Prospects for the use of big data, artificial intelligence, machine learning, neural networks, and deep learning in the diagnosis and treatment of malignant tumors of the genitourinary system: a review

全文:

Introduction. The term "big data" in scientific literature primarily denotes a large volume of information accumulated in the Internet space [1]. Recently, in Russian and foreign sources, one can increasingly come across the statement that this definition also implies the latest technologies that analyze and process information accumulated in the global digital space [2]. "Big data systems consist of components for information extraction, preliminary processing, reception and integration, data analysis, interface and visualization" [3]. It should be noted that medicine was the first to turn its attention to digital information technologies, as it was in dire need of systematization and storage of large amounts of information [4]. Currently, the digitalization of medicine is rapidly increasing, which makes it possible to both accumulate and systematize medical information material in the global digital field [5]. Big data leads to the need to find semantic categories in the created information space, which is constantly changing, and to find relationships between the data obtained. Understanding these relationships and the ability to extract information hidden in big data is a top priority in working with Big Data [6]. To manage diverse medical information in real time, advanced tools are needed that can consolidate, extract, analyze and visualize diverse incoming information. Traditional methods and techniques, as well as familiar software, no longer meet the high requirements of modern medicine. The main tool for working with Big Data is artificial intelligence (AI) or a computer system that imitates the ability of people to learn and solve problems. The larger the structured database becomes, the more powerful the AI ​​becomes. Since 2015, about ninety percent of all the world's information has been generated [7]. Machine learning (ML) is a part of artificial intelligence (AI), a learning model based on configured algorithms based on data. Neural networks are hardware or software mathematical models that imitate the work of neural networks of living organisms. In essence, it is a type of machine learning. Neural networks use one algorithm in their work. They absorb a huge amount of information, process it, analyze it, pass it through a network of artificial neurons and generate their answer to the task set by the researcher. Deep learning (DL) is an advanced type of ML. DL uses a large number of neural networks. They create a more powerful, intelligent and productive computer system than a single neural network [8]. What is important to note is that DL uses a mathematical algorithm to improve learning based on experience. Deep learning helps to interpret the information dataset, as well as guide the doctor to make accurate decisions. Currently, artificial intelligence, using mathematical algorithms to imitate human cognitive abilities, is able to solve such complex healthcare problems as diagnosis, treatment and prevention of cancer. The exponential growth of artificial intelligence over the past decade indicates that it is a potential platform for making optimal decisions with the help of superintelligence, when the human mind is limited to processing huge data in a narrow time range [9]. The purpose of the study was to provide a broad overview of the current state of artificial intelligence (AI) tools for decision-making, diagnosis, choice of treatment tactics and prediction of results in oncourological practice. It is necessary to show how the latest AI-based technologies provide undoubted effective assistance in the diagnosis, treatment and prevention of cancer, giving hope for a cure to many people. 1. Prospects and problems of using big data and artificial intelligence in precision and accurate medicine in oncourology Big data is represented by a combination of three components: volume, velocity and variety. Therefore, in the literature you can find the definition of Big data as the unity of three Vs [10]. In the field of oncology, big data includes many sources of information (volume): diagnostic, including clinical, radiological, pathomorphological; data on surgical intervention, systemic therapy, radiation therapy, information on the body's reaction, complications, etc. Big data also includes patient medical histories, patient vital signs obtained both from sensors in real time and entered into electronic medical histories [11]. It is interesting to note that the measurement results can come from the patients themselves, who were recorded using various computer applications or mobile devices [12].  A large percentage of big data is scientific research information, regulatory and legal information, medical insurance information, etc. In the information field, information is collected and systematized at a tremendous speed (velocity). Their significant diversity and heterogeneity require strict standardization and unambiguous definitions (variety) [13]. A number of researchers believe that in oncology, the main data are those obtained from the patient and presented in digital form. They include various information: gender, age, family history, clinical signs, treatment and concomitant diseases. An important component is the results of instrumental diagnostic methods: X-ray studies CT, MRI, PET-CT, ultrasound, results of fluid and tissue tests: histopathological, immunohistochemical, DNA and RNA sequencing, blood tests, etc. The second thing that researchers note is that big data includes computational analysis of patient information obtained by processing the received parameters. They include radiomics and digital image analysis, as well as genetic expression and mutation analysis [14]. Patient data is processed using machine learning and usually contains large computer files of structured materials. Big data contains a lot of information that comes from scientific and practical medical periodicals and other Internet sources. The authors emphasize that, according to statistics, a huge amount of information is stored in big data per cancer patient compared to data taken from a small cohort of patients typed by a certain type of cancer [15]. Medical information comes from numerous sources in various formats using all sorts of terms and in different languages, which is one of the main difficulties for generation. Due to the heterogeneity of formats and the lack of a common vocabulary, the availability of big data for medical data analysis and decision support systems causes certain difficulties [16]. Therefore, special vocabulary and visual standards for describing clinical signs and pathologies, various procedures, drugs, "logical observation identifier codes (LOINC), International Classification of Diseases (ICD9 and ICD10), Systematized Nomenclature of Medical and Clinical Terms (SNOMED-CT), Current Procedural Terminology, 4th edition (CPT 4), ATC - Anatomical Therapeutic Chemical Classification of Drugs, Gene Ontology (GO) and others" were developed [17]. In the Russian Federation, the integration of medical information is one of the most popular and complex tasks that must be solved in the near future. To implement the informatization of healthcare, developers of Russian information systems must take into account international standards, which will greatly simplify the collection and analysis of the world's medical information array [18]. Special attention is focused on the development of tools for working with big data. Significant progress has been noted in this area. Special algorithms have been created for complex analysis, which allow interpreting incoming information in real time. A set of predictive analytics methods Advanced analytics helps to identify various pathological changes or deviations in indicators and associate them with certain patterns, thus helping to detect the disease in the early stages of the pathological process [19]. For example, the use of an evolutionary built-in algorithm in whole-genome sequencing of DNA of cells obtained from blood plasma contributed to the detection of colorectal cancer in the early stages of the disease [20, 21]. It should be noted that with the help of Big Data technologies it is possible to predict the outcome of an oncological disease against the background of combining clinical data and genomic analysis [22]. Assessing the risk of cancer can be facilitated by combining large sets of genomic data and environmental information. In this case, big data will become useful in developing and changing disease prevention strategies, new approaches aimed at improving the environment and changing the behavior of people in risk groups. In oncology practice, big data is necessary to monitor the effectiveness of various treatment methods, such as chemotherapy or immunotherapy. This will help to increase the degree of personalization of medicine and replenish the necessary knowledge on the effectiveness of various treatment regimens [15]. Clinical oncology with the help of AI has recently been supplemented with new molecular strategies. One of them is next-generation sequencing or parallel sequencing (NGS), which allows for one-time reading of the nucleotide sequence of DNA and RNA in many parts of the genome. This platform has a high throughput and is designed to generate a large amount of information. Novikova E.I. and Snigireva G.P. in their work refer to a number of available technologies for NGS (manufacturers are given in brackets): HiSeq, MiSeq and NextSeq 500 (Illumina), Roche 454 GS, Ion torrent (Thermo Fisher Scientific) and SOLiD (Applied Biosystems) [23]. The obtained genetic results are processed using special software, which makes it possible to identify germline and somatic mutations in target genes and discover new and rare genetic variants associated with oncological diseases. Developers of NGS technologies offer computer programs that identify and classify genetic pathologies, and also provide information on their clinical significance. The results of next-generation sequencing allow the algorithm to offer the most effective therapy taking into account individual genetic factors, since precision anti-cancer drugs are aimed at a specific type of cancer cells, taking into account their genetic variability [24]. In their works, the researchers provide examples of using NGS sequencing in combination with AI to search for somatic and germline mutations associated with oncopathology. Thus, in urothelial bladder cancer, about 55 mutations were detected using this technology. Of these, 49 were registered for the first time [23]. It is worth paying attention to a promising direction in precision medicine - radiomics or virtual biopsy, which combines several areas of knowledge. Radiomics is primarily based on tumor visualization using radiation diagnostics. The obtained radiological images of a malignant tumor are processed using mathematical modeling and deep learning. Digital image analysis (grayscale and volumetric spatial arrangement of pixels and voxels) of the texture of various tissue pathologies is used to create image biomarkers [25]. In cancer diagnostics, they are used to assess the homogeneity of the tumor tissue under study throughout its depth. Image parameters quantitatively determine the phenotypic characteristics of the entire tumor. This solves two very important problems in oncology - early diagnosis of complex heterogeneous tumors and, as a result, more accurately selected treatment [26]. Radiomics extracts and uses a large number of medical imaging characteristics in a non-invasive and cost-effective way. In precision oncology and cancer treatment, it allows the use of prognostic and reliable machine learning methods for a personalized approach to each patient. It is possible to digitize and analyze not only radiological images but also histological preparations using a neural network. Cellular components are stained in a certain way on the thinnest section, and then analyzed under a microscope. The extent and nature of pathological changes in the cells are also determined on the preparation. These features are recorded on a digital slide, which makes it possible to study them further. High-precision neural networks and innovative software help the doctor analyze the digital slide made from the histological section. Neural networks automatically mark the tissue and write a potential diagnosis, which significantly speeds up the diagnosis process, makes it more accurate and reduces the risk of medical error by tens of times [27]. In addition, the doctor can view the digital image by moving along the slide on the monitor screen thanks to WSI technology - visualization of the entire slide. In prostate cancer (PC), after digitizing the image of the histological section, the neural network accurately determines the Gleason score, which has great prognostic value. It should be noted that the histological section has so-called zones of interest in which pathological changes are most pronounced. The neural network marks the image and indicates the characteristic signs of a certain condition [28]. The accuracy and quality of cancer diagnosis and treatment is growing as computer technology and biotechnology develop. The greatest achievements have been achieved in the field of genomic technologies. They contribute to the development of targeted and precision medicine, which allows for the adaptation of treatment to the personalized data of each patient, according to their genetic, biomarker and phenotypic characteristics. At the same time, they improve predictive medicine for the entire subpopulation of people with similar characteristics [29]. In the Russian Federation, “the state policy in the field of precision medicine is reflected in the Concept of predictive, preventive and personalized medicine aimed at developing individual approaches to the patient, including before the development of diseases” (Order of the Ministry of Health of Russia dated April 24, 2018, No. 186). 

The main goal of precision medicine is the ability to unite patients by identical indicators. Based on this, it is possible to more accurately predict the onset of the disease and its outcome, as well as correctly select strategies for treating oncological diseases [30]. Today, big data and artificial intelligence create the basis for daily diagnostics, biomedical research and treatment of cancer patients. They provide unprecedented potential for improving the quality and effectiveness of medical care. 2. The role of artificial intelligence in the diagnosis and treatment of malignant tumors of the genitourinary system. Artificial intelligence is beginning to be actively used in all areas of oncourology. The rapid growth of publications over the past decade in this area convincingly testifies to this. Of great interest is the world scientific information on the achievements in this field of medicine, the goals and objectives that it sets for itself, as well as their innovative solutions. Prostate cancer (PC). It is natural that the greatest amount of information was analyzed on prostate cancer, which is one of the most common oncopathologies in men both in Russia and abroad. Therefore, the tasks of early diagnostics and timely and accurate treatment are the highest priority. Let us note the innovations that have become available with the advent of artificial intelligence and the latest technologies in oncourological practice over the past ten years. PSA screening for prostate cancer was optimized at the State Budgetary Educational Institution of Higher Professional Education “Saratov State Medical University named after V.I. Razumovsky” of the Ministry of Health of the Russian Federation in the city of Saratov. Scientists decided to create an artificial neural network (ANN) to predict transrectal biopsy of the prostate gland. A whole series of scientific works were carried out [31]. AI revealed the possibilities of determining the risk group according to D'Amico in the presence of adenocarcinoma. The authors were also faced with the task of assessing the prognostic effectiveness of the neural network. A three-layer perceptron served as a format for designing the ANN. The artificial neural network was trained on the characteristics obtained from 398 patients who underwent diagnostics at the S.R. Mirotvortsev Clinical Hospital of the SSMU. All men underwent transrectal prostate biopsy. The input variables for training the ANN were: patient age, total and free serum PSA levels, PSA kinetics (doubling time and growth rate), prostate volume and PSA density, hypoechoic zones during transrectal ultrasonography, and changes detected during digital rectal examination. The output parameters were: the number of tissue columns with adenocarcinoma, Gleason scores, and the expected histological conclusion. The validation group for testing the prognostic efficiency of the ANN included 88 patients. The ANN algorithm developed by the scientists confirmed its high efficiency in predicting adenocarcinoma (the accuracy was about 94%, the sensitivity was about 98%). The predicted and real risk according to D'Amico coincided by 70%. This ANN model can be used to differentiate patients by risk groups before invasive biopsy [32]. Attempts to create a neural network for prostate cancer prediction and treatment have been made repeatedly. The work of I.R. Ayupov and co-authors, who developed a neural network algorithm for diagnosing prostate cancer, is of interest. Scientists built a mathematical model trained on experimental knowledge based on the Neural Network Toolbox Matlab. Clinical and pathological characteristics of 220 patients before surgery were input variables. The results of treatment according to the stage of the malignant process were also used to train and test the neural network. The trained neural network was tested on 64 patients at the S.P. Botkin Hospital. Scientists confirmed an increase in the reliability of prognosis of prostate cancer by 14%. The authors noted the exact correspondence of real medical parameters and neural network calculations. They created the Urostat software, as well as an algorithm for treating prostate cancer in a medical institution depending on the stage of the oncological process [33].

Recent years in the field of digitalization of medicine can be called explosive in several directions at once. Among them: the incredible growth of big data and the capacity of modern computer systems, the creation of powerful graphics processors and other innovations in the field of high-tech equipment. These trends have made it possible to conduct machine learning at a high level of accuracy and reliability. Along with this, the growth of investments in the digital healthcare system of Russia compared to 2015 by 2023 has grown more than 15 times. At present, there are about 50 different AI systems in the Russian Federation working in the field of medicine and healthcare. Such as Celsus, Botkin.ai, RADLogics, Doctor Tomo, Doctor AIzimov, OneCell, PathVision.ai, Onqueta and others work in the field of oncology. Based on the PathVision.ai platform, a neural network has been created that helps analyze the state of prostate cells on histological preparations. This algorithm can currently be called the "gold standard" for diagnosing prostate cancer. The technology became possible thanks to the colossal array of accumulated big data obtained in the study of histological sections in prostate cancer and used for deep machine learning. PathVision.ai offers its innovative development - convolutional neural networks. They are built on the SkipNet algorithm. This is deep learning with convolutional networks of 34 levels, which has undergone validation, testing and training on data networks, including a multi-million data array formed from 22 thousand categories. The practical significance of the proposed network lies in the early diagnosis and typing of prostate cancer according to Gleason. The neural network also carries out marking on histological preparations, identifying a number of prostate diseases, including prostatic intraepithelial neoplasia, which is considered a precancerous condition [28]. In 2020, the creation of the OneCell digital platform began in the Russian Federation. Not only Russian specialists were invited to develop this project. Scientists from Israel and Norway took part in it. The goal of the project was to create a digital hardware and software complex OneCell, which has three components: equipment, AI tools and a telemedicine platform. Each category was handled by an independent group of specialists. A major achievement of the new digital platform can be considered the creation of its own equipment, including a unique digital camera. Domestic equipment is five times cheaper than imported, but no less effective. The AI ​​of big data of the new software accelerates the work of a histomorphologist by 10 times, who instantly identifies the pathology of the histosection, while making the diagnosis more accurate. The company plans to unite all medical oncology institutions into a single national diagnostic digital hub, from small regional laboratories to large oncology centers that will use AI technologies and accumulated big data information [34]. It is interesting to note the new IT product Onqueta. It was recently tested at the molecular genetic research laboratory of the S. Berezina Medical Institute and is under development. This algorithm has the ability to assess the probability of germline inheritance of cancer. Artificial intelligence will help to establish a high risk of hereditary predisposition to oncology through simple testing, without resorting to medical care. The algorithm gives instructions for undergoing further medical examination. This will greatly simplify the task of early cancer diagnosis. In 2022, 1551 packages from the Hereditary Oncological Syndromes in the Russian Federation database were tested using Onqueta. Two cohorts were identified: patients with pathogenic mutations (731) and with a low probability of germline mutations (820). "The application has a sensitivity of 94% and a specificity of 97.5%" [35]. If in domestic oncourology the use of AI is rather at the development stage, then in the foreign segment scientists show significant results of diagnosis and treatment in this area of ​​​​oncology. Medical research platforms such as PubMed and Medline display numerous scientific publications, which are both reviews on the use of AI in oncourology [36, 37] and original research articles. The most interesting information is from the last five years, since the use of digital technologies has acquired a global scale during this time. Strom et al. proposed a technique based on deep machine learning to accurately identify and type prostate cancer according to the Gleason scale. The model was trained on approximately 7 thousand digital slides made from tumor histosections of about a thousand men and tested on more than 1.5 thousand biopsy samples taken from 246 patients. The scientist demonstrated an accuracy of 0.997 (AUC) for differentiating malignant and benign tumors. The results of the Gleason score correlated with the results presented by histologists [38]. The use of AI in the treatment of prostate cancer is also noted. According to statistics, approximately twenty percent of patients experience severe toxicity during chemotherapy with docetaxel, which is a first-line drug for the treatment of metastatic castrator-resistant prostate cancer (mCRPC). Deng et al. used AI to create an algorithm to identify patients who do not tolerate this drug well. In their studies, the scientists used data taken from 1,600 patients with phase III mCRPC. The artificial intelligence was trained on 78 parameters. These were laboratory parameters, metastasis data, clinical features of the patient, and medical history. The scientists integrated survival status and the severity of adverse events into their model. The proposed method is an innovative way to complement and stratify treatment discontinuation information. Critical stratification biomarkers were additionally identified in determining treatment discontinuation. The proposed model has great potential to improve future personalized treatment in mCRPC [39]. Porpiglia et al. proposed an AI model that can be used in surgical treatment of PCa. It helps in effective intraoperative detection of extracapsular tumor invasion into the neurovascular bundle during radical prostatectomy [40]. Preoperatively, MRI images of the prostate gland are obtained and then used as a three-dimensional (3D) image during robotic prostatectomy. The surgeon sees the area of ​​the prostate gland affected by the tumor in 3D reconstruction during the surgery. This method can also potentially be used as an adjunct in robotic partial nephrectomy, especially for endophytic or dorsally located tumors. This type of intraoperative navigation with visualization can help to avoid positive surgical margins and maximize organ preservation [41]. These examples have shown that the use of big data and various AI tools has great potential in the early diagnosis and treatment of PCa, including both chemotherapy and surgery. Malignant testicular tumors. There are not enough scientific papers describing the use of AI in testicular tumors. Baessler et al. proposed the use of computed tomography based on machine learning to diagnose possible metastatic lymph node involvement in testicular malignancies. Using the proposed technique, lymph nodes were identified as metastatic or unchanged before lymphadenectomy in patients who received chemotherapy for advanced disease. The model gave a correct classification with an accuracy of 0.81 (area under the curve AUC), a sensitivity of 88%, and a specificity of 72% [42]. Bladder cancer (BC). A number of authors have studied the prediction of bladder cancer progression using digital technologies. Studies by James W F Catto and al. presented a comparative analysis of neuro-fuzzy modeling (NFM), artificial neural networks (ANN) and traditional statistical methods in predicting the clinical course of BC. NFM is a type of AI methods that has a transparent functional layer and is devoid of many of the disadvantages of ANN. In a cohort of 109 patients diagnosed with bladder cancer, experimental molecular biomarkers, including p53 and mismatch repair proteins, as well as conventional clinicopathological data, were studied. For the methods used, models were created to predict the presence and timing of tumor recurrence. AI methods predicted recurrence with an accuracy of 88% to 95%. These indicators exceeded the statistical methods of the study (71-77%; P <0.0006). NFM was more accurate than ANN in predicting the timing of recurrence (P = 0.073) [43]. Another study was also indicative: M. F. Abbod et al. compared the prediction of neuro-fuzzy modeling (NFM), artificial neural networks (ANN) and traditional statistical methods for predicting BC. In a cohort of bladder cancer patients (117 people), molecular biomarkers, including p53 expression and gene methylation and conventional clinicopathological parameters, were studied. Prediction models for the presence and timing of tumor progression were created for all three models. AI methods predicted progression with an accuracy of 88-100%. This result was much more accurate than logistic regression. [44]. James W F Catto and colleagues developed a new method for gene expression microarray analysis that combines two forms of artificial intelligence (AI), neural phase modeling and artificial neural networks, and tested it in a cohort of patients with bladder cancer. The scientists used AI and statistical analysis to identify genes that were associated with tumor development. The microarray dataset was taken from 66 tumors, with a total of 2800 genes. The genes selected by the AI ​​were then examined in a second cohort of patients (262) using immunohistochemistry. The AI ​​identified 11 cancer development genes (odds ratio, 0.70; 95% confidence interval, 0.56-0.87; p = 0.0004). In the statistical analysis group, the odds ratio was 1.24; 95%; confidence interval, 0.96-1.60; p = 0.09. The expression of six AI-selected genes (LIG3, FAS, KRT18, ICAM1, DSG2, and BRCA2) was determined using antibodies and successfully identified tumor progression (matching index: 0.66; log-rank test: p=0.01). The AI-selected genes were more accurate in determining aggressive bladder cancer development than pathological criteria in determining progression (Cox multivariate analysis: p=0.01) [45]. In the presented review by Raghav K Pai [37], published in PubMed, interesting material was collected on the latest developments in the use of AI in bladder cancer. Among the authors, it is worth noting a group of scientists led by I. Sokolov, who developed a machine learning-based method that was able to detect bladder cancer using cell images extracted from urine samples. This method showed 94% diagnostic accuracy [46]. Artificial intelligence can also help in choosing the treatment for bladder cancer. Cystectomy is the “gold standard” for the treatment of muscle-invasive bladder cancer, and neoadjuvant chemotherapy before cystectomy has been shown to improve treatment outcomes [47]. Cha KH et al. conducted a study that analyzed different radionics-based prediction models. Deep learning was also used to accurately classify tumors according to their response to chemotherapy. The model used more than 6,000 paired pre- and post-treatment regions of interest obtained from CT scans and compared the results with the results of two expert radiologists. The differences between deep learning algorithms and the work of radiologists in predicting the outcome of chemotherapy did not reach statistical significance. The study showed that AI can be used to predict treatment outcomes, allowing doctors to save resources and avoid chemotherapy in patients with chemoresistant bladder cancer [48]. Ann-Christin Woerl et al. conducted an interesting study on muscle-invasive bladder cancer. It is the second most common malignant neoplasm of the genitourinary system and is associated with high morbidity and mortality. Recently, molecular subtypes of urothelial cancer have been identified that determine the course and prognosis of the disease. The authors attempted to predict the molecular subtype of bladder cancer samples using standard histomorphology based on deep machine learning. The model was trained on data from two cohorts of patients with bladder cancer. 407 cases with urothelial carcinoma of the bladder from the Cancer Genome Atlas and 16 cases with patients with primarily resected muscle-invasive bladder cancer. As a result, 423 digitized images were used to train, test and validate the DL algorithm for molecular subtype prediction. The deep learning model showed higher accuracy and performance than expert histomorphologists. Various methods of visualization of hematoxylin and eosin stained histosections allowed the authors to identify new histopathological features that were most relevant to their model. DL can be used to predict important molecular features in patients with muscle-invasive BC from slide analysis alone, significantly increasing the potential for clinical management of this disease [49].

Kidney cancer. Kidney cancer is the most lethal type of malignant neoplasm in oncourology. In 25% of patients, the development of metastasis is observed. The most common types are clear cell, papillary and chromophobe. Differentiation of kidney cancer by type is of great diagnostic and therapeutic importance, since there are differences in the factors of prognosis and treatment of various subtypes [37]. ML and DL algorithms based on CT texture analysis were used for differentiation by a number of researchers. The authors aimed to predict the nuclear grade and identify certain genetic mutations for diagnosis, relapse prediction and survival [50]. For example, Kocak et al. used a machine learning method based on CT texture analysis. The authors predicted and identified the nuclear grade according to Fuhrman of clear cell RC and compared the indicators with the results of percutaneous biopsy. The trained algorithm showed the maximum prognostic significance; it differentiated nuclear grades in 85.1% of cases of clear cell renal cell carcinoma [51]. Chinese scientists Lin et al. developed a machine learning model capable of predicting the Fuhrman grade in clear cell renal cell carcinoma of varying degrees of malignancy (RCC). They used single- and 3-phase computed tomography with a marked area of ​​interest as the basis for training. The model based on 3-phase CT turned out to be more effective (area under the ROC curve (AUC) = 0.87) and achieved the best diagnostic performance for differentiating low- and high-grade clear cell renal cell carcinoma. When using classifiers based on the specifics of each individual phase: pre-passage - AUC = 0.84, corticomedullary - AUC = 0.80 and nephrographic - AUC = 0.82 [52]. In addition to diagnostics, AI can be used to optimize the treatment strategy for RC. Buchner et al. proposed an ANN-based model trained on multiple parameters, including patient age, body mass index, histological parameters, and treatment. Using these data, the algorithm was able to accurately predict 36-month patient survival according to different types of therapy with an accuracy of 91% in a validation cohort [53]. The results demonstrate the ability of artificial intelligence to develop personalized pharmacological treatment plans for patients with kidney cancer based on multiple prognostic factors and automated analysis of the results. Conclusion. Digitalization of medicine has developed significantly over the past decade. This was facilitated by the accumulation of a huge array of medical information, the emergence of large digital platforms, innovative digital technologies and equipment: graphic processors, microscopes, computers of enormous power, various sensors, coordinated devices intended for medical institutions and all kinds of gadgets for individual use. Artificial intelligence has enormous potential for use in the medical field. Therefore, the attention of specialists is currently focused on the development of various AI models: machine learning, neural networks and deep learning. By imitating human cognitive abilities, they are able to solve complex medical problems, such as early diagnosis, treatment and prevention of cancer. The huge genetic and epigenetic variability of cancer makes it one of the most complex and multifaceted diseases, presenting significant difficulties for study. A systematic approach based on big data, models and artificial intelligence tools create enormous research potential in all areas of oncological practice, including oncourology. The use of AI in prostate, kidney, bladder and testicular cancer represents great potential. AI increases the chances of early diagnosis, further prognosis of the disease and the most effective treatment through the use of trained models. Training is based on NGS sequencing, various visualization methods (digital radiological and histological images), the use of genetic biomarkers of cancer, PSA, Gleason scores and other parameters. AI reduces the subjectivity of the diagnostic process, reduces the time for processing information, helps the doctor in making a decision, promotes personalized treatment and accurate precision medicine. It is important to note that the use of digital innovative technologies does not exclude traditional diagnostics, treatment methods and medical presence. On the contrary, AI facilitates consolidated interaction between doctors and scientists in real time. They can exchange information, conduct a joint visual assessment of the patient and collectively discuss the choice of the most effective treatment.

作者简介

Alexander Khachaturyan

编辑信件的主要联系方式.
Email: centrforward@mail.ru
ORCID iD: 0000-0003-3774-2879

Cand. Sci. (Med.)

参考

补充文件