Jurnal Teknik Informatika dan Sistem Informasi ISSN 2407-4322
Vol. 10, No. 2, Juni 2023, Hal. 589-597 E- ISSN 2503-2933 589
The Challenges Of Implementing Medical Practice In The Artificial Intelligence Era
Yuliana
Fakultas Kedokteran, Universitas Udayana; Denpasar, 0361 222510 e-mail: [email protected]
Abstrak
Tulisan ini bertujuan untuk menjelaskan tantangan praktek medis di tengah era kecerdasan buatan. Metode yang digunakan adalah narrative literature review. Tulisan dicari dari Science Direct, Google Scholar, dan PubMed. Kriteria inklusi adalah penelitian dan tinjauan pustaka. Kriteria eksklusi adalah tidak tersedianya teks secara lengkap. Hasil telaah pustaka menunjukkan bahwa kecerdasan buatan telah diaplikasikan di berbagai aktivitas sehari-hari. Namun, di samping sisi positif kecerdasan buatan, ada banyak hal yang perlu diperhatikan jika akan mengimplementasikan kecerdasan buatan pada praktek medis dan pelayanan kesehatan. Masalah yang dihadapi adalah kesulitan saat permulaan, kurangnya pemahaman, posisi pasien dan dokter, proses pengambilan keputusan, reliabilitas data, dan pertanggungjawaban. Kecerdasan buatan memiliki celah dalam mengatasi privasi data. Di sisi lain, sistem harus selalu diperbaharui untuk fungsi monitoring. Oleh karena itu, tantangan dalam mengimplementasikan kecerdasan buatan adalah kerja yang terintegrasi dari pengembang kecerdasan buatan, klinisi, regulator (pemegang kebijakan), dan para peneliti.
Keterbatasan tulisan ini adalah berupa tinjauan pustaka naratif, sehingga informasi yang diperoleh tidak seluas dan selengkap penelitian. Kontribusi tulisan ini adalah untuk praktek medis dan pengembangan penelitian di bidang kecerdasan buatan. Kebaharuan tulisan ini adalah meliputi perkembangan dan tantangan penerapan kecerdasan buatan di bidang medis dan pelayanan kesehatan.
Kata kunci—kecerdasan buatan, medis, pelayanan kesehatan
Abstract
This paper aims to describe the challenges of healthcare practice in the artificial intelligence era. This is a narrative literature review. Literature was searched from Science Direct, Google Scholar, and PubMed. The inclusion criteria are research and review. The exclusion criteria are unavailable in full-text journals. Artificial intelligence (AI) is applied in large activities of daily life. However, besides the positive sides of AI, there are still many concerns about implementing AI in medical practice and healthcare services. The sources of the problems are uneasiness at the beginning, lack of understanding, the position of the physicians and patients, the decision-making process, reliability of the data, and accountability. AI has some gaps in handling the privacy of the data. On the other side, the system should be updated for monitoring functions. Therefore, the challenge in implementing AI in the healthcare practice is the integrated work of the AI developer, clinicians, regulators, and researchers. This paper is a narrative literature review, therefore the information would not be as complete as a research study. The contribution of the paper is to the healthcare practice and research development in artificial intelligence. Novelty: This paper includes the development and challenges of artificial intelligence in medical science and healthcare practice.
Keywords— artificial intelligence, healthcare, medical.
590 Jatisi ISSN 2407-4322 Vol. 10, No. 2, Juni 2023, Hal. 589-597 E-ISSN 2503-2933
1. INTRODUCTION
Artificial intelligence (AI) is applied in large activities in daily life. Examples are navigation, prediction of traffic jams, translation, social media, etc. However, the application in the healthcare area is still limited in some countries. AI cannot replace human labor [1]. The positive sides of AI in healthcare are early detection of diseases, increasing the accuracy of diagnosis, early detection, also diagnostic, and therapy [2]–[5]. Besides the benefits of AI in healthcare practice, the concerns regarding some skepticism should be depicted [1].
Software as a medical device (SaMD) is one special software for medical purposes. The Food and Drug Administration (FDA) ensures effective technology for patients and healthcare.
Therefore, manufacturers have to submit an appropriate marketing application before distributing any medical devices. There are three categories of medical devices, namely class III (high-risk), class II, and class I (low-risk). Class II needs FDA clearance, while class III has to be accompanied by FDA approval. FDA clearance needs at least 510.000 submissions that been passed the review and clearance process from FDA. To pass the FDA clearance needs a substantially equivalent quality to a similar approved device. FDA approval means the manufacturers must submit premarket applications and clinical test results to gain FDA approval. FDA has a special committee called the Center for Devices and Radiological Health (CDRH). The tasks are to publish guidance about a risk-based approach in determining the requirement of a premarket submission [1].
During the COVID-19 pandemic, AI has been adopted in some places. However, there are some challenges and barriers regarding its implementation. An important aspect in considering using AI is the possibility of delivering health services without jeopardizing the health workers’ safety. AI could bring efficiency to physicians and patients. Successful AI implementation needs a comprehensive analysis of the Big Data repositories. The AI system can be leveraged by automating repetitive administrative tasks such as billing procedures.
Meanwhile, the patients may use AI in locating a physician [6].
The development and implementation of AI are different among countries. In China, the
‘New Generation Artificial Intelligence Development Plan’’ was released in 2017. This step is important to strengthen privacy, respect, and accountability for human welfare. Meanwhile, in Europe, there is a regulatory framework proposal for high-risk AI systems. A code of conduct is provided for non-high-risk AI systems. In the United Kingdom, there are several-point frameworks for the guidance of AI as follows: avoid unintended consequences, ensure fairness, data safety, accountability principle for the users, law compliance, and monitoring system.
However, to make the data among countries more accurate and representative, the International Classification of Diseases, Tenth Revision, and Clinical Modification (ICD-10-CM) should be applied in the AI algorithm. Electronic medical records are essential in increasing data safety and timely. AI must show accurate and sound data to help the decision-making process for clinicians [1].
Health inequities also happen among the population with different social backgrounds.
Lower socioeconomic status means less access to healthcare. This group also has worse nutrition status. This condition drives the group into health inequities and discrimination. On the other side, AI also causes inequalities among many patients in the US. A well-designed AI model plays an essential role in reducing discrimination. AI model can be used for diagnosis, prognosis, treatment, monitoring, and prediction [7]. However, poorly designed AI may increase the gap in health discrepancy. Therefore, the AI model must be well-designed to prevent unnecessary problems in the future [8].
Developing AI programs has many challenges. Most programmers develop systems based on the deductive method. However, it is impractical. Therefore, the deep learning system is developed now [9]. The deep learning system is divided into many algorithms. One of them
Jatisi ISSN 2407-4322
Vol. 10, No. 2, Juni 2023, Hal. 589-597 E- ISSN 2503-2933 591
that is commonly used is an artificial neural network. It is usually implemented in clinical problems [10].
Artificial neural networks are comprised of many types. However, the most common is deep neural networks. This system classifies the data of the patient based on the diagnosis.
However, bias can occur and lead to a poorer prediction of health outcomes. Therefore, the development of health AI products must include the participation of government and non- government organizations, activism, and international levels. These steps will create more reliable data for AI products [10].
Sunarti et al. in 2021 studied the opportunity of AI and the risk of its implementation in health services. They collected some kinds of literatures from three databases, namely Web of Science, EBSCOhost, and Google Scholar. The Joanna Briggs Institute was used to assess the quality of the studies. The results revealed that AI can be implemented to improve diagnostics, treatment, and prevention. However, the challenges are patient autonomy rights, early adoption, and ethical problems that might happen during AI implementation in the healthcare system [11].
Therefore, this paper will review the challenges of healthcare practice in the AI era.
2. METHODOLOGY
This is a narrative literature review. Articles were taken from the Science Direct and Google Scholar databases. Inclusion criteria are review and research articles. Exclusion criteria are not peer-reviewed articles and unavailable full-text articles. Articles were read twice to reduce the bias. The selected articles were summarized and narrated descriptively. There were 17 selected articles.
3. RESULT AND DISCUSSION
The concerns in using AI are reliability, liability, and accountability. Reliable AI technology has to be refined and modified using new data. Real-time monitoring is critical. This system is used for identifying possible failure and their causes. Quick correction is essential to maintain the reliability of the AI system. The reliability score of the AI system will be compared with the existing systems. The FDA label is needed to show the efficacy of the AI algorithms [1].
The quality, efficacy, indication, and adverse effects of the AI products need to be revealed to the users to maintain accountability. Therefore, adequate training must be provided for the physicians and personnel who tackle the AI algorithm. The FDA and current evidence- based practice have to be the guidelines. The higher risk of AI algorithms needs meticulous monitoring after implementation [1].
There are some clinical challenges associated with the implementation of artificial intelligence in clinical practices as shown in Figure 1. Some challenges are identification problems, regulation, interpretability, assessment, also patient engagement. Transitioning to electronic medical records is time-saving. This method could help clinicians in making decisions promptly especially when clinical decisions should be made promptly by a combination of clinical judgment and augmented intelligence algorithms. Bias will be lessened when the augmented intelligence models can perform the tasks accurately [1].
592 Jatisi ISSN 2407-4322 Vol. 10, No. 2, Juni 2023, Hal. 589-597 E-ISSN 2503-2933
Figure 1. Clinical Challenges Associated With The Implementation of Artificial Intelligence In Clinical Practice [1]
Whole Slide Imaging (WSI) as an advanced device, could capture the entire resection, cytological preparations, and biopsy at diagnostic resolution. Those digital assets are valuable for diagnosis. Quantitative data can be extracted from the image by automated segmentation and pixel analysis methods. Thus, consistency and reproducibility are accomplished. Even subvisual clues can be identified by image analysis. This advantage is important for images that cannot be seen by the naked eye. WSI got FDA clearance in 2017 [1].
Digitization of pathology helps pathologists to integrate digital scanners with laboratory systems, and also dispatch digital slides to staff and outside. This procedure is also known as digital pathology. It is useful in diagnostic time-saving for pathologists. Therefore, the digital transformation of pathology is increasing from time to time. The reasons behind this situation are a shortage of pathologists, pathology workloads, increased cancer screening programs, and increasing complexity of pathology tests. The benefits of using AI applications in the pathologic department are efficiency, reproducibility, and improvements. Thus, there will be a major improvement in pathology AI in the coming years. Continued innovation is needed to gain those improvements [1].
The real transformation of digital pathology is marked by advanced technologies, digital transformation, AI innovation, and acceleration of AI pathology usage. The technologies have the potency to reach further evolution in the future. Therefore, computational pathology and biomedical imaging are developed recently. The development of neural networks is needed to enhance the various computational pathology tasks. U-Net has been used in several applications for data augmentation. Meanwhile, MVPNet was used for breast cancer diagnosis. U-Net and MVPNet are parts of the deep learning network. These methods used multiple viewing parts for better diagnosis. The benefit of using MVPNet are fewer parameters, comparable performance, and simultaneous processes of local and global features for diagnosis. Another method, A ResNet, is based on a 101-layer deep learning network that showed high efficiency in breast cancer metastasis discrimination [1]. The process of identifying anomalies is shown in Figure 2 [3]. Figure 3 showed the process of molecular profiling [3].
Jatisi ISSN 2407-4322
Vol. 10, No. 2, Juni 2023, Hal. 589-597 E- ISSN 2503-2933 593
Figure 2. Unsupervised Learning In AI [3]
Figure 3. Process of Molecular Profiling [3]
Liability means ensure that all stakeholders (developers, physicians, and researchers) do a continuous evaluation of the effectiveness and safety. Although the AI has received FDA approval, it should be monitored during the post-market period. When any adverse events and/or system failures are found, the developers must report and investigate them thoroughly.
The principle of equity means the absence of unfairness among groups of people in any different economic, social, or geographical condition [1].
Some concerns about using AI in healthcare are cybersecurity, including handling personal data; the understanding of how the operating system of the AI, including the roles of clinicians and patients in the decision-making process; and the reliability of the data. The
594 Jatisi ISSN 2407-4322 Vol. 10, No. 2, Juni 2023, Hal. 589-597 E-ISSN 2503-2933
continuous monitoring and updating of the system. Those concerns need to be addressed by using an integrated network among clinicians, developers, and researchers. The steps are critical for developing the best service for the patient [1].
The fast development of AI in health care needs concerns in ethical and legal aspects.
The policy for disclosure of the data might compromise the privacy of the customer. Privacy should be well protected. The steps in protecting privacy are creating safer applications or systems, also upgrading the literacy of the consumers. The data often include sensitive aspects of a patient’s physical and/or mental health that have to be protected [12]. The data contain three important aspects, namely variety, volume, and veracity. These aspects are essential in predicting healthcare [13]. Innovation in AI systems will transform physicians into high-tech educated and good providers in the future [14]. Comprehensive data have a valuable role in applying AI in clinical practice. This action is important for the efficiency of the system [15].
AI can be used in health care services in the form of data processing, machine learning, and the interpretation of diagnostic images such as X-rays, computed tomography (CT), and magnetic resonance imaging (MRI). The demand for prompt diagnosis and treatment makes the use of AI valuable. Therefore, innovative AI technologies are needed to support healthcare services [15]. However, when bias occurs, it needs a feedback loop as shown in Figure 4 [10].
Figure 4. Feedback Loop Mechanism [10]
Virtual Reality as part of Artificial intelligence could be used to stimulate special environments and social interaction virtually. Craving and nicotine use could be controlled by using virtual reality. However, more studies are needed. A combination of virtual reality and Non-Invasive Brain Stimulation (NIBS) can be given to patients for a better emotion simulation.
Modification of cortical pathways and plasticity can be accomplished by transcranial magnetic stimulation. Phobia and Post Traumatic Stress Disorders (PTSD) revealed an improvement with these methods. Supportive communication is given through online education methods [16].
Convolutional neural networks (CNN) are useful to process multiple arrays of data.
Four benefits of using CNN are shared weights, pooling, local connections, and multiple layers.
The process is similar to speech and text from sounds to phones, words, and sentences. Face
Jatisi ISSN 2407-4322
Vol. 10, No. 2, Juni 2023, Hal. 589-597 E- ISSN 2503-2933 595
recognition is one of the important abilities of CNN. The application of face recognition includes autonomous mobile robots. An example of the CNN process is shown in Figure 5. It is useful for a diagnostic pattern [9].
The difference between using machine learning and deep learning for dental cases is shown in Figure 6. Machine learning relies on expert knowledge. While deep learning can perform feature extraction and selection for classification. Although the deep learning method uses several hidden layers, it can learn relevant images [17].
Figure 5. The Example of Vision Deep Convolution Neural Network Process [9]
Figure 6. Difference Between Machine Learning and Deep Learning In Diagnosis Dental Case [17]
4. CONCLUSION
In conclusion, the challenges of healthcare practice in the AI era are ensuring the privacy of the data, maintaining the stability of the result in using AI, including reducing the potency of bias. AI can be used in health care service in the form of data processing, machine learning, and the interpretation of diagnostic images such as X-rays, computed tomography (CT), and magnetic resonance imaging (MRI). The demand for prompt diagnosis and treatment makes the use of AI valuable. Therefore, innovative AI technologies are needed to support
596 Jatisi ISSN 2407-4322 Vol. 10, No. 2, Juni 2023, Hal. 589-597 E-ISSN 2503-2933
healthcare services. However, when bias occurs, it needs a feedback loop. Feedback, continuous update, and monitoring have to be done to ensure the reliability and validity of the AI.
5. SUGGESTION
It is suggested to have further studies in the future about AI implementation in healthcare in many clinical situations settings and various diseases in the population.
ACKNOWLEDGMENT
The author acknowledges the committee of the 2 nd International Conference on Advanced Information Technology and Communication (IC-AITC 2) who has prepared the conference very well.
REFERENCE
[1] G. Bazoukis, J. Hall, J. Loscalzo, and E. M. Antman, “The Inclusion of Augmented Intelligence In Medicine: A Framework For Successful Implementation,” Cell Reports Med., pp. 1–8, 2022, doi: 10.1016/j.xcrm.2021.100485.
[2] R. Alizadehsani, M. Roshanzamir, M. Abdar, A. Beykikhoshk, A. Khosravi, and M.
Panahiazar, “A Database For Using Machine Learning and Data Mining Techniques For Coronary Artery Disease Diagnosis,” Sci. Data, Vol. 6, No. 227, pp. 1–13, 2019, doi:
10.1038/s41597-019-0206-3.
[3] A. Serag et al., “Translational AI and Deep Learning in Diagnostic Pathology,” Front.
Med., Vol. 6, No. October, pp. 1–15, 2019, doi: 10.3389/fmed.2019.00185.
[4] C. Wu, X. Zhao, M. Welsh, K. Costello, K. Cao, and A. A. Tayoun, “Using Machine Learning to Identify True Somatic Variants from Next-Generation Sequencing,” Mol.
Diagnostics Genet., Vol. 66, No. 1, pp. 239–246, 2020, doi:
10.1373/clinchem.2019.308213.
[5] S. Le et al., “Pediatric Severe Sepsis Prediction Using Machine Learning,” Front.
Pediatr., Vol. 7, No. October, pp. 1–8, 2019, doi: 10.3389/fped.2019.00413.
[6] A. S. Young, “AI In Healthcare Startups and Special Challenges,” Intell. Med., Vol. 6, pp. 1–3, 2022, doi: 10.1016/j.ibmed.2022.100050.
[7] E. L. Henriksen, J. F. Carlsen, I. M. M. Vejborg, M. B. Nielsen, and C. A. Lauridsen,
“The Efficacy of Using Computer-Aided Detection (CAD) for Detection of Breast Cancer In Mammography Screening: A Systematic Review,” Acta radiol., Vol. 60, No. 1, pp. 13–
18, 2019, doi: 10.1177/0284185118770917.
[8] I. Y. Chen, P. Szolovits, and M. Ghassemi, “Can AI Help Reduce Disparities in General Medical and Mental Health Care?,” AMA J. Ethics, Vol. 21, No. 2, pp. 167–179, 2019.
Jatisi ISSN 2407-4322
Vol. 10, No. 2, Juni 2023, Hal. 589-597 E- ISSN 2503-2933 597
[9] Y. Lecun, Y. Bengio, and G. Hinton, “Deep Learning,” Nature, Vol. 521, pp. 436–444, 2015, doi: 10.1038/nature14539.
[10] C. M. Moore, “The Challenges of Health Inequities and AI,” Intell. Med., Vol. 6, pp. 1–
5, 2022, doi: 10.1016/j.ibmed.2022.100067.
[11] S. Sunarti, F. Fadzlul, M. Naufal, M. Risky, K. Febriyanto, and R. Masnina, “Artificial Intelligence in Healthcare : Opportunities and Risk For Future ,” Gac. Sanit., Vol. 35, pp. S67–S70, 2021, doi: 10.1016/j.gaceta.2020.12.019.
[12] S. Gerke and D. Rezaeikhonakdar, “Privacy Aspects of Direct-To-Consumer Artificial Intelligence / Machine Learning Health Apps,” Intell. Med., Vol. 6, No. February, pp. 1–
5, 2022, doi: 10.1016/j.ibmed.2022.100061.
[13] S. Gerke, B. Babic, T. Evgeniou, and I. G. Cohen, “The Need For A System View To Regulate Arti Ficial Intelligence/ Machine Learning-Based Software As Medical Device,”
npj Digit. Med., Vol. 3, No. 53, pp. 1–4, 2020, doi: 10.1038/s41746-020-0262-2.
[14] V. B. Kolachalama, “Machine Learning and Pre-Medical Education,” Artif. Intell. Med., Vol. 129, No. April, pp. 1–3, 2022, doi: 10.1016/j.artmed.2022.102313.
[15] A. J. London, “Artificial Intelligence In Medicine : Overcoming Or Recapitulating Structural Challenges To Improving Patient Care ?,” Cell Reports Med., Vol. 3, No. 5, pp. 1–8, 2022, doi: 10.1016/j.xcrm.2022.100622.
[16] M. Vella, “Preventing, Reducing and Treating Problematic Drug Use With Digital Technology,” Malta Med. J., Vol. 34, No. 03, pp. 83–96, 2022.
[17] R. H. Putra, C. Doi, N. Yoda, E. R. Astuti, and K. Sasak, “Current Applications and Development of Artificial Intelligence For Digital Dental Radiography,”
Dentomaxillofacial Radiol., Vol. 50, pp. 1–12, 2021, doi: 10.1259/dmfr.20210197.