiology, ophthalmology, and oncology[6, 7]. Machine learning, particularly deep learning, has also made significant advancements in digital pathology and radiology research, revolutionizing early cancer detection[8]. For instance, the application of AI in radiology diagnosis aids doctors in prioritizing critical cases and enhances the efficiency of diagnosis and treatment. Given that clinical diagnosis often entails analyzing extensive examination results and images, AI algorithms can rapidly process and analyze large datasets, potentially enabling earlier disease detection [9].
Computer aided-detection
Since the well practice of Da Vinci Surgical System, researches on AI-assisted surgery have developed rapidly. For instance, algorithmic decision support tools, computer-aided navigation, and medical robots are being ulilized in spinal surgery [10]。Neural Networks enable high-fidelity 3D reconstruction of the spine to automatically segment and detect vertebrae, thereby minimizing associated risk factors of manual measurement[11]。Computer vision-based machine learning applications optimize computer-aided navigation systems used by spinal surgeons, allowing doctors to avoid potential obstacles in the optimal screw trajectory provided during navigation[12]。In the field of neurosurgery, research has shown that intelligent surgical robots operated through automatic dynamic motion can assist doctors in performing precise tasks efficiently. These innovative approaches exhibit great potential in the realm of precision medicine [13]。
Risk Prediction
As a powerful and adaptable framework capable of identifying intricate patterns in large-scale clinical and molecular datasets, AI holds great promise for improve risk prediction and drive the development of precision medicine. Although still in the experimental phase , researchers are optimistic about the integration of AI in risk prediction, combining multi-modal data such as clinical, traditional methods and genomics to form more accurate and personalized predictions[14]. By taking a holistic approach to analyzing patient data and understanding disease progression, AI-driven disease risk prediction has been investigated in various conditions, including osteoporotic fractures[15], cardiovascular disease[16], and more.
Medicine
AI has found applications in various sectors of the pharmaceutical industry, including drug discovery and development, drug reuse, improving pharmaceutical productivity, and clinical trials.
For example, computer-aided drug design (CADD), which, when combined with AI, has garnered attention for its ability to accelerate and reduce the costs associated with drug developmen [17]. Leveraging AI technologies like machine learning, it becomes possible to predict the activity and toxicity of therapeutic molecules within the human body. This, in turn, expedites the screening process for potential drugs, allowing for more efficient identification of compounds that may not fare well in clinical trials [18].
Health care
Healthcare encompasses nutrition, physical health and mental health management, and with AI technology we can establish a more precise and personalized health care service system. An illustrative example is a mobile cardiovascular healthcare system enhanced by AI algorithms. It utilizes mobile phone monitoring and automated analysis to provide doctors with real-time and accurate ECG diagnosis results through cloud servers. This allows ordinary individuals and doctors to access valuable heart health scores [19].The application of AI in mental health can bring benefits such as patients triage, improved quality of care services, and faster treatment or interventions [20].For example, by processing “natural language” data such as ones’ blog from social media, it is possible to observe and measure a person's mental health distress and identify at-risk individuals[21].
Hospital management
Although hospital technologists and administrators will show doubts to the use of AI tools, AI-based approaches, such as hierarchical diagnosis and treatment schemes, and patient referral services, etc., are promising to reduce the huge pressure on human resources deployment in countries like China, which has a great demand of outpatient services. Intelligent hospital management models[22] can send timely patient reminders automatically, keep the environment clean, and logistics robots, in particular, can help transport medical supplies and medical equipment without any physical labor. A survey found that 64.7% of hospital staff believe that the use of AI to manage the inventory of drugs and hospital daily supplies, can connect it to the needs of logistics and supply chains in a timely manner, thus reducing human errors in manual processing[23].
2.2 Challenges
While AI can bring many benefits, the challenges it faces are even more important to ignore. Compared to doctors, AI systems face the fact of high false-positive rate in the real-word clinics[24]. Surveys show that many doctors have doubts about the quality and safety of AI[6]. And, even if AI can complete the designed and specified tasks and achieve an acceptable accuracy rate, can it really bring better help to patients?
Interpretability
AI is composed of complex computational models that simulate human behavior and brain activity. Just as we cannot explain how every structure and function of the human brain corresponds to its behavior, the relationship between the results of AI computation and its input variables is also difficult to explain. The more complex the AI algorithm, the less interpretability it becomes. For example, deep learning, which is highly effective in disease or drug screening, is called "black box" because of its complex internal nested models, making it difficult for people to intuitively understand its internal composition and how it selects key factors that affect the results. Especially in the medical field, every decision can potentially impact a patient's health and future life . In clinical practice, it is difficult for doctors to make medical diagnoses based on results that lack explanation. For patients, it is unbelievable to accept a heartless result based solely on big data calculations.
Issue about the dataset
When we want to use AI to reduce costs, both the AI system and the devices that support such systems can be very expensive. When AI needs to achieve acceptable accuracy, it means a large amount of input, complex models, and massive computational power. For example, a medical image may contain billions of pixels, and that's before any calculations have even begun. Obviously, regular computers and ordinary neural network algorithms are difficult complete such massive data calculations. Typically, people reduce the computational requirements of big data by cropping images and focusing only on smaller regions of interest[25]. However, this indicates that the medical image input is artificially weighted.
Another problem is that explainable AI is usually based on supervised learning, which requires the data itself to have classification labels in order to determine the correctness of its calculations. Typically, these labels are provided by medical experts, however, considering the time cost, labor cost, and number of personnel, it is impossible for medical experts alone to complete the labeling of large-scale data. Therefore, a significant portion of the labels inevitably come from non-professionals hired by the company, which raises issues related to label accuracy and data privacy.
Data privacy and safety
Can data collected by AI be employed ethically in every instance? This issue is complicated by the lack of industry guidelines for the ethical use of AI and machine learning in healthcare. Beyond the previously mentioned privacy concerns, the persistent threat of hacking and data breaches underscores the difficulty of guaranteeing data security in the realm of big data. Although traditional data gathering typically mandates the explicit consent of patients, the pervasive nature of data collection raises the possibility that individuals might inadvertently mistake AI for a human entity and unwittingly authorize it to gather their information. Once medical entities have procured data, there's the potential for companies to engage in AI processing—refining data and crafting models—without patient consent. A case in point involves the collaboration between Google DeepMind and National Health Service (NHS) Foundation, where identifiable patient records were shared without direct patient approval for the development of an app to flag potential kidney injuries[26]. Is there also the risk of hackers deliberately penetrating and tampering with algorithms to skew patients' treatment regimes? Imagine, for instance, the peril of someone remotely administering an excessive dose of medication via a patient's infusion pump.
Bias
AI has the transformative potential to broaden access to medical services across diverse populations. However, it concurrently risks magnifying pre-existing biases, stemming from the historical data it is trained on[9]。uch biases may arise due to gaps in the datasets—where certain demographics are underrepresented—or from subjective labels assigned during the data curation phase. It is a recognized issue that AI tools deployed in healthcare may lack the ability to generalize effectively to novel data variants they have yet to encounter, which can result in poor performance or ingrained biases impactfully affecting minority groups. Furthermore, the variability in expert opinions during the manual data labelling process can introduce inconsistencies that may propagate covert biases within AI systems trained on these datasets.
Trust from patients and healthcare professionals
How to establish trust in AI systems is still a long-term exploration question. The clinical environment deeply values AI systems that are transparent, intelligible, user-friendly, and capable of meaningful interpretation, yet these remain pressing hurdles for AI integration. Beyond the issue of interpretability lies the critical concern of error scrutiny.
Clinicians and patients are unlikely to place their confidence in a model whose accuracy has not been rigorously validated. Furthermore, the repeatability of AI computations is a pressing question. Can the system consistently reproduce results, particularly when applied to varying datasets? The reliability of a research algorithm's performance across different datasets is not yet a certainty. Compounding the issue is the fact that research datasets, code, and trained models often remain proprietary, thus complicating the process of external validation.
Responsibility
When medical professionals stray from medical norms and cause harm, the path to accountability is clear—they must assume responsibility. But the assignment of blame becomes perplexing when errors emerge post-clinical validation of artificial intelligence, such as faults in AI software or AI-augmented devices used in patient care. Pinpointing who's accountable — the creators, overseers, vendors, or medical staff — is a contentious issue.
Additionally, when AI's tailored suggestions do not align with established medical practices, the splitting of responsibility turns into a thorny issue. How do you slice the pie of liability when doctors make calls based on AI's advice versus their own clinical experience? This quandary extends to patients exercising self-care who are faced with AI suggestions that diverge from their own knowledge and preferences. And in the event of a mishap, the question looms: who shoulders the blame? Thus, while human oversight in AI-augmented decision-making is imperative, it brings with it a diminished sense of individual agency and novel risks to navigate.
Regulation
As policymakers grapple with responsibility allocation, the regulatory landscape for AI deployment in healthcare demands rigorous attention. It's clear that flawed AI algorithms carry the potential for grave patient harm, potentially scaling up to medical misadventures. While an individual doctor's mistake might endanger a single patient, the fallout from AI-induced errors can be far-reaching[27] , underscoring the need for thorough validation of AI applications. Regulatory bodies bear the weighty task of not only proving the efficacy and merits of AI in medical practices but also safeguarding patient data security and privacy.
Moreover, AI systems are dynamic; they evolve and possibly cultivate novel methodologies or errors as they digest new datasets. Regulatory protocols rooted in the assessment of static initial model parameters will fall short in catching these adaptations. Hence, it's critical that regulatory agencies refine their oversight frameworks to stay abreast of such evolving systems, ensuring that certifications remain robust and relevant.
3. AI, ecology and health
3.1 Ecological impact and health influence of AI
AI in healthcare is creating a ripple effect of positive change for our planet. As anthropogenic climate change accelerates, the healthcare sector, responsible for a non-negligible 1%-5% of the global environmental footprint and 4.4% of greenhouse gas emissions, is exploring ways to green its operations. In the face of such data, medical organizations across the globe are proactively reevaluating and restructuring their operational modalities to mitigate their environmental impacts [28].
Measures being considered include the reduction in combustion of fossil fuels and the phasing out of deleterious substances such as mercury in favor of less harmful alternatives. But, it is the generation and subsequent management of massive medical waste that poses a most pressing environmental challenge within the sector. In particular, the widespread adoption of nitrile gloves, valued for their imperviousness and thus serving as a protective barrier in shielding health workers from disease and harmful substances, has clinical importance. However, the synthetization of nitrile involves petrochemicals, and constituents of these gloves have been identified by the International Agency for Research on Cancer (IARC) as potential carcinogens[29], raising issues regarding their lifecycle from production to disposal that have environmental and health repercussions. The advent of COVID-19 pandemic has led to a rapid increase of personal protective equipment waste, notably disposable masks and gloves, and thereby amplifying the strain on waste management infrastructures. The consequent augmentation of environmental pollution via landfilling, coupled with the potential broadening of viral transmission vectors due to improper waste management, underscores a critical global challenge [30].
The integration of AI technologies within the healthcare sector offers a promising avenue for diminishing extensive resource and energy expenditures. For instance, the introduction of an AI-driven medical waste management framework can lead to the automated categorization of waste streams, the implementation of advanced treatment methodologies, and the refinement of logistical pathways for waste transport, thus enhancing overall efficiency and reducing environmental load. Auto Robots could take up the slack in hospitals, allowing less human hustle and more focus on patient care. The deployment of advanced speech recognition AI systems could cut the need for paperwork, allowing for deeper doctor-patient interactions. In surgical environments, AI-equipped robotic systems represent a leap forward in precision medicine. By incorporating surgical monitoring models, these systems aim to reduce intraoperative errors linked to manual positioning, potentially increasing the accuracy and success rates of surgical procedures while concurrently decreasing the repetitive consumption of surgical supplies. Dentistry illustrates an application where AI's impact can be profound: conventional way of dental prosthetics often involves the physical creation and transportation of tooth models, a procedure both material-intensive and logistically demanding. AI changes the game completely with AI-driven modeling, 3D reconstructions and cloud servers. This approach offers significant reductions in material waste and transportation emissions. Furthermore, AI's role in pharmacological research and drug testing is transformative. The utilization of predictive models in drug discovery allows for a substantial decrease in experimental resource usage by simulating pharmacokinetics and pharmacodynamics, thus conserving laboratory materials and reducing the generation of hazardous biomedical waste.
3.2 Enhancing human health with the help of AI
When medical AI applications are deployed in a manner that enhances ecological sustainability, they concurrently confer benefits to public health through the promotion of environmental determinants of health. The omnipresence of digital technology in the modern age is evident as a considerable portion of the population engages extensively with the internet, smartphones, and social media. This constant interaction with digital platforms facilitates a shift in human behaviors and habits by leveraging the capabilities of computing technology to foster healthier lifestyle choices through continuous environmental cues and behavioral reinforcement strategies.
Health-oriented applications on smartphones, utilizing AI algorithms, function to provide personalized feedback to promote the acquisition of beneficial habits such as regular physical activity, balanced sleep patterns, and nutritious dietary intake while concurrently discouraging adverse behaviors like excessive alcohol consumption and high meat consumption[31]. Utilizing an AI system that integrates a mix of tools like mobile apps, SMS, wearables, social networks, and web-based communication platforms is promising to offer vital support to individuals with mental health conditions, such as depression and anxiety. By guiding patients towards healthier behavior patterns, enabling comprehensive behavior modification programs, and facilitating direct feedback to mental health professionals, these technologies can significantly boost the impact of therapeutic interventions [32].
So, Will these behavior affect the ecological environment? Of course, yes. For instance, adopting healthy behaviors and habits, such as avoiding smoking and maintaining a nutritious diet, can lower the chances of falling ill, consequently lessening the strain on healthcare resources and energy consumption. Embracing healthy practices like jogging or cycling can also inspire individuals to opt for more eco-friendly modes of transportation, ultimately reducing carbon dioxide emissions from vehicles. As a result, the improvement in the ecological environment directly safeguards human health, creating a virtuous cycle. In this context, AI plays a role in personalized alerts, highlighting the potential risks associated with unhealthy behaviors and beliefs, thereby influencing people's decision-making process.
4. Conclusion
The dual nature of AI technology in healthcare—its potential benefits and inherent challenges—demands a keen focus on enhancing the medical process through AI while upholding stringent safety measures. Puerta and his team have adeptly brought forward an insightful proposition: "A necessary step in the digitalization of our environments is to include the users in the decision loop, following a more human-centric paradigm." This recommendation underscores a key tenet: AI isn't a cure-all — the critical task of decision-making still falls squarely on human shoulders. Yet, the hope is that with AI's support, those human decisions become increasingly advantageous and informed.
References
1. Nickson, D., et al., Prediction and diagnosis of depression using machine learning with electronic health records data: a systematic review. BMC Med Inform Decis Mak, 2023. 23(1): p. 271.
2. Zhang, J., et al., Intelligent speech technologies for transcription, disease diagnosis, and medical equipment interactive control in smart hospitals: A review. Comput Biol Med, 2023. 153: p. 106517.
3. Poder, T.G., J.F. Fisette, and V. Déry, Speech Recognition for Medical Dictation: Overview in Quebec and Systematic Review. J Med Syst, 2018. 42(5): p. 89.
4. Ma, A.C., Z. Meng, and X. Ding, Performance Review of Intelligent Guidance Robot at the Outpatient Clinic Setting. Cureus, 2021. 13(8): p. e16840.
5. Gottliebsen, K. and G. Petersson, Limited evidence of benefits of patient operated intelligent primary care triage tools: findings of a literature review. BMJ Health Care Inform, 2020. 27(1).
6. Lenharo, M., An AI revolution is brewing in medicine. What will it look like? Nature, 2023. 622(7984): p. 686-688.
7. Lenharo, M., An AI revolution is brewing in medicine. What will it look like? (1476-4687 (Electronic)).
8. Elkhader, J. and O. Elemento, Artificial intelligence in oncology: From bench to clinic. Semin Cancer Biol, 2022. 84: p. 113-128.
9. Rajpurkar, P., et al., AI in health and medicine. Nat Med, 2022. 28(1): p. 31-38.
10. Manickam, P., et al., Artificial Intelligence (AI) and Internet of Medical Things (IoMT) Assisted Biomedical Systems for Intelligent Healthcare. Biosensors (Basel), 2022. 12(8).
11. Chang, M., et al., The Role of Machine Learning in Spine Surgery: The Future Is Now. Front Surg, 2020. 7: p. 54.
12. Nguyen, N.Q., et al., Machine vision augmented reality for pedicle screw insertion during spine surgery. J Clin Neurosci, 2020. 72: p. 350-356.
13. Ko, S., et al., Intelligent control of neurosurgical robot MM-3 using dynamic motion scaling. Neurosurg Focus, 2017. 42(5): p. E5.
14. Olier, I., et al., How machine learning is impacting research in atrial fibrillation: implications for risk prediction and future management. Cardiovasc Res, 2021. 117(7): p. 1700-1717.
15. Li, Z., et al., AI algorithms for accurate prediction of osteoporotic fractures in patients with diabetes: an up-to-date review. J Orthop Surg Res, 2023. 18(1): p. 956.
16. Mohd Faizal, A.S., et al., A review of risk prediction models in cardiovascular disease: conventional approach vs. artificial intelligent approach. Comput Methods Programs Biomed, 2021. 207: p. 106190.
17. Vemula, D., et al., CADD, AI and ML in drug discovery: A comprehensive review. Eur J Pharm Sci, 2023. 181: p. 106324.
18. Paul, D., et al., Artificial intelligence in drug discovery and development. Drug Discov Today, 2021. 26(1): p. 80-93.
19. Fu, Z., et al., Artificial-Intelligence-Enhanced Mobile System for Cardiovascular Health Management. Sensors (Basel), 2021. 21(3).
20. Zidaru, T., E.M. Morrow, and R. Stockley, Ensuring patient and public involvement in the transition to AI-assisted mental health care: A systematic scoping review and agenda for design justice. Health Expect, 2021. 24(4): p. 1072-1124.
21. Islam, M.R., et al., Depression detection from social network data using machine learning techniques. Health Inf Sci Syst, 2018. 6(1): p. 8.
22. Mi, D., et al., Exploring intelligent hospital management mode based on artificial intelligence. Front Public Health, 2023. 11: p. 1182329.
23. Bas, T.G., et al., Opinions Related to the Potential Application of Artificial Intelligence (AI) by the Responsible in Charge of the Administrative Management Related to the Logistics and Supply Chain of Medical Stock in Health Centers in North of Chile. Int J Environ Res Public Health, 2023. 20(6).
24. Kanagasingam, Y., et al., Evaluation of Artificial Intelligence-Based Grading of Diabetic Retinopathy in Primary Care. JAMA Netw Open, 2018. 1(5): p. e182665.
25. Courtiol, P., et al., Deep learning-based classification of mesothelioma improves prediction of patient outcome. Nature Medicine, 2019. 25(10): p. 1519-1525.
26. Powles, J. and H. Hodson, Google DeepMind and healthcare in an age of algorithms. Health and Technology, 2017. 7(4): p. 351-367.
27. Casey Ross, I.S. IBM’s Watson supercomputer recommended ‘unsafe and incorrect’ cancer treatments, internal documents show. 2018 [cited 2018 July 25]; Available from: https://www.statnews.com/2018/07/25/ibm-watson-recommended-unsafe-incorrect-treatments/.
28. Vatovec, C., L. Senier, and M. Bell, An ecological perspective on medical care: environmental, occupational, and public health impacts of medical supply and pharmaceutical chains. Ecohealth, 2013. 10(3): p. 257-67.
29. IARC, Acrylonitrile. IARC Monographs on the Evaluation of Carcinogenic Risks to Humans 71. 1999: p. 43–108，109-225.
30. Chen, C., et al., What medical waste management system may cope With COVID-19 pandemic: Lessons from Wuhan. Resour Conserv Recycl, 2021. 170: p. 105600.
31. Hughes, C., The Use of Technology to Change Behavior: Synoptic Abstract. 2017.
32. Moon, K., M. Sobolev, and J.M. Kane, Digital and Mobile Health Technology in Collaborative Behavioral Health Care: Scoping Review. JMIR Ment Health, 2022. 9(2): p. e30810.
