From gastric cancer prevention to global health care, this is the way to implementing artificial intelligence in medicine

Abstract

The use of artificial intelligence (AI) models for gastric cancer prevention shows promise beyond traditional diagnostic approaches. Modern digital technologies can be applied to global healthcare. The implementation of deep machine learning has significantly increased the efficiency of computer vision. The introduction of AI is very effective in the field of diagnostic recognition of pathology in endoscopic, radiological and histological images. Robotic surgery has good development prospects. Also, many methods of treating diseases are implemented using AI technologies. Comparative studies of the effectiveness of a human doctor and AI have a high level of evidence. Comparative studies report high levels of performance for AI vs human clinicians in selected tasks, although evidence varies by task and dataset. AI demonstrates higher efficiency than 5-10 highly qualified experts in many areas of medicine. The main areas of medicine actively use AI: Diagnostic recognition of X-ray computed tomographic and magnetic resonance images, endoscopic and histological patterns. The use of AI has begun in the field of targeted treatment. The development of robot-associated surgery continues. There is a good prospect for using AI not only for recognizing X-ray computer images, in endoscopy, histology and targeted treatment, but also for subjective methods of examining patients. For example, the development of AI models for questioning patients by correspondence or conversation between the interlocutor - a doctor and the interlocutor-a patient has begun. The number of assistants (interlocutors, digital agents) can be more than two. The language of communication can also be any.

Key Words: Artificial intelligence; Deep machine learning; X-ray computer images; Magnetic resonance images; Endoscopic; Histological patterns; Targeted treatment; Robot-associated surgery

Core Tip: Modern digital technologies can be applied in global healthcare. The introduction of deep machine learning has significantly increased the effectiveness of computer vision. Pathology recognition in endoscopic, radiological, and histological images holds promise in the diagnosis of internal diseases. Physical diagnostic methods (visual examination, palpation, percussion, and auscultation) using artificial intelligence (AI) have demonstrated high diagnostic effectiveness. Comparative studies of the effectiveness of human physicians and AI have a high level of evidence. Development of AI models for interviewing patients via correspondence or in a conversational format between a physician and a patient has begun.

INTRODUCTION

The topic of developing, implementing, and using digital technologies is rapidly spreading across all areas of healthcare and medical science. The rapid growth in the number of publications on artificial intelligence (AI) in medicine demonstrates that global healthcare is poised for the progressive implementation and use of machine learning models and deep learning in preventive, diagnostic, therapeutic, and prognostic medical technologies[1,2].

POSSIBILITIES OF IMPLEMENTING MACHINE LEARNING IN MEDICINE AT THE PRESENT STAGE

AI is being rapidly developed and integrated into clinical practice. Few areas of medicine remain untouched by AI. Therapeutics, surgery, diagnostics, medical statistics, health management, biomedical research, and medical education increasingly employ digital technologies, including convolutional, recurrent, and graph neural networks[3]. AI facilitates improvements in the efficiency and quality of medical care across diverse healthcare systems and supports the advancement of personalized and preventive medicine[4]. Although AI applications in colorectal surgery remain preliminary, AI-enhanced robotic surgery has the potential to transform surgical practice in the coming years, as suggested by recent systematic reviews[5]. Nevertheless, AI-assisted robotic systems do not supplant the clinician–patient relationship; procedures such as spinal surgery continue to require meticulous perioperative care and empathic communication, which influence postoperative recovery[6]. AI can assist decision-making in emergency surgery, yet the deployment of digital technologies in emergency settings raises complex legal and ethical issues. Clinical decisions in the operating room and at the bedside are unlikely to be made without physician oversight[7]. The transformation of step-by-step methodologies with input from the most experienced experts plays a primary role during revolutionary changes. International professional societies can establish consensus committees to transform methodology and develop protocols, practical recommendations, indications and contraindications, and their systematic updates while considering legal and ethical standards for implementing digital technologies in medicine[8]. There is extensive practical experience in robotic surgery in pediatrics and urology[9,10]. The advantages of robotic surgery over laparoscopic liver surgery remain debated. With improved tactile feedback, AI will further enhance robotic capabilities in the future[11]. The number of scientific publications on robotic surgery grows annually, with the United States leading in publications and citations. Leading clinical areas for robotic surgery include gastrointestinal, gynecological, pulmonary, and thyroid diseases. AI combined with robotic surgery will become a key trend in modern surgical methodology[12]. AI is in high demand for training residents in robotic surgery methodology[13]. Robotic endovascular surgery often replaces traditional endovascular technologies in cardiac surgery, including atrial septal defect repair, left atrial myxoma resection, and mitral valve repair[14]. The methodology for implementing AI in robotic surgery includes a standard set of tools. As in other areas, computer vision for image recognition is carried out using deep learning. Choksi et al[15] used 211 videos to train a robot; the model was tested on ten robotic inguinal hernia procedures[15]. Robotic-assisted lymphatic supermicrosurgery has gained popularity recently, offering advantages in dexterity, precision, and ergonomics[16]. There is growing interest in AI-assisted robotic approaches in neurosurgery. A systematic review of computer vision methodologies found frequent use of convolutional neural networks (65%); recurrent neural networks and hybrid architectures also demonstrate excellent results[17]. AI and robot-assisted surgery have shown high efficiency in transplantology. Modern robotic digital technologies can be used not only for the most complex surgical operations in transplantology, but can also be used for training transplant surgeons[18]. The innovative robotic dental implant system is successfully implemented in oral implant surgery. The robotic arm of the EC66 model has six degrees of freedom. It operates using optical navigation technology and digital correction of the implant direction. The robotic arm is carefully calibrated. The eyes of the robotic system are also calibrated. Optimized calibration methods allow for increased angular accuracy and therefore the innovative robotic system is more efficient than a human hand[19]. Pharmacotherapy methods are being transformed by AI implementation. Challenges in drug discovery stem from multifactorial heterogeneity, including genetic polymorphism and diverse genetic risks. AI based on deep learning significantly facilitates the search for genetic targets. Promising results are expected in pharmacotherapy for arterial hypertension and drug discovery for dementia[20,21]. AI shows potential for predicting treatment outcomes in breast cancer[22] and atrial fibrillation[23]. In the pharmaceutical sector, AI reduces drug development costs, enables prediction of therapeutic outcomes, models pharmacokinetic profiles, and assesses drug toxicity[24]. AI has great potential to improve personalized treatment, enhance targeted pharmacotherapy, and individualize radiotherapy dosing[25]. Treatment stratification and gene therapy should follow personalized-medicine principles. AI may effectively support new treatments for chronic lung diseases, gene therapies for cystic fibrosis, chronic obstructive pulmonary disease, asthma, idiopathic pulmonary fibrosis, and acute respiratory distress syndrome[26]. AI-based image analysis for recognizing polyploid giant cancer cells in breast cancer has been successfully used to predict chemoresistance, metastasis, and survival based on biomarker analysis[27]. Using AI to identify oncogenic variants improves targeted therapy effectiveness in precision oncology[28]. Integrating AI into clinical decision support systems enhances strategies to overcome antibiotic resistance during antibacterial therapy in severe infectious and septic diseases[29]. A methodology combining AI-guided analytics and nanotechnology enables better personalization of postoperative pain management protocols in spinal surgery[30]. AI has made significant progress in the treatment of surgical and medical patients, and even greater progress has been achieved in disease diagnosis.

USING AI IN DISEASE DIAGNOSTICS AND REHABILITATION

The methodology of diagnostic image recognition obtained during X-ray, magnetic resonance imaging, computed tomography, endoscopic, and ultrasound examinations, as well as electrocardiography and other monitoring methods, has changed significantly due to the introduction of AI, particularly deep learning. Diagnostic accuracy has increased substantially, and predictive modeling of disease outcomes has become possible[31,32]. AI also has potential in diagnostics using positron emission tomography[33]. The use of AI substantially improves the diagnostic accuracy of panoramic tomographic examinations in dentistry[34]. AI, based on deep learning, significantly improves the accuracy of retinal optical coherence tomography analysis for cognitive impairment and neurodegenerative diseases, and may help monitor disease progression[35,36]. To improve the accuracy of ophthalmic diagnostics and glaucoma detection, it is necessary to use AI models[37]. Rehabilitation medicine aims to restore patients’ functions after various treatments and is especially important for postoperative care. Experience with AI-enabled robotic rehabilitation is likely to increase the effectiveness of rehabilitation medicine.

USING AI IN SUBJECTIVE METHODS OF PATIENT EXAMINATION

The use of AI has significant potential for interviewing patients to obtain information about the patient’s complaints, medical history and life history (medical biography). Medical interviews allow to reduce the time for questioning each patient and increase the accuracy of diagnosis[38,39]. AI-powered chatbot model GPT-4 can effectively interview patients on complaints and medical history. The GPT-4 model has the potential to be a versatile tool for use in various medical disciplines after deep learning[40]. The ChatGPT AI dialog model is a very effective way of maintaining medical records, including subjective methods of patient examination[41]. There is already successful experience using ChatGPT for assessing low back pain[42]. The ChatGPT can be successfully used for dynamic monitoring of a patient with diabetes mellitus[43]. The ChatGPT-4 model has the potential to be adapted to non-English speaking patients, in this sense it can be adapted to other languages[44]. A tablet computer can be a tool for patient communication with AI. Most patients prefer to use a tablet computer than to fill out a paper questionnaire[45]. The use of tablet computers by patients to collect medical anamnesis will make it much easier for doctors to obtain medical records[46]. The authors of a systematic review strongly support the use of AI for subjective assessment of patients in primary care[47]. AI models learn speech and text linguistic features, allowing them to be used in different languages[48]. Collecting patient information using a tablet computer is a widely used tool[49,50]. The results of many studies of conversational agents are characterized by contradictory results. Two-thirds of the studies have shown positive results from using AI models to dialogue with patients. There was no consensus on the effectiveness of different methods of patient dialogue and AI. However, the results of studying the capabilities of conversational agents can be assessed as promising for use in subjective diagnostics[51]. Voice chatbots have been successfully used during the Covid-19 pandemic. According to the authors, this experience of using voice AI was effective and high-quality in serving patients[52]. Conversational AI has been shown to be a positive force in providing care to patients with atrial fibrillation[53].

USING AI IN CLINICAL OBJECTIVE METHODS OF PATIENT EXAMINATION

General external examination (status praesens objectivus)

External examination is the cornerstone of the traditional physical examination. AI models employing computer vision can perform many components of the examination effectively. Routine clinical techniques can be supplemented or partly replaced by various sensors that complement existing medical practice. Such systems can capture traditional physical-examination maneuvers and assess phenotypic features (e.g., gait and posture). AI will facilitate the transition between in-office and remote monitoring by detecting changes in a patient’s clinical status in a timely manner[54].

Palpation

Based on the CASIT tactile-feedback system developed at the University of California, Los Angeles, various AI models can be developed to implement traditional patient palpation techniques. The CASIT system includes tactile sensors, a tactile controller, a driver, and actuators; using these components, it can determine the texture, rigidity, compliance, shape, and edges of tissues and organs[55].

Percussion

Ayodele et al[56] developed a device that can receive medical percussion sounds and analyze them using the MobileNetv2 convolutional neural network.

Auscultation

Auscultation is one of the primary methods of physical examination. AI-assisted auscultation has significant potential to improve diagnostic accuracy for many conditions. The DigitaLung project, a system employing deep learning and neural networks for lung auscultation, has the potential to enhance differential diagnosis in pulmonology[57]. An AI-enabled electronic stethoscope could substantially improve diagnostic efficiency in cardiology and cardiac surgery[58,59].

USING AI IN LABORATORY METHODS OF PATIENT EXAMINATION

The fundamentals of machine learning are being actively introduced into laboratory diagnostics, extending to routine clinical laboratory tests, biochemical analyses, and modern molecular-genetic studies[60,61]. Contemporary highly automated laboratory diagnostics, whose future development is closely linked to digital technologies, are likely to employ AI. The pace of development and implementation of future AI models in laboratory diagnostics depends on the intensity of cooperation between information-technology specialists and laboratory diagnosticians[62]. There is strong potential for AI in chemical laboratory research[63], and high expectations for its use in microbiological investigations[64]. Personalized medicine based on diagnostic genomics is impractical without AI, since the analysis of whole-genome sequencing data involves very large volumes of information[65]. AI is poised to revolutionize clinical laboratory diagnostics in line with the demands of personalized medicine[66,67]. According to participants at the Clinical Microbiology Conference held in Oceanside, California, on 1-2 February 2024, laboratory diagnostics using AI represents the “Laboratory of the Future”[68]. A revolutionary development is AI-assisted karyotyping in cytogenetics laboratories, which alters the paradigm of chromosome-analysis methodology[69].

AI in instrumental diagnostic methods

Technical progress in medical imaging for instrumental diagnostic examinations has achieved significant advances. Diagnostic methodology has been transformed by applying AI to modern visualization techniques for diverse pathological processes. This primarily concerns computed tomography and magnetic resonance imaging. AI is also used in diagnostic modalities such as electrocardiography, electroencephalography, electromyography, and ultrasonography[70-72]. The greatest diagnostic gains have been reported for AI-assisted endoscopic detection of pathological changes in the stomach[73-77] and colon[78,79]. AI applications in histopathological imaging have likewise proven highly effective[80-82].

DISCUSSION

Although experts predict a 5%-10% reduction in United States healthcare costs as a result of implementing AI in medicine, patient attitudes across 43 countries were mixed. The majority of respondents expressed a favorable view of digital technologies in medicine, but women were less positive than men: 73% preferred decisions to be made by a doctor[83]. The use of AI in diagnostics significantly increases the potential for predicting disease outcomes, treatment responses, and dynamic patient monitoring[84]. The development of personalized medicine and progress in robotic surgery depend largely on the extent of AI implementation[85-87]. In the future, AI technologies and models will cover all areas of medical activity, including biomedical research. At the same time, emotional and legal aspects of interactions between clinicians, patients, and AI must be carefully regulated[88-90]. A proposed general model of primary contact with a diagnostic algorithm is shown in Figure 1. According to the European Medicines Agency (EMA), leveraging the benefits of AI in healthcare requires significant support while simultaneously ensuring safety. Risks related to accountability, confidentiality, and bias must also be mitigated. The establishment of international-level training centers for effective experimentation is recommended as a promising way to overcome these contradictions. The operating principles of these training centers should be based on the support and development of AI products in medicine, in full compliance with legal and ethical regulations. The United States Food and Drug Administration (FDA) shares a similar view. The FDA emphasizes the need for strict oversight to ensure the safety, efficacy, and reliability of the use of AI models in medicine. To achieve promising goals, cause-and-effect relationships must be subject to feedback loops. Experimentation stimulates the development of a research program on tools and technologies. International collaboration will stimulate the creation of training programs and collaboration platforms based at training centers[91,92]. Modern personalized medicine is unthinkable without AI. The implementation of digital technologies by dividing patients into groups with similar characteristics will enable accurate diagnostics, targeted treatment, and prevention in a personalized medicine setting. Patient stratification using mechanistic models and machine learning will significantly facilitate diagnosis and treatment in personalized medicine. AI models make it possible to effectively structure the physiological mechanisms of diseases and identify patterns in large data sets. Model validation is crucial in this context. Collin et al[93] recommend the SPIRIT-AI and CONSORT-AI checklists as guidelines for reporting clinical trial data for digital models. They also recommend the AIMe registry for AI in biomedical research (https://aime-registry.org) to help developers of new machine learning models navigate data, methods, and reproducibility. The predictive capabilities of clinical medicine will significantly increase within the context of personalized medicine using digital technologies. For example, the introduction of a cancer signaling network atlas into practical oncology, which describes in detail the molecular and genetic mechanisms of cancer, will significantly improve the effectiveness of cancer prevention[93]. I would like to express my confidence that the recommendations proposed by Collin et al[93], as well as the recommendations of the EMA and FDA, will contribute to the development of precision digital medicine in the future. Human participation in digitally enabled systems offers promising potential for implementation in medicine. To implement such projects, Räz et al[94] propose a roadmap for the future of explainable AI in medicine. Explainable AI is constrained by two requirements. Explanatory information must be accurate and relevant. Explanations must also take into account the user’s level of professional training. For example, the level of a physician researcher and the level of a nurse require different explanations. A preliminary diagnosis generated by explainable AI, based on patient complaints, medical history, and physical data obtained using various sensors, must be explained in a way that takes into account the physician's experience and specialization. The use of language models will allow for the extraction and concise presentation of valuable patient information contained in voluminous medical records, thereby reducing the physician's cognitive load. The key question in this context is whether a full-fledged dialogue between humans and AI is possible. The use of large language models, such as ChatGPT, allows for a positive answer to this question. At this stage, such models are still inferior to traditional medical models. The ability of AI to analyze facial expressions, tone of voice, and communication characteristics allows us to assume a good dialogue prospect between the patient, doctor, and AI[94].

Figure 1 Future model of preventive, diagnostic and therapeutic prehospital care. AI: Artificial intelligence.

CONCLUSION

The diagnostic and therapeutic activities of the doctor are in competitive relations with the use of AI. Each of them has its own advantages and disadvantages. As a result of interaction, progress in medicine will become very dynamic.