Medical Artificial Intelligence Systems Research Articles

New possibilities of artificial intelligence in medicine: a narrative review

The purpose of the narrative review is to provide a descriptive analysis of the emerging capabilities of artificial intelligence (AI) to improve the diagnosis, prevention and treatment of various diseases. The article discusses which modern AI tools can be used in clinical practice, healthcare organization and medical education. The paper considers various aspects of medical AI systems, which are mainly computer support systems for medical decision-making in the process of clinical work. Much attention is paid to the possibilities of generative AI in medicine. Potential applications of AI in clinical practice have been investigated, highlighting promising prospects for both practitioners and their patients. The limitations associated with the use of AI in various fields of medicine are described, and possible ways of solving them are suggested. The problems of information security and ethical constraints associated with the introduction of AI are outlined. The broad integration of AI into public health will enhance clinical and management decision support, speed up disease diagnosis, and improve the overall quality and accessibility of healthcare services.

Ethics and artificial intelligence

The introduction of artificial intelligence (AI) into radiology promises to enhance efficiency and improve diagnostic accuracy, yet it also raises manifold ethical questions. These include data protection issues, the future role of radiologists, liability when using AI systems, and the avoidance of bias. To prevent data bias, the datasets need to be compiled carefully and to be representative of the target population. Accordingly, the upcoming European Union AI act sets particularly high requirements for the datasets used in training medical AI systems. Cognitive bias occurs when radiologists place too much trust in the results provided by AI systems (overreliance). So far, diagnostic AI systems are used almost exclusively as "second look" systems. If diagnostic AI systems are to be used in the future as "first look" systems or even as autonomous AI systems in order to enhance efficiency in radiology, the question of liability needs to be addressed, comparable to liability for autonomous driving. Such use of AI would also significantly change the role of radiologists.

МЕТОДОЛОГИЯ СОЗДАНИЯ ИНДИВИДУАЛЬНОЙ МОДЕЛИ ЗДОРОВЬЯ ЧЕЛОВЕКА С ИСПОЛЬЗОВАНИЕМ ИСКУССТВЕННОГО ИНТЕЛЛЕКТА

Background. In the modern world, the role of artificial intelligence in healthcare is becoming increasingly significant, providing new opportunities to transform traditional methods of diagnosis, treatment and medical data management. This technological breakthrough not only improves the efficiency of medical procedures, but also opens up new per-spectives in the prevention and treatment of diseases. The aim. To emphasize the need to create a personalized health model using artificial intelligence to help an individual achieve and maintain optimal health and well-being. In the field of personalized treatment, artificial intelligence plays an important role, taking into account the unique characteristics of each patient. Algorithms analyze genetic information, medical history and responses to previous therapies to develop optimal treatment plans. This opens the way to individualized medicine, where the approach to each patient is based on his or her unique characteristics. Despite all the positive aspects, the introduction of artificial intelligence in healthcare also raises questions of data privacy, ethical issues and technology security. However, if these issues are resolved, artificial intelligence promises to significantly improve the quality and accessibility of medical treatment, opening new horizons in healthcare. Results. In this article, we describe breakthroughs in artificial intelligence technologies and biomedical applications, identify problems of using and further development in medical artificial intelligence systems and summarize the economic, legal and social consequences of using artificial intelligence in healthcare, and propose a scheme for constructing a model of individual human health using artificial intelligence. Conclusion. The results of the analysis of modern scientific literature allow us to draw a conclusion about the potential for creating more effective and personalized approaches to the problem of individual health using integrated artificial intelligence technologies. The proposed methodology can serve as the basis for the development of innovative deci-sion support systems in medicine and improving the quality of medical care.

Revolutionizing Medical Practice: The Impact of Artificial Intelligence (AI) on Healthcare

The twenty-first century has witnessed significant advancements in informatics, reshaping our understanding of data processing and accessibility. Artificial intelligence (AI), encompassing techniques such as machine learning (ML), deep learning (DP), and neural networks (NN), is poised to revolutionize medicine. AI holds the capability of analyzing vast amounts of data, extracting meaningful insights, and making accurate predictions, thereby empowering industries to make informed decisions, drive innovation, and enhance efficiency. The landscape of medical AI has evolved significantly, demonstrating expert-level disease detection from medical images and promising breakthroughs across various industries. AI revolutionizes medical practice by leveraging advanced algorithms and machine learning capabilities to improve diagnostics, treatment planning, and overall patient care. However, the deployment of medical AI systems in regular clinical practice still needs to be tapped, presenting complex ethical, technical, and human-centered challenges that must be addressed for successful implementation. While AI algorithms have shown efficacy in retrospective medical investigations, their translation into practical medical settings has been limited, raising concerns about their usability and interaction with healthcare professionals. Moreover, the representativeness of retrospective datasets in real-world medical practice is subject to filtering and cleaning biases. Integrating AI into clinical medicine holds great promise for transforming healthcare delivery, improving patient care, and revolutionizing aspects such as diagnosis, treatment planning, drug discovery, personalized treatment, and medical imaging. With advanced algorithms and machine learning capabilities, AI and robotics in Healthcare can analyze large volumes of medical data, extract meaningful insights, and provide accurate predictions, empowering healthcare professionals to make informed decisions and optimize resource allocation. The availability of extensive clinical, genomics, and digital imaging data, coupled with investments from healthcare institutions and technology giants, underscores the potential of AI in healthcare. This review article explores AI's powerful potential to revolutionize healthcare delivery across multiple domains, emphasizing the need to overcome challenges and harness its transformative capabilities in clinical practice.

Commentary] Artificial Intelligence, or Artifact Intelligence? Most AI Is Not Ready for Practice

This commentary emphasizes the limitations of artificial intelligence (AI) in medical diagnosis, highlighting instances of erroneous diagnoses due to inadequate training data and a lack of explainability. It stresses the need for rigorous data curation and the cost implications of developing safe medical AI systems. The authors caution against overenthusiastic adoption of AI in medicine and call for awareness of these limitations among medical professionals and stakeholders, underscoring the importance of a cautious and resource-intensive approach to AI in healthcare.

Commentary] Artificial Intelligence, or Artifact Intelligence? Most AI Is Not Ready for Practice

This commentary emphasizes the limitations of artificial intelligence (AI) in medical diagnosis, highlighting instances of erroneous diagnoses due to inadequate training data and a lack of explainability. It stresses the need for rigorous data curation and the cost implications of developing safe medical AI systems. The authors caution against overenthusiastic adoption of AI in medicine and call for awareness of these limitations among medical professionals and stakeholders, underscoring the importance of a cautious and resource-intensive approach to AI in healthcare.

The Fidelity of Artificial Intelligence to Multidisciplinary Tumor Board Recommendations for Patients with Gastric Cancer: A Retrospective Study.

Due to significant growth in the volume of information produced by cancer research, staying abreast of recent developments has become a challenging task. Artificial intelligence (AI) can learn, reason, and understand the enormous corpus of literature available to the scientific community. However, large-scale studies comparing the recommendations of AI and a multidisciplinary team board (MTB) in gastric cancer treatment have rarely been performed. Therefore, a retrospective real-world study was conducted to assess the level of concordance between AI and MTB treatment recommendations. Treatment recommendations of Watson for Oncology (WFO) and an MTB were retrospectively analyzed 322 patients with gastric cancer from January 2015 to December 2018 and the degree of agreement between them was compared. The patients were divided into concordance and non-concordance groups and factors affecting the concordance rate were analyzed. The concordance rate between the AI and MTB was 86.96%. The concordance rates for each stage were 96.93% for stage I, 88.89% for stages II, 90.91% for stage III, and 45.83% for stage IV, respectively. In the multivariate analysis, age (p-value = 0.000), performance status (p-value = 0.003 for performance score 1; p-value = 0.007 for performance score 2; p-value = 0.000 for performance score 3), and stage IV (p-value = 0.017) had a significant effect on concordance between the MTB and WFO. Factors affecting the concordance rate were age, performance status, and stage IV gastric cancer. To increase the validity of future medical AI systems for gastric cancer treatment, their supplementation with local guidelines and the ability to comprehensively understand individual patients is essential.

Artificial Intelligence in Healthcare- An Overview

Artificial intelligence (AI) has been developing fleetly in recent times in terms of software algorithms, tackle preparation, and operations in a vast number of areas. In this review, we epitomize the rearmost of operations of AI in biomedicine, including complaint diagnostics, living backing, biomedical information processing, and biomedical exploration. The end of this review is to keep track of new scientific accomplishments, to understand the vacuity of technologies, to appreciate the tremendous eventuality of AI in biomedicine, and to give experimenters in affiliated field’s alleviation. It can be asserted that, just like AI itself, the operation of AI in biomedicine is still in its early stage. New progress and improvements will continue to push the frontier and widen the compass of AI operations, and fast developments are envisaged in the near future.AI in healthcare is an umbrella term to describe the application of machine learning (ML) algorithms and other cognitive technologies in medical settings. In the simplest sense, AI is when computers and other machines mimic human cognition, and are capable of learning, thinking, and making decisions or taking actions. Artificial intelligence (AI) is gradationally changing medical practice. With recent progress in digitized data accession, machine literacy and computing structure, AI operations are expanding into areas that were preliminary allowed to be only the fiefdom of mortal experts. In this Review composition, we outline recent breakthroughs in AI technologies and their biomedical applications, identify the challenges for further progress in medical AI systems, and epitomize the profitable, legal and counteraccusations of AI in healthcare.

Transparency in Medical Artificial Intelligence Systems

Many of the artificial intelligence (AI) systems used nowadays have a very high level of accuracy but fail to explain their decisions. This is critical, especially in sensitive areas such as medicine and the health area at large but also for applications of the law, finance etc., where explanations for certain decisions are needed and are often useful and valuable as the decision itself. This paper presents a review of four different methods for creating transparency in AI systems. It also suggests a list of criteria under which circumstances one should use which methods.

Deployment of Real-time Natural Language Processing and Deep Learning Clinical Decision Support in the Electronic Health Record: Pipeline Implementation for an Opioid Misuse Screener in Hospitalized Adults.

The clinical narrative in electronic health records (EHRs) carries valuable information for predictive analytics; however, its free-text form is difficult to mine and analyze for clinical decision support (CDS). Large-scale clinical natural language processing (NLP) pipelines have focused on data warehouse applications for retrospective research efforts. There remains a paucity of evidence for implementing NLP pipelines at the bedside for health care delivery. We aimed to detail a hospital-wide, operational pipeline to implement a real-time NLP-driven CDS tool and describe a protocol for an implementation framework with a user-centered design of the CDS tool. The pipeline integrated a previously trained open-source convolutional neural network model for screening opioid misuse that leveraged EHR notes mapped to standardized medical vocabularies in the Unified Medical Language System. A sample of 100 adult encounters were reviewed by a physician informaticist for silent testing of the deep learning algorithm before deployment. An end user interview survey was developed to examine the user acceptability of a best practice alert (BPA) to provide the screening results with recommendations. The planned implementation also included a human-centered design with user feedback on the BPA, an implementation framework with cost-effectiveness, and a noninferiority patient outcome analysis plan. The pipeline was a reproducible workflow with a shared pseudocode for a cloud service to ingest, process, and store clinical notes as Health Level 7 messages from a major EHR vendor in an elastic cloud computing environment. Feature engineering of the notes used an open-source NLP engine, and the features were fed into the deep learning algorithm, with the results returned as a BPA in the EHR. On-site silent testing of the deep learning algorithm demonstrated a sensitivity of 93% (95% CI 66%-99%) and specificity of 92% (95% CI 84%-96%), similar to published validation studies. Before deployment, approvals were received across hospital committees for inpatient operations. Five interviews were conducted; they informed the development of an educational flyer and further modified the BPA to exclude certain patients and allow the refusal of recommendations. The longest delay in pipeline development was because of cybersecurity approvals, especially because of the exchange of protected health information between the Microsoft (Microsoft Corp) and Epic (Epic Systems Corp) cloud vendors. In silent testing, the resultant pipeline provided a BPA to the bedside within minutes of a provider entering a note in the EHR. The components of the real-time NLP pipeline were detailed with open-source tools and pseudocode for other health systems to benchmark. The deployment of medical artificial intelligence systems in routine clinical care presents an important yet unfulfilled opportunity, and our protocol aimed to close the gap in the implementation of artificial intelligence-driven CDS. ClinicalTrials.gov NCT05745480; https://www.clinicaltrials.gov/ct2/show/NCT05745480.

Levels of explicability for medical artificial intelligence: What do we normatively need and what can we technically reach?

Definition of the problemThe umbrella term “explicability” refers to the reduction of opacity of artificial intelligence (AI) systems. These efforts are challenging for medical AI applications because higher accuracy often comes at the cost of increased opacity. This entails ethical tensions because physicians and patients desire to trace how results are produced without compromising the performance of AI systems. The centrality of explicability within the informed consent process for medical AI systems compels an ethical reflection on the trade-offs. Which levels of explicability are needed to obtain informed consent when utilizing medical AI?ArgumentsWe proceed in five steps: First, we map the terms commonly associated with explicability as described in the ethics and computer science literature, i.e., disclosure, intelligibility, interpretability, and explainability. Second, we conduct a conceptual analysis of the ethical requirements for explicability when it comes to informed consent. Third, we distinguish hurdles for explicability in terms of epistemic and explanatory opacity. Fourth, this then allows to conclude the level of explicability physicians must reach and what patients can expect. In a final step, we show how the identified levels of explicability can technically be met from the perspective of computer science. Throughout our work, we take diagnostic AI systems in radiology as an example.ConclusionWe determined four levels of explicability that need to be distinguished for ethically defensible informed consent processes and showed how developers of medical AI can technically meet these requirements.

A large language model for electronic health records

There is an increasing interest in developing artificial intelligence (AI) systems to process and interpret electronic health records (EHRs). Natural language processing (NLP) powered by pretrained language models is the key technology for medical AI systems utilizing clinical narratives. However, there are few clinical language models, the largest of which trained in the clinical domain is comparatively small at 110 million parameters (compared with billions of parameters in the general domain). It is not clear how large clinical language models with billions of parameters can help medical AI systems utilize unstructured EHRs. In this study, we develop from scratch a large clinical language model—GatorTron—using >90 billion words of text (including >82 billion words of de-identified clinical text) and systematically evaluate it on five clinical NLP tasks including clinical concept extraction, medical relation extraction, semantic textual similarity, natural language inference (NLI), and medical question answering (MQA). We examine how (1) scaling up the number of parameters and (2) scaling up the size of the training data could benefit these NLP tasks. GatorTron models scale up the clinical language model from 110 million to 8.9 billion parameters and improve five clinical NLP tasks (e.g., 9.6% and 9.5% improvement in accuracy for NLI and MQA), which can be applied to medical AI systems to improve healthcare delivery. The GatorTron models are publicly available at: https://catalog.ngc.nvidia.com/orgs/nvidia/teams/clara/models/gatortron_og.

Against explainability requirements for ethical artificial intelligence in health care

It is widely accepted that explainability is a requirement for the ethical use of artificial intelligence (AI) in health care. I challenge this Explainability Imperative (EI) by considering the following question: does the use of epistemically opaque medical AI systems violate existing legal standards for informed consent? If yes, and if the failure to meet such standards can be attributed to epistemic opacity, then explainability is a requirement for AI in healthcare. If not, then based on at least one metric of ethical medical practice (informed consent), explainability is not required for the ethical use of AI in healthcare. First, I show that the use of epistemically opaque AI applications is compatible with meeting accepted legal criteria for informed consent. Second, I argue that human experts are also black boxes with respect to the criteria by which they arrive at a diagnosis. Human experts can nonetheless meet established requirements for informed consent. I conclude that the use of black-box AI systems does not violate patients’ rights to informed consent, and thus, with respect to informed consent, explainability is not required for medical AI.

Clinical Comparable Corpus Describing the Same Subjects with Different Expressions.

Medical artificial intelligence (AI) systems need to learn to recognize synonyms or paraphrases describing the same anatomy, disease, treatment, etc. to better understand real-world clinical documents. Existing linguistic resources focus on variants at the word or sentence level. To handle linguistic variations on a broader scale, we proposed the Medical Text Radiology Report section Japanese version (MedTxt-RR-JA), the first clinical comparable corpus. MedTxt-RR-JA was built by recruiting nine radiologists to diagnose the same 15 lung cancer cases in Radiopaedia, an open-access radiological repository. The 135 radiology reports in MedTxt-RR-JA were shown to contain word-, sentence- and document-level variations maintaining similarity of contents. MedTxt-RR-JA is also the first publicly available Japanese radiology report corpus that would help to overcome poor data availability for Japanese medical AI systems. Moreover, our methodology can be applied widely to building clinical corpora without privacy concerns.

Identity Threats as a Reason for Resistance to Artificial Intelligence: Survey Study With Medical Students and Professionals.

BackgroundInformation systems based on artificial intelligence (AI) have increasingly spurred controversies among medical professionals as they start to outperform medical experts in tasks that previously required complex human reasoning. Prior research in other contexts has shown that such a technological disruption can result in professional identity threats and provoke negative attitudes and resistance to using technology. However, little is known about how AI systems evoke professional identity threats in medical professionals and under which conditions they actually provoke negative attitudes and resistance.ObjectiveThe aim of this study is to investigate how medical professionals’ resistance to AI can be understood because of professional identity threats and temporal perceptions of AI systems. It examines the following two dimensions of medical professional identity threat: threats to physicians’ expert status (professional recognition) and threats to physicians’ role as an autonomous care provider (professional capabilities). This paper assesses whether these professional identity threats predict resistance to AI systems and change in importance under the conditions of varying professional experience and varying perceived temporal relevance of AI systems.MethodsWe conducted 2 web-based surveys with 164 medical students and 42 experienced physicians across different specialties. The participants were provided with a vignette of a general medical AI system. We measured the experienced identity threats, resistance attitudes, and perceived temporal distance of AI. In a subsample, we collected additional data on the perceived identity enhancement to gain a better understanding of how the participants perceived the upcoming technological change as beyond a mere threat. Qualitative data were coded in a content analysis. Quantitative data were analyzed in regression analyses.ResultsBoth threats to professional recognition and threats to professional capabilities contributed to perceived self-threat and resistance to AI. Self-threat was negatively associated with resistance. Threats to professional capabilities directly affected resistance to AI, whereas the effect of threats to professional recognition was fully mediated through self-threat. Medical students experienced stronger identity threats and resistance to AI than medical professionals. The temporal distance of AI changed the importance of professional identity threats. If AI systems were perceived as relevant only in the distant future, the effect of threats to professional capabilities was weaker, whereas the effect of threats to professional recognition was stronger. The effect of threats remained robust after including perceived identity enhancement. The results show that the distinct dimensions of medical professional identity are affected by the upcoming technological change through AI.ConclusionsOur findings demonstrate that AI systems can be perceived as a threat to medical professional identity. Both threats to professional recognition and threats to professional capabilities contribute to resistance attitudes toward AI and need to be considered in the implementation of AI systems in clinical practice.

Promoting the application of federated learning in medical imaging artificial intelligence

Medical image-based artificial intelligence (AI) systems have shown great potential in assisting disease diagnosis and treatment. However, the challenges, such as data silos, privacy security and standardization, seriously impedes the application of AI in disease diagnosis and treatment. By integrating federated learning technology and FAIR data principle, it is possible to resolve the aforementioned obstacles. Then, it is able to maximize the value of multicenter data to develop a more efficient and accurate disease diagnosis and treatment AI systems, and promote the clinical application of medical image-based AI systems in the field of disease diagnosis and treatment.

Liability from the use of medical artificial intelligence: a comparative study of English and Taiwanese tort laws

Background: Modern artificial intelligence applications are appearing in healthcare and medical practices. Artificial intelligence is used both in medical research and on patients via medical devices. The aim of this paper is to examine and compare English and Taiwanese tort laws in relation to medical artificial intelligence. Methods: The methodologies employed are legal doctrinal analysis and comparative law analysis. Results: The investigation finds that English tort law treats wrong diagnostic or wrong advice as negligent misstatement, and mishaps due to devices as a physical tort under the negligence rule. Negligent misstatement may occur in diagnosis or advisory systems, while a negligent act may occur in products used in the treatment of the patient. Product liability under English common law applies the same rule as negligence. In Taiwan, the general principles of tort law in Taiwan’s Civil Code for misstatement and negligent action apply, whereas the Consumer Protection Act provides for additional rules on product liability of traders. Conclusions: Safety regulations may be a suitable alternative to tort liability as a means to ensure the safety of medical artificial intelligence systems.

Designing medical artificial intelligence for in- and out-groups

Medical artificial intelligence (AI) is expected to deliver worldwide access to healthcare. Through three experimental studies with Chinese and American participants, we tested how the design of medical AI varies between in- and out-groups. Participants adopted the role of a medical AI designer and decided how to develop medical AI for in- or out-groups based on their experimental condition. Studies 1 (pre-registered: N = 191) revealed that Chinese participants were less likely to adopt human doctors' assistance in medical AI system when targeting patients from US (i.e., out-groups) than for patients from China (i.e., in-groups). Study 2 (N = 190) revealed that US participants were less likely to adopt human doctors' assistance in medical AI system when targeting patients from China (i.e., out-groups) than for patients from US (i.e., in-groups). Study 3 revealed that Chinese medical students (N = 160) selected a smaller training database for AI when diagnosing diabetic retinopathy among US patients (i.e., out-groups) than for Chinese patients (i.e., in-groups), and this effect was stronger among medical students from higher (vs. lower) socioeconomic backgrounds. This AI design inequity was mediated by individuals’ underestimation of out-group heterogeneity. Overall, our evidence suggests that out-group stereotype shapes the design of medical AI, unwittingly undermining healthcare quality. The current findings underline the need for more robust data on medical AI development and intervention research addressing healthcare inequity.

Lessons Learned from Designing an AI-Enabled Diagnosis Tool for Pathologists

Despite the promises of data-driven artificial intelligence (AI), little is known about how we can bridge the gulf between traditional physician-driven diagnosis and a plausible future of medicine automated by AI. Specifically, how can we involve AI usefully in physicians' diagnosis workflow given that most AI is still nascent and error-prone (\eg in digital pathology)? To explore this question, we first propose a series of collaborative techniques to engage human pathologists with AI given AI's capabilities and limitations, based on which we prototype Impetus --- a tool where an AI takes various degrees of initiatives to provide various forms of assistance to a pathologist in detecting tumors from histological slides. We summarize observations and lessons learned from a study with eight pathologists and discuss recommendations for future work on human-centered medical AI systems.

Critical evaluation of deep neural networks for wrist fracture detection

Wrist Fracture is the most common type of fracture with a high incidence rate. Conventional radiography (i.e. X-ray imaging) is used for wrist fracture detection routinely, but occasionally fracture delineation poses issues and an additional confirmation by computed tomography (CT) is needed for diagnosis. Recent advances in the field of Deep Learning (DL), a subfield of Artificial Intelligence (AI), have shown that wrist fracture detection can be automated using Convolutional Neural Networks. However, previous studies did not pay close attention to the difficult cases which can only be confirmed via CT imaging. In this study, we have developed and analyzed a state-of-the-art DL-based pipeline for wrist (distal radius) fracture detection—DeepWrist, and evaluated it against one general population test set, and one challenging test set comprising only cases requiring confirmation by CT. Our results reveal that a typical state-of-the-art approach, such as DeepWrist, while having a near-perfect performance on the general independent test set, has a substantially lower performance on the challenging test set—average precision of 0.99 (0.99–0.99) versus 0.64 (0.46–0.83), respectively. Similarly, the area under the ROC curve was of 0.99 (0.98–0.99) versus 0.84 (0.72–0.93), respectively. Our findings highlight the importance of a meticulous analysis of DL-based models before clinical use, and unearth the need for more challenging settings for testing medical AI systems.