Kudryavtsev, Nikita D

Petraikin A.V., Pickhardt P.J., Belyaev M.G., Belaya Z.E., Pisov M.E., Bukharaev A.N., Zakharov A.A., Kudryavtsev N.D., Bobrovskaya T.M., Semenov D.S., Akhmad E.S., Artyukova Z.R., Abuladze L.R., Nizovtsova L.A., Blokhin I.A., et. al.

Artyukova Z.R., Petraikin A.V., Kudryavtsev N.D., Petryaykin F.A., Semenov D.S., Sharova D.E., Vladzymyrskyy A.V., Vasilev Y.A., Belaya Z.E.

BACKGROUND: Osteoporosis is often diagnosed at the stage with complications, i.e., low-energy fractures. Vertebral compression fractures, which are complications of osteoporosis and predictors of subsequent fractures, are often asymptomatic. Compression fractures can be found by computed tomography performed for other indications with vertebral morphometry. Approaches to using artificial intelligence algorithms designed for diagnosing vertebral compression fractures were analyzed. AIM: Testing artificial intelligence algorithms to conduct morphometric analysis of vertebrae on chest computed tomography scans and assess the possibility of their implementation in medical organizations of the Moscow Healthcare Department. MATERIALS AND METHODS: To set a clinical task for artificial intelligence algorithms, basic diagnostic requirements in the area of “vertebral compression fractures (osteoporosis)” were formulated. The testing of the artificial intelligence algorithms included the following stages: self-testing, functional and calibration testing, practical evaluation, and operation testing. The first three stages of testing were performed using previously generated datasets. At practical evaluation and operation testing, artificial intelligence algorithms analyzed the data from computed tomography performed in medical organizations. The expert group of radiologists assessed the diagnostic accuracy and functional capacity of the AI algorithms at all stages. The resulting quantitative metrics of the accuracy of artificial intelligence algorithms were compared with the required target values. RESULTS: From June 2021 to June 2022, two artificial intelligence algorithms (Nos. 1 and 2) with different methods of detecting compression fractures were tested. Both artificial intelligence algorithms successfully passed the self-testing (6 tests), functional (5 tests), and calibration (100 tests) stages. The area under the ROC curve for artificial intelligence algorithm No. 1 was 0.99 (95% CI, 0.98–1), and for artificial intelligence algorithm No. 2, it was 0.91 (95% CI, 0.85–0.96). Artificial intelligence algorithm No. 1 passed the practical evaluation stage without any significant remarks, whereas algorithm No. 2 was sent for fine-tuning. After the operation testing stage, the following accuracy metrics were obtained: the areas under the ROC curve for artificial intelligence algorithm Nos. 1 and 2 were 0.93 (95% CI, 0.89–0.96) and 0.92 (95% CI, 0.90–0.94), respectively. At all stages, both artificial intelligence algorithms demonstrated sufficient metrics for clinical validation. CONCLUSION: Artificial intelligence algorithms for the automatic diagnosis of vertebral compression fractures have been tested, demonstrating the high quality of their operation. artificial intelligence algorithms can be applied as a supplementary tool in the medical decision support system.

Solovev A.V., Sinitsyn V.E., Sokolova M.V., Kudryavtsev N.D., Vladzymyrskyy A.A., Semenov D.S.

BACKGROUND: The state of health of the pulmonary system and its impact on the overall well-being of the individual is an important aspect of modern medicine. Despite continuous progress in diagnostics and technology, epidemiologic data on pulmonary trunk health at the population level in Russia remain understudied. In the context of this problem, the present study is an in-depth population-based analysis of the status of pulmonary trunk dilatation using modern technology and artificial intelligence [1]. Pulmonary trunk dilatation (≥29 mm) may be associated with various pathologies including arterial hypertension, chronic obstructive pulmonary disease, heart failure, and other diseases of the circulatory system [2]. AIM: The aim of the study was to assess the prevalence of pulmonary trunk dilatation in the Moscow population using artificial intelligence technologies. MATERIALS AND METHODS: The study was conducted between September 2022 and February 2023 in the population of Moscow. A large amount of chest CT data was analyzed, including information on 134,218 patients (61,514 men and 72,704 women). Artificial intelligence technologies were used to automatically process this data. RESULTS: The results show that 49,227 (36.7%) patients — 23,720 (38.6%) men and 25,507 (35.1%) women — had evidence of pulmonary trunk dilatation. The analysis shows gender and age differences in the incidence of the pathology. The distribution of pulmonary trunk dilatation in the population shows age dependence. The percentage of patients with signs of pulmonary trunk dilatation increases with age: from 18.1% in the group of young people to 62.2% in the group of elderly people. CONCLUSIONS: The study provides the first epidemiological data on pulmonary trunk dilatation in Moscow and emphasizes the importance of further research in this area. The findings may serve as a basis for the development of effective diagnostic and treatment strategies, as well as for further research in the field of artificial intelligence in medicine.

Kudryavtsev N.D., Kozhikhina D.D., Goncharova I.V., Shulkin I.M., Sharova D.E., Arzamasov K.M., Vladzymirskyy A.V.

Kudryavtsev N.D., Petraikin A.V., Ahkmad E.S., Kiselev F.A., Burashov V.V., Mukhortova A.N., Soldatov I.V., Shkoda A.S.

The global outbreak of COVID-19 has posed unprecedented challenges to healthcare systems worldwide. Healthcare administrators had to make quick and effective decisions to ensure high quality of medical care standards in new conditions. The need to form a reserve bed fund during the pandemic was due to the high load on city hospitals in Moscow. Due to this fact, temporary reserved hospitals for COVID-19 patients were organized in non-core facilities, such as ice arenas, shopping malls, and exhibition pavilions. This urgency prompted a search for solutions that could provide the necessary level of diagnosis and treatment appropriate to specialized medical facility. Given the technical and time constraints associated with the installation of a fixed computer tomographic scanner, the deployment of mobile computer tomographic scanners emerged as a viable option. The study aims to share insights gained from using a mobile computer tomographic scanner within a temporary backup hospital setting to treating patients with COVID-19 coronavirus infection. The paper discusses the features, advantages, and disadvantages of mobile computer tomography. It also presents hardware and control room layouts, along with the placement options for the computer tomography device. The research includes the results of dosimetry studies and provides a clinical assessment of the applicability of this type of diagnostic devices.

Kudryavtsev N.D., Vladzymyrskyy A.V.

Kudryavtsev N.D., Bardasova K.A., Khoruzhaya A.N.

Speech recognition technology is a promising tool for healthcare systems. This technology has a fairly long history of use in US and European healthcare systems, beginning in the 1970s. However, it only became widespread at the beginning of the 21st century, replacing medical transcriptionists. For Russian healthcare system, speech recognition technology is a new tool. Its active development began only in the early 2010s, and its active implementation in healthcare began in the late 2010s. Such a delay is due to the peculiarities of the Russian language and the limitation of computing power present at the beginning of the XXI century. Currently, speech recognition technology is used to fill out medical reports by voice, it allows you to reduce the preparation time of the protocol of radiological studies in comparison with the traditional (keyboard) text input. The review described a brief history of the development and application of speech recognition technology in radiology. Key scientific studies confirming the effectiveness of its use in US and European healthcare systems are described. Our experience in the use of speech recognition technology is demonstrated and its effectiveness is evaluated. The prospects for further development and application of this technology in Russian health care are described.

Kudryavtsev N.D., Sharova D.E., Vladzymyrskyy A.V.

BACKGROUND: Speech recognition is becoming increasingly common in the national healthcare system. One of the first specialties to implement this technology on a large scale was radiology. However, the efficiency of voice input and its effect on the length of time required to complete medical records remain unresolved. AIM: To assess the efficiency of speech recognition in generating radiological protocols of different modalities and types. METHODS: The retrospective study was conducted at the Moscow Reference Center of the Research and Practical Clinical Center for Diagnostics and Telemedicine Technologies of the Moscow Department of Health. A total of 12,912 radiological reports on fluorography, mammography, chest computed tomography (CT), contrast-enhanced magnetic resonance imaging (MRI) of the brain, and contrast-enhanced CT of the abdomen and pelvis were included in the study by simple random sampling. The size of all samples exceeded 766 reports, calculated with regard to the size of the general population of over 100,000 reports. The Voice2Med software was used to fill in the radiological protocols. Intergroup comparison was performed using the MannWhitney U-test with a statistical significance level of 0.05. RESULTS: The average duration of generating fluorographic protocols in the keyboard and voice input groups was 189.9 s (0:03:09) and 236.2 s (0:03:56), respectively (p 0.0001). For mammographic reports, the duration was 387.1 s (0:06:27) and 444.8 s (0:07:24), respectively (p 0.0001). For radiographic reports, it amounted to 247.8 s (0:04:07) and 189.0 s (0: 03:09), respectively (p 0.0001), and for chest CT, it was 379.7 s (0:06:19) and 382.7 s (0:06:22), respectively (p=0.12). For MRI of the brain, the protocols were generated for 709.9 s (0:11:49) and 559.9 s (0: 09:19), respectively (p 0.0001), and for contrast-enhanced chest, abdominal, and pelvic CT scans, it took 2714.6 s (0:45:15) and 1778.4 s (0:29:38), respectively. Voice input slowed down the preparation time of mammographic and fluorographic protocols. This is due to the use of a structured electronic medical document in medical facilities to describe the results of the examinations. Speech recognition showed the greatest efficiency in generating MRI and CT protocols. Such reports contain a large number of pathological changes, both target and incidental findings, which requires a detailed description by the radiologist in the examination protocol. CONCLUSIONS: Speech recognition in generating radiological protocols showed different efficiency depending on the modality and type of the radiological protocol filled in using the voice input system. This approach is optimal for describing CT and MRI scans.

Artyukova Z.R., Kudryavtsev N.D., Ikryannikov E.O., Titova A.V., Balashov M.K., Petraikin A.V.

BACKGROUND: The FRAX tool (a 10-year fracture risk assessment) is recommended to diagnose osteoporosis and optimize the number of patients who need to undergo X-ray densitometry. Due to various circumstances, the integration of a full-fledged FRAX tool into the digital circuits of the Moscow City Health Department is problematic. AIM: The study aimed to develop a calculator of the 10-year probability of osteoporotic fractures to optimize the routing of patients for examination. METHODS: An optimized Half-FRAX calculator was created based on the FRAX tool from the University of Sheffield, which was developed using the results of population studies of the Russian Federation. All data used in the original FRAX algorithm, i.e. sex, age, height, weight, and T-criterion (if available) and other important parameters such as a history of fractures, parental hip fractures, smoking, rheumatoid arthritis, secondary osteoporosis, and glucocorticoid and alcohol intake were included in the risk assessment calculator. An algorithm for interaction with the FRAX website was developed and implemented to verify critical levels of patient stratification by multiple consecutive enumerations of different combinations of body mass index (BMI) measurements (0.1 discretization) and age (1-year discretization). Data from clinical guidelines were taken as thresholds. RESULTS: When implementing the developed algorithm by modeling various combinations of BMI, T-criterion, and risk factors (RF), the absence of RFs and BMI 25 (upper limit of normal) in women was shown to guarantee the exclusion from the orange zone where densitometry should be performed. In men, BMI was not a RF. If a RF was present, a patient was recommended to consult a doctor. If no T-criterion was present, but a RF was detected, the patient was indicated for densitometry. Similar results were reported for women with the same indices. In the absence of the RF and with a T-criterion 2.5, low fracture risk factor was indicated for both men and women. CONCLUSIONS: An optimized Half-FRAX calculator for the 10-year probability of major osteoporotic fractures was developed, which may optimize the routing of patients for densitometry and reduce the burden on radiology departments in Moscow. This will allow patients to be timely referred to the clinical specialists for consultations. Half-FRAX is integrated into the Osteoporosis Digital Platform (https://telemedai.ru/cifrovaya-platforma-osteoporoz/half-frax).

Petraikin A.V., Artyukova Z.R., Kudryavtsev N.D., Semenov D.S., Smorchkova A.K., Repin S.S., Akhmad E.S., Petriaikin F.A., Nisovtsova L.A., Vladzimirskyy A.V.

Objective: to conduct the study of age distribution of bone mineral density (BMD) by the database of dualenergy X-ray absorptiometry (DXA) and to compare it with datа of population NHANES study. Material and methods. We used data from the densitometry of three-zone (total hip (TH), femoral neck (FN), and lumbar spine) measured by DXA from two outpatient clinics. The obtained data were compared with NHANES III for TH and FN and with NHANES 2005-08 for lumbar spine. The BMD value was corrected with the calibration coefficient for each DXA scanner. Adjustments were also made for the population distribution by sex and age. Results. Compared with NHANES for FN and TH, the obtained BMD values were significantly decreased for patients aged less than 50 years (men and women). The BMD values for FN and TH were unsignificantly decreased in men older 50 years. In women older 50 years unsignificantly decreased BMD values for FN and a significantly increased BMD values for TH were observed. The BMD values were decreased for lumbar spine in men and women throughout this age interval (more than 50 years old). Conclusion. The population BMD distribution in men and women was assessed by DXA method. The obtained dependence of the BMD for FN in women older 50 years was in good agreement with the results given by Russian and foreign authors.

Artyukova Z.R., Kudryavtsev N.D., Petraikin A.V., Abuladze L.R., Smorchkova A.K., Akhmad E.S., Semenov D.S., Belyaev M.G., Belaya Z.E., Vladzimirskyy A.V., Vasiliev Y.A.

Goal: To develop a method for automated assessment of the volumetric bone mineral density (BMD) of the vertebral bodies using an artificial intelligence (AI) algorithm and a phantom modeling method.Materials and Methods: Evaluation of the effectiveness of the AI algorithm designed to assess BMD of the vertebral bodies based on chest CT data. The test data set contains 100 patients aged over 50 y.o.; the ratio between the subjects with/without compression fractures (Сfr) is 48/52. The X-ray density (XRD) of vertebral bodies at T11-L3 was measured by experts and the AI algorithm for 83 patients (205 vertebrae). We used a recently developed QCT PK (Quantitative Computed Tomography Phantom Kalium) method to convert XRD into BMD followed by building calibration lines for seven 64-slice CT scanners. Images were taken from 1853 patients and then processed by the AI algorithm after the calibration. The male to female ratio was 718/1135.Results: The experts and the AI algorithm reached a strong agreement when comparing the measurements of the XRD. The coefficient of determination was R2=0.945 for individual vertebrae (T11-L3) and 0.943 for patients (p=0.000). Once the subjects from the test sample had been separated into groups with/without Сfr, the XRD data yielded similar ROC AUC values for both the experts – 0.880, and the AI algorithm – 0.875. When calibrating CT scanners using a phantom containing BMD samples made of potassium hydrogen phosphate, the following averaged dependence formula BMD =0.77*HU-1.343 was obtained. Taking into account the American College Radiology criteria for osteoporosis, the cut-off value of BMD<80 mg/ml was 105.6HU; for osteopenia BMD<120 mg/ml was 157.6HU. During the opportunistic assessment of BMD in patients aged above 50 years using the AI algorithm, osteoporosis was detected in 31.72% of female and 18.66% of male subjects.Conclusions: This paper demonstrates good comparability for the measurements of the vertebral bodies’ XRD performed by the AI morphometric algorithm and the experts. We presented a method and demonstrated great effectiveness of opportunistic assessment of vertebral bodies’ BMD based on computed tomography data using the AI algorithm and the phantom modeling.

Kudryavtsev N.D., Semenov D.S., Kozhikhina D.D., Vladzymyrskyy A.V.

Vladzymyrskyy A.V., Kudryavtsev N.D., Kozhikhina D.D., Shulkin I.M., Morozov S.P., Ledikhova N.V., Klyashtornyy V.G., Goncharova I.V., Novikov A.V., Vnukova O.M.

Методы лучевой диагностики все более масштабно применяются при массовых профилактических осмотрах (скринингах) для выявления различных патологических состояний. С целью повышения эффективности программ скрининга в ряде ведущих стран мира предусмотрены двойные описания результатов исследований, что неоспоримо увеличивает рабочую нагрузку на врачей-рентгенологов. В связи с эти крайне актуальной становится задача автоматизированного анализа результатов скрининговых исследований. ЦЕЛЬ ИССЛЕДОВАНИЯ Оценить влияние делегирования полномочий по выполнению первого описания медицинскому программному обеспечению на основе технологий искусственного интеллекта (ИИ) на длительность процесса двойного описания результатов флюорографии. МАТЕРИАЛ И МЕТОДЫ Исследование выполнено на базе Московского референс-центра лучевой диагностики (ГБУЗ «НПКЦ ДиТ ДЗМ»). В исследование включено 13 901 флюорографическое исследование. Реализованы два сценария двойного описания исследований: в первом случае участвовали врач-рентгенолог и алгоритм ИИ, во втором — два врача-рентгенолога. Просмотр результатов ИИ и описание исследований проводилось в Едином радиологическом информационном сервисе Единой медицинской информационно-аналитической системы города Москвы. Выполнен статистический анализ данных. РЕЗУЛЬТАТЫ По сценарию №1 проведено двойное описание 1435 результатов флюорографии, по сценарию №2 — 12 446. В первом сценарии врач, получив данные машинного анализа, затрачивал на подготовку заключения в среднем 0,9±3,0 мин. Во втором сценарии продолжительность работы врача, осуществлявшего первый просмотр, составила 0,8±2,1 мин; второй просмотр — 0,4±1,0 мин. Общая длительность проведения двойного описания в формате «врач + ИИ» колебалась от 0,7 до 1241,2 мин, составив в среднем 199,3±330,3 мин. Во втором сценарии общая длительность проведения двойного описания составила 1 838,5±3671,4 мин. ЗАКЛЮЧЕНИЕ Делегирование первого описания алгоритму искусственного интеллекта принципиально ускоряет предоставление результатов флюорографии, повышая их доступность для обследованных лиц и медицинских работников, направляющих пациентов на обследование. Актуальнейшим вопросом становится точность работы соответствующих технологий искусственного интеллекта, а обязательность их регистрации в качестве медицинского изделия не подлежит дальнейшему обсуждению.

Petraikin A.V., Akhmad E.S., Semenov D.S., Artyukova Z.R., Kudryavtsev N.D., Petriaikin F.A., Nizovtsova L.A.

Background. Dual-energy X-ray absorptiometry (DXA) is an effective method for bone mineral density (BMD) and subcutaneous fat percentage estimation. The constant development of new densitometry techniques, the demographic change and the higher potential of artificial intelligence in healthcare enhance requirements for the high-quality measurements in DXA. This study aimed to develop a quality control method for DXA scanners and compare four DXA systems with different X-ray geometries and manufacturers when simulating fat-water environments. Methods. We evaluated the accuracy (relative error (%) and precision (CV%)) of the bone mineral density (BMD) measurements, performed by the four DXA scanners: 2 with narrow-angle fan beam (64- and 16-channel detectors (DXA-1, DXA-2)); 1 with wide-angle fan beam (DXA-3); 1 with pencil beam (DXA-4). We used a PHK (PHantom Kalium) designed to imitate spine. The PHK contained four vertebras filled with a K2HPO4 solution in various concentrations (50-200 mg/ml). The PHK also included paraffin patches (thickness 40 mm) to simulate the fat layer. Results. DXA-1 and DXA-2 demonstrated the best CV% ranged from 0.56% to 1.05%. The least % was observed when scanning PHK with fat layer on DXA-1 and DXA-2 (1.74% and 0.85%) and DXA-4 (1.47%). DXA-3 produced significantly lower BMD ( = -14.56%, p = 0.000). After removing the fat layer, we observed reduction (p = 0.000) of BMD for DXA- 1 and DXA-2 ( = -5.11% and -6.12% respectively) and weak deviation (p = 0.80) for DXA-4 (0.87%). For DXA-3, removal of the fat layer also resulted in a significant reduction in BMD ( = -16.44%, p = 0.000). The subcutaneous fat modeling showed that all these DXA systems automatically determine the percentage of fat in the scanned area with weak underestimation: for DXA-1, DXA-2 and DXA-4 the % were -5,9%, -6,3% and -2,3% respectively. CV% were 0.15%; 0.39%; 1.6%, respectively. Conclusions. We proved a significant underestimation of the BMD measurements across the entire range of simulated parameters for the DXA scanners when the model did not include the subcutaneous fat layer. All models demonstrated high accuracy in measuring the fat layer, with the exception of the DXA-3 model, which was not assessed in these studies.

Andrianova M.G., Kudryavtsev N.D., Petraikin A.V.

.

Solovev A.V., Vasilev Y.A., Sinitsyn V.E., Vladzymyrskyy A.V., Ivanova G.V.

INTRODUCTION: In modern clinical practice, the diagnosis establishment and monitoring of pulmonary arteries are important components of care for patients with cardiovascular and respiratory diseases. However, in Russia, there remains a limited amount of epidemiological data on the prevalence of pulmonary artery dilation, especially at the population level.This study aims to fill this gap by providing population-based data on the prevalence of pulmonary artery dilation in the city of Moscow. With the help of advanced artificial intelligence technologies, this work seeks to offer a comprehensive assessment of the pathology, covering both its main characteristics and the effects of various factors such as age and gender.OBJECTIVE: To investigate the prevalence of pulmonary artery dilation in Moscow using the data from computed tomography (CT) scans of the chest analysed by means of an automatic artificial intelligence technology.MATERIALS AND METHODS: The study involved the analysis of a large volume of chest CT scans acquired from September 2022 to February 2023. The total study sample comprised 134,218 patients. Artificial intelligence technologies were applied for the automatic detection of signs of pulmonary artery dilation.RESULTS: An analysis of 125,878 CT scans, including 57,913 men and 67,965 women, revealed that 34.4% of patients (43,242 individuals) showed signs of pulmonary artery dilation. Among them, the proportion of men with this pathology was 35.6% (20,630 individuals), while for women it was 33.3% (22,612 individuals). The prevalence of the pathology increased with age, starting at 14.8% among younger patients and reaching 62.7% among the elderly. The prevalence of pulmonary artery dilation among the population of Moscow was 794.7 cases per 100,000 people.DISCUSSION: The results of this research allowed us to draw conclusions regarding the meaning of pulmonary artery dilation as a predictor of pulmonary hypertension. The results demonstrated a correlation between the frequency of pathology occurrence and gender and age groups, with a more pronounced association observed in women. Detailed analysis also revealed corellations of pulmonary artery diameter with age and gender.CONCLUSION: The study provided the first population-based data on the prevalence of pulmonary artery dilation in Moscow. The importance of early diagnosis to prevent severe complications, especially in patients with chronic lung diseases, is emphasized. The study results can provide a basis for screening strategies and treatment approaches for this pathology.

Artyukova Z.R., Petraikin A.V., Kudryavtsev N.D., Petryaykin F.A., Semenov D.S., Sharova D.E., Vladzymyrskyy A.V., Vasilev Y.A., Belaya Z.E.

BACKGROUND: Osteoporosis is often diagnosed at the stage with complications, i.e., low-energy fractures. Vertebral compression fractures, which are complications of osteoporosis and predictors of subsequent fractures, are often asymptomatic. Compression fractures can be found by computed tomography performed for other indications with vertebral morphometry. Approaches to using artificial intelligence algorithms designed for diagnosing vertebral compression fractures were analyzed. AIM: Testing artificial intelligence algorithms to conduct morphometric analysis of vertebrae on chest computed tomography scans and assess the possibility of their implementation in medical organizations of the Moscow Healthcare Department. MATERIALS AND METHODS: To set a clinical task for artificial intelligence algorithms, basic diagnostic requirements in the area of “vertebral compression fractures (osteoporosis)” were formulated. The testing of the artificial intelligence algorithms included the following stages: self-testing, functional and calibration testing, practical evaluation, and operation testing. The first three stages of testing were performed using previously generated datasets. At practical evaluation and operation testing, artificial intelligence algorithms analyzed the data from computed tomography performed in medical organizations. The expert group of radiologists assessed the diagnostic accuracy and functional capacity of the AI algorithms at all stages. The resulting quantitative metrics of the accuracy of artificial intelligence algorithms were compared with the required target values. RESULTS: From June 2021 to June 2022, two artificial intelligence algorithms (Nos. 1 and 2) with different methods of detecting compression fractures were tested. Both artificial intelligence algorithms successfully passed the self-testing (6 tests), functional (5 tests), and calibration (100 tests) stages. The area under the ROC curve for artificial intelligence algorithm No. 1 was 0.99 (95% CI, 0.98–1), and for artificial intelligence algorithm No. 2, it was 0.91 (95% CI, 0.85–0.96). Artificial intelligence algorithm No. 1 passed the practical evaluation stage without any significant remarks, whereas algorithm No. 2 was sent for fine-tuning. After the operation testing stage, the following accuracy metrics were obtained: the areas under the ROC curve for artificial intelligence algorithm Nos. 1 and 2 were 0.93 (95% CI, 0.89–0.96) and 0.92 (95% CI, 0.90–0.94), respectively. At all stages, both artificial intelligence algorithms demonstrated sufficient metrics for clinical validation. CONCLUSION: Artificial intelligence algorithms for the automatic diagnosis of vertebral compression fractures have been tested, demonstrating the high quality of their operation. artificial intelligence algorithms can be applied as a supplementary tool in the medical decision support system.

Akhmedzyanova M.R., Zelikman M.I., Kruchinin S.A., Lantukh Z.A., Soldatov I.V., Vasilev Y.A., Druzhinina Y.V., Tolokonskiy A.O.

Vasilev Y.A., Arzamasov K.M., Vladzymyrskyy A.V., Kolsanov A.V., Shulkin I.M., Bobrovskaya T.M., Pestrenin L.D.

The purpose of research. Radiation diagnostics is central to the detection of malignant neoplasms. Recently, the implementation of screening programs has faced a number of obstacles, including staff shortages and limited funding. The introduction of artificial intelligence (AI)-based systems capable of absolutely accurate sorting of research into two categories - "normal" and "not normal", seems to be a promising solution to these problems. However, before they are widely used, it is critically important to verify their ability to guarantee the safety and high quality of the screening process. The aim of the study is to evaluate the possibility of using autonomous sorting of mammographic examination results in real clinical conditions.  Methods. The study was carried out in 2 stages. At the first stage, 25,892 mammographic studies processed by the AI service were retrospectively analyzed. A ROC analysis of these results was carried out in order to assess the possibility of configuring the AI service for 100% sensitivity. At the prospective stage, the results of 82,372 mammograms were analyzed. All studies were processed by AI services configured for 100% sensitivity. The tasks of the AI services included the sorting of mammography results into the categories "normal" and "not normal". Next, the decisions of AI services and radiologists on categorization were compared. Results. According to the results of a retrospective study, when configuring the AI service for 100% sensitivity, the specificity was 39%. In the course of a prospective study, the proportion of defects (false attribution of research results to the "norm" category) was 0.08%, the specific weight of clinically significant defects in AI services was 0.02%, which is significantly lower than that of a radiologist. Conclusion. The use of autonomous sorting of mammographic research results in clinical practice is possible in order to optimize the diagnostic process during preventive measures, as well as under the condition of monitoring the quality of artificial intelligence technologies. Keywords: artificial intelligence, mammography, preventive examinations, radiation diagnostics. Conflict of interest: The author declares the absence of obvious and potential conflicts of interest related to the publication of this article. 

Gusev A.V., Artemova O.R., Vasiliev Y.A., Vladzymyrskyy A.V.

Introduction. Healthcare is one of the priority sectors for the deployment of artificial intelligence (AI) technologies worldwide, including Russia. A key area of AI deployment is the integration of AI-base software as a medical device (AI SaMD) into the Unified digital systems of the healthcare sector of the Russian Federation.Aim. Research of the results of the deployment of AI SaMD in healthcare of the Russian Federation in 2023.Materials and methods. The State Register of Medical Devices and Organizations (individual entrepreneurs) engaged in the production and manufacture of medical devices was used as information about AI SaMD registered in Russia. As information on the deployment of AI SaMD, data from monitoring to the federal project “Creating a single digital system in healthcare” was used, including reports from constituent entities of the Russian Federation upon these activities. The results of the implementation of AI SaMD in Moscow were obtained according to data from the Moscow Department of Health as part of an experiment on the use of innovative technologies in the field of computer vision for the analysis of medical images.Results. As of January 1, 2024, Roszdravnadzor registered 26 AI SaMD, 77 % of them were developed by 13 Russian companies. At the end of 2023, 84 (94 %) constituent entities of the Russian Federation met the minimum established target for the purchase of AI SaMD. Within the framework of public procurement procedures provided by law, 106 government contracts were signed for the purchase and deployment of AI SaMD for a total amount of 448 million 430 thousand rubles.Conclusion. In 2023, the Russian healthcare system made a significant breakthrough in terms of the practical deployment of AI SaMD. Completed procurement and deployment projects are the basis for subsequent industry development.

Vasilev Y.A., Petraikin A.V., Semenov D.S., Uchevatkin A.A., Abuladze L.R., Bazhin A.V., Sharova D.E.

Background. Magnetic resonance imaging (MRI) is one of the leading modalities for imaging of the musculoskeletal system. There are major problems in MRI of the hand, including lack of specialized coils and reliable fixation device for the hand, uncomfortable posture of the patient, motion artifacts, and small anatomical structures. These factors inevitably lead to incorrect interpretation of the study. Aim. To improve the quality of MRI of the hand by developing an approach to the study: selection of the coil and scanning protocol, as well as positioning and fixation of the patients hand. Materials and Methods. The positioning device was developed to prevent hand movements. Studies were performed using two types of coils. A comparative evaluation of the images was performed, as well as evaluation by the musculoskeletal radiologist. Results: А head coil is more appropriate when it is necessary to scan the entire hand, for example, in rheumatic diseases. A knee coil is more appropriate when it is necessary to study the small anatomical structures (including the wrist) due to a smaller field of view and higher resolution. Based on the obtained data, we formulated guidelines for MRI of the hand: selection of scanning parameters, sequences, and coils. We also suggested a specialized device for fixation of the patient's hand to prevent motion artifacts. Conclusion: There are a number of factors that directly affect qualitative MRI study of the hand: MRI safety, scanning parameters, fixation of the hand. The guidelines offered in this manuscript, as well as the use of the developed specialized fixation device can improve the quality of MRI of the hand.

Pezzutti D.L., Makary M.S.

Vasilev Y.A., Tyrov I.A., Vladzymyrskyy A.V., Shulkin I.M., Arzamasov K.M.

Vinokurova O.O., Vinokurov A.S., Petryaykin A.V., Zimina V.N., Yudin A.L.

The review presents modern ideas about X-ray examination for diagnosis of lung diseases including tuberculosis in the pregnant. The use of X-ray diagnostic tools in the pregnant is limited due to the lack of information about modern capabilities of equipment and special protection, and often by X-ray phobia among patients and physicians. The article presents data on the physical parameters of modern X-ray methods (digital radiography, low-dose CT) and highlights methods free of ionizing radiation, which are gradually entering phthisiologic practice.

Kanwal K., Asif M., Khalid S.G., Liu H., Qurashi A.G., Abdullah S.

Community-acquired pneumonia is one of the most lethal infectious diseases, especially for infants and the elderly. Given the variety of causative agents, the accurate early detection of pneumonia is an active research area. To the best of our knowledge, scoping reviews on diagnostic techniques for pneumonia are lacking. In this scoping review, three major electronic databases were searched and the resulting research was screened. We categorized these diagnostic techniques into four classes (i.e., lab-based methods, imaging-based techniques, acoustic-based techniques, and physiological-measurement-based techniques) and summarized their recent applications. Major research has been skewed towards imaging-based techniques, especially after COVID-19. Currently, chest X-rays and blood tests are the most common tools in the clinical setting to establish a diagnosis; however, there is a need to look for safe, non-invasive, and more rapid techniques for diagnosis. Recently, some non-invasive techniques based on wearable sensors achieved reasonable diagnostic accuracy that could open a new chapter for future applications. Consequently, further research and technology development are still needed for pneumonia diagnosis using non-invasive physiological parameters to attain a better point of care for pneumonia patients.

Zhou Q., Zhang Z., Xia Y., Li J., Liu S., Fan L.

Lung injury is caused by various physical and chemical damages that lead to the destruction of the structural integrity or dysfunction of the lungs, resulting in a series of symptoms or diseases. Given its complex clinical manifestations and varying degrees of severity, accurate diagnosis and evaluation of lung injury are particularly crucial for the selection of appropriate treatment programs. Imaging examination, including chest radiography, computed tomography (CT), magnetic resonance imaging (MRI), positron emission tomography (PET), and so on, are vital tools for diagnosing lung injury and assessing its functional or morphological changes and progression. With the advancement of medical imaging technology, multimodal imaging offers a more diverse assessment of lung injury. This article reviews the progress of multimodal imaging in lung injury, encompassing routine and new imaging techniques, multimodal fusion imaging, and the application of AI, all of which can provide valuable references for clinical diagnosis and management of lung injury.

D’Angelo T., Lanzafame L.R., Carerj M.L., Micari A., Mazziotti S., Booz C.

Kudryavtsev N.D., Kozhikhina D.D., Goncharova I.V., Shulkin I.M., Sharova D.E., Arzamasov K.M., Vladzymirskyy A.V.

Leung M.S., Anam Z., Abraham K., Sean Sarma V., Hamadah Al-Assam H.J.

Solovev A.V., Vasilev Y.A., Sinitsyn V.E., Petraikin A.V., Vladzymyrskyy A.V., Shulkin I.M., Sharova D.E., Semenov D.S.

BACKGROUND: Aortic aneurysms are known as “silent killers” because this potentially fatal condition can be asymptomatic. The annual incidence of thoracic aortic aneurysms and ruptures is approximately 10 and 1.6 per 100,000 individuals, respectively. The mortality rate for ruptured aneurysms ranges from 94% to 100%. Early diagnosis and treatment can be life-saving. Artificial intelligence technologies can significantly improve diagnostic accuracy and save the lives of patients with thoracic aortic aneurysms. AIM: This study aimed to assess the efficacy of artificial intelligence technologies for detecting thoracic aortic aneurysms on chest computed tomography scans, as well as the possibility of using artificial intelligence as a clinical decision support system for radiologists during the primary interpretation of radiological images. MATERIALS AND METHODS: The results of using artificial intelligence technologies for detecting thoracic aortic aneurysms on non-contrast chest computed tomography scans were evaluated. A sample of 84,405 patients 18 years old was generated, with 86 cases of suspected thoracic aortic aneurysms based on artificial intelligence data selected and retrospectively assessed by radiologists and vascular surgeons. To assess the age distribution of the aortic diameter, an additional sample of 968 cases was randomly selected from the total number. RESULTS: In 44 cases, aneurysms were initially identified by radiologists, whereas in 31 cases, aneurysms were not detected initially; however, artificial intelligence aided in their detection. Six studies were excluded, and five studies had false-positive results. Artificial intelligence aids in detecting and highlighting aortic pathological changes in medical images, increasing the detection rate of thoracic aortic aneurysms by 41% when interpreting chest computed tomography scans. The use of artificial intelligence technologies for primary interpretations of radiological studies and retrospective assessments is advisable to prevent underdiagnosis of clinically significant pathologies and improve the detection rate of pathological aortic enlargement. In the additional sample, the incidence of thoracic aortic dilation and thoracic aortic aneurysms in adults was 14.5% and 1.2%, respectively. The findings also revealed an age-dependent diameter of the thoracic aorta in both men and women. CONCLUSION: The use of artificial intelligence technologies in the primary interpretation of chest computed tomography scans can improve the detection rate of clinically significant pathologies such as thoracic aortic aneurysms. Expanding retrospective screening based on chest computed tomography scans using artificial intelligence can improve the diagnosis of concomitant pathologies and prevent negative consequences.

Vasilev Y., Tyrov I., Vladzymyrskyy A., Arzamasov K., Shulkin I., Kozhikhina D., Pestrenin L.

BACKGROUND: In recent years, the availability of medical datasets and technologies for software development based on artificial intelligence technology has resulted in a growth in the number of solutions for medical diagnostics, particularly mammography. Registered as a medical device, this program can interpret digital mammography, significantly saving time, material, and human resources in healthcare while ensuring the quality of mammary gland preventive studies. AIM: This study aims to justify the possibility and effectiveness of artificial intelligence-based software for the first interpretation of digital mammograms while maintaining the practice of a radiologists second description of X-ray images. MATERIALS AND METHODS: A dataset of 100 digital mammography studies (50 absence of target pathology and 50 ― presence of target pathology, with signs of malignant neoplasms) was processed by software based on artificial intelligence technology that was registered as a medical device in the Russian Federation. Receiver operating characteristic analysis was performed. Limitations of the study include the values of diagnostic accuracy metrics obtained for software based on artificial intelligence technology versions, relevant at the end of 2022. RESULTS: When set to 80.0% sensitivity, artificial intelligence specificity was 90.0% (95% CI, 81.798.3), and accuracy was 85.0% (95% CI, 78.092.0). When set to 100% specificity, artificial intelligence demonstrated 56.0% sensitivity (95% CI, 42.269.8) and 78.0% accuracy (95% CI, 69.986.1). When the sensitivity was set to 100%, the artificial intelligence specificity was 54.0% (95% CI, 40.267.8), and the accuracy was 77.0% (95% CI, 68.885.2). Two approaches have been proposed, providing an autonomous first interpretation of digital mammography using artificial intelligence. The first approach is to evaluate the X-ray image using artificial intelligence with a higher sensitivity than that of the double-reading mammogram by radiologists, with a comparable level of specificity. The second approach implies that artificial intelligence-based software will determine the mammogram category (absence of target pathology or presence of target pathology), indicating the degree of confidence in the obtained result, depending on the corridor into which the predicted value falls. CONCLUSIONS: Both proposed approaches for using artificial intelligence-based software for the autonomous first interpretation of digital mammograms can provide diagnostic quality comparable to, if not superior to, double-image reading by radiologists. The economic benefit from the practical implementation of this approach nationwide can range from 0.6 to 5.5 billion rubles annually.

Artyukova Z.R., Kudryavtsev N.D., Petraikin A.V., Abuladze L.R., Smorchkova A.K., Akhmad E.S., Semenov D.S., Belyaev M.G., Belaya Z.E., Vladzimirskyy A.V., Vasiliev Y.A.

Goal: To develop a method for automated assessment of the volumetric bone mineral density (BMD) of the vertebral bodies using an artificial intelligence (AI) algorithm and a phantom modeling method.Materials and Methods: Evaluation of the effectiveness of the AI algorithm designed to assess BMD of the vertebral bodies based on chest CT data. The test data set contains 100 patients aged over 50 y.o.; the ratio between the subjects with/without compression fractures (Сfr) is 48/52. The X-ray density (XRD) of vertebral bodies at T11-L3 was measured by experts and the AI algorithm for 83 patients (205 vertebrae). We used a recently developed QCT PK (Quantitative Computed Tomography Phantom Kalium) method to convert XRD into BMD followed by building calibration lines for seven 64-slice CT scanners. Images were taken from 1853 patients and then processed by the AI algorithm after the calibration. The male to female ratio was 718/1135.Results: The experts and the AI algorithm reached a strong agreement when comparing the measurements of the XRD. The coefficient of determination was R2=0.945 for individual vertebrae (T11-L3) and 0.943 for patients (p=0.000). Once the subjects from the test sample had been separated into groups with/without Сfr, the XRD data yielded similar ROC AUC values for both the experts – 0.880, and the AI algorithm – 0.875. When calibrating CT scanners using a phantom containing BMD samples made of potassium hydrogen phosphate, the following averaged dependence formula BMD =0.77*HU-1.343 was obtained. Taking into account the American College Radiology criteria for osteoporosis, the cut-off value of BMD<80 mg/ml was 105.6HU; for osteopenia BMD<120 mg/ml was 157.6HU. During the opportunistic assessment of BMD in patients aged above 50 years using the AI algorithm, osteoporosis was detected in 31.72% of female and 18.66% of male subjects.Conclusions: This paper demonstrates good comparability for the measurements of the vertebral bodies’ XRD performed by the AI morphometric algorithm and the experts. We presented a method and demonstrated great effectiveness of opportunistic assessment of vertebral bodies’ BMD based on computed tomography data using the AI algorithm and the phantom modeling.

Zhang J., Liu F., Xu J., Zhao Q., Huang C., Yu Y., Yuan H.

BackgroundAcute vertebral fracture is usually caused by low-energy injury with osteoporosis and high-energy trauma. The AOSpine thoracolumbar spine injury classification system (AO classification) plays an important role in the diagnosis and treatment of the disease. The diagnosis and description of vertebral fractures according to the classification scheme requires a great deal of time and energy for radiologists.PurposeTo design and validate a multistage deep learning system (multistage AO system) for the automatic detection, localization and classification of acute thoracolumbar vertebral body fractures according to AO classification on computed tomography.Materials and MethodsThe CT images of 1,217 patients who came to our hospital from January 2015 to December 2019 were collected retrospectively. The fractures were marked and classified by 2 junior radiology residents according to the type A standard in the AO classification. Marked fracture sites included the upper endplate, lower endplate and posterior wall. When there were inconsistent opinions on classification labels, the final result was determined by a director radiologist. We integrated different networks into different stages of the overall framework. U-net and a graph convolutional neural network (U-GCN) are used to realize the location and classification of the thoracolumbar spine. Next, a classification network is used to detect whether the thoracolumbar spine has a fracture. In the third stage, we detect fractures in different parts of the thoracolumbar spine by using a multibranch output network and finally obtain the AO types.ResultsThe mean age of the patients was 61.87 years with a standard deviation of 17.04 years, consisting of 760 female patients and 457 male patients. On vertebrae level, sensitivity for fracture detection was 95.23% in test dataset, with an accuracy of 97.93% and a specificity of 98.35%. For the classification of vertebral body fractures, the balanced accuracy was 79.56%, with an AUC of 0.904 for type A1, 0.945 for type A2, 0.878 for type A3 and 0.942 for type A4.ConclusionThe multistage AO system can automatically detect and classify acute vertebral body fractures in the thoracolumbar spine on CT images according to AO classification with high accuracy.

Dong Q., Luo G., Lane N.E., Lui L., Marshall L.M., Kado D.M., Cawthon P., Perry J., Johnston S.K., Haynor D., Jarvik J.G., Cross N.M.

Osteoporosis affects 9% of individuals over 50 in the United States and 200 million women globally. Spinal osteoporotic compression fractures (OCFs), an osteoporosis biomarker, are often incidental and under-reported. Accurate automated opportunistic OCF screening can increase the diagnosis rate and ensure adequate treatment. We aimed to develop a deep learning classifier for OCFs, a critical component of our future automated opportunistic screening tool.The dataset from the Osteoporotic Fractures in Men Study comprised 4461 subjects and 15,524 spine radiographs. This dataset was split by subject: 76.5% training, 8.5% validation, and 15% testing. From the radiographs, 100,409 vertebral bodies were extracted, each assigned one of two labels adapted from the Genant semiquantitative system: moderate to severe fracture vs. normal/trace/mild fracture. GoogLeNet, a deep learning model, was trained to classify the vertebral bodies. The classification threshold on the predicted probability of OCF outputted by GoogLeNet was set to prioritize the positive predictive value (PPV) while balancing it with the sensitivity. Vertebral bodies with the top 0.75% predicted probabilities were classified as moderate to severe fracture.Our model yielded a sensitivity of 59.8%, a PPV of 91.2%, and an F1 score of 0.72. The areas under the receiver operating characteristic curve (AUC-ROC) and the precision-recall curve were 0.99 and 0.82, respectively.Our model classified vertebral bodies with an AUC-ROC of 0.99, providing a critical component for our future automated opportunistic screening tool. This could lead to earlier detection and treatment of OCFs.

Kudryavtsev N.D., Semenov D.S., Kozhikhina D.D., Vladzymyrskyy A.V.

Vladzymyrskyy A.V., Kudryavtsev N.D., Kozhikhina D.D., Shulkin I.M., Morozov S.P., Ledikhova N.V., Klyashtornyy V.G., Goncharova I.V., Novikov A.V., Vnukova O.M.

Методы лучевой диагностики все более масштабно применяются при массовых профилактических осмотрах (скринингах) для выявления различных патологических состояний. С целью повышения эффективности программ скрининга в ряде ведущих стран мира предусмотрены двойные описания результатов исследований, что неоспоримо увеличивает рабочую нагрузку на врачей-рентгенологов. В связи с эти крайне актуальной становится задача автоматизированного анализа результатов скрининговых исследований. ЦЕЛЬ ИССЛЕДОВАНИЯ Оценить влияние делегирования полномочий по выполнению первого описания медицинскому программному обеспечению на основе технологий искусственного интеллекта (ИИ) на длительность процесса двойного описания результатов флюорографии. МАТЕРИАЛ И МЕТОДЫ Исследование выполнено на базе Московского референс-центра лучевой диагностики (ГБУЗ «НПКЦ ДиТ ДЗМ»). В исследование включено 13 901 флюорографическое исследование. Реализованы два сценария двойного описания исследований: в первом случае участвовали врач-рентгенолог и алгоритм ИИ, во втором — два врача-рентгенолога. Просмотр результатов ИИ и описание исследований проводилось в Едином радиологическом информационном сервисе Единой медицинской информационно-аналитической системы города Москвы. Выполнен статистический анализ данных. РЕЗУЛЬТАТЫ По сценарию №1 проведено двойное описание 1435 результатов флюорографии, по сценарию №2 — 12 446. В первом сценарии врач, получив данные машинного анализа, затрачивал на подготовку заключения в среднем 0,9±3,0 мин. Во втором сценарии продолжительность работы врача, осуществлявшего первый просмотр, составила 0,8±2,1 мин; второй просмотр — 0,4±1,0 мин. Общая длительность проведения двойного описания в формате «врач + ИИ» колебалась от 0,7 до 1241,2 мин, составив в среднем 199,3±330,3 мин. Во втором сценарии общая длительность проведения двойного описания составила 1 838,5±3671,4 мин. ЗАКЛЮЧЕНИЕ Делегирование первого описания алгоритму искусственного интеллекта принципиально ускоряет предоставление результатов флюорографии, повышая их доступность для обследованных лиц и медицинских работников, направляющих пациентов на обследование. Актуальнейшим вопросом становится точность работы соответствующих технологий искусственного интеллекта, а обязательность их регистрации в качестве медицинского изделия не подлежит дальнейшему обсуждению.

Morozov S.P., Vladzymyrskyy A.V., Shulkin I.M., Ledikhova N.V., Arzamasov K.M., Andreychenko A.E., Logunova T.A., Omelyanskaya O.V., Gusev A.V.

Andrianova M.G., Kudryavtsev N.D., Petraikin A.V.

.

Shelepa A.A., Petraikin A.V., Artyukova Z.R., Abuladze L.R., Kudryavtsev N.D., Ahmad E.S., Semenov D.S., Zakharov A.A., Belyaev M.G.

.

Lauritzen A.D., Rodríguez-Ruiz A., von Euler-Chelpin M.C., Lynge E., Vejborg I., Nielsen M., Karssemeijer N., Lillholm M.

Background Developments in artificial intelligence (AI) systems to assist radiologists in reading mammograms could improve breast cancer screening efficiency. Purpose To investigate whether an AI system could detect normal, moderate-risk, and suspicious mammograms in a screening sample to safely reduce radiologist workload and evaluate across Breast Imaging Reporting and Data System (BI-RADS) densities. Materials and Methods This retrospective simulation study analyzed mammographic examination data consecutively collected from January 2014 to December 2015 in the Danish Capital Region breast cancer screening program. All mammograms were scored from 0 to 10, representing the risk of malignancy, using an AI tool. During simulation, normal mammograms (score < 5) would be excluded from radiologist reading and suspicious mammograms (score > recall threshold [RT]) would be recalled. Two radiologists read the remaining mammograms. The RT was fitted using another independent cohort (same institution) by matching to the radiologist sensitivity. This protocol was further applied to each BI-RADS density. Screening outcomes were measured using the sensitivity, specificity, workload, and false-positive rate. The AI-based screening was tested for noninferiority sensitivity compared with radiologist screening using the Farrington-Manning test. Specificities were compared using the McNemar test. Results The study sample comprised 114 421 screenings for breast cancer in 114 421 women, resulting in 791 screen-detected, 327 interval, and 1473 long-term cancers and 2107 false-positive screenings. The mean age of the women was 59 years ± 6 (SD). The AI-based screening sensitivity was 69.7% (779 of 1118; 95% CI: 66.9, 72.4) and was noninferior (P = .02) to the radiologist screening sensitivity of 70.8% (791 of 1118; 95% CI: 68.0, 73.5). The AI-based screening specificity was 98.6% (111 725 of 113 303; 95% CI: 98.5, 98.7), which was higher (P < .001) than the radiologist specificity of 98.1% (111 196 of 113 303; 95% CI: 98.1, 98.2). The radiologist workload was reduced by 62.6% (71 585 of 114 421), and 25.1% (529 of 2107) of false-positive screenings were avoided. Screening results were consistent across BI-RADS densities, although not significantly so for sensitivity. Conclusion Artificial intelligence (AI)-based screening could detect normal, moderate-risk, and suspicious mammograms in a breast cancer screening program, which may reduce the radiologist workload. AI-based screening performed consistently across breast densities. © RSNA, 2022 Online supplemental material is available for this article.

Andreychenko A.E., Logunova T.A., Gombolevskiy V.A., Nikolaev A.E., Vladzymyrskyy A.V., Sinitsyn V.E., Morozov S.P.

AbstractIn recent years, there has been tremendous interest in the use of artificial intelligence (AI) in radiology in order to automate the interpretation. However, uncontrolled and widespread use of AI solutions may have negative consequences. Therefore, before implementing such technologies in healthcare, thorough training of personnel, adaptation of information systems, and standardized datasets for an external validation are required. All this necessitates a formation of a unique unified methodology. The best practices of AI introduction in diagnostic radiology are still subject to debate and require new results of a scientific-practical research with the assessment of implementation conditions.This work discusses expected issues and potential solutions for the introduction of computer vision-based technologies for automatic analysis of radiological examinations with an emphasis on the real-life experience gained during simultaneous AI implementation into practice of more than a hundred state radiology departments in 2020-2021 in Moscow, Russia (an experiment). The experiment used end-user software testing approaches, quality assurance of AI-based radiological solutions, and accuracy assessment of the AI-empowered diagnostic tools on local data. The methods were adapted and optimized to ensure a successful real-life radiological AI deployment on the extraordinary large scale. The experiment involved in total around thousand diagnostic devices and thousand radiologists. AI deployment was associated with additional options in a routine radiologist’s workflow: triage; additional series formed by AI with indication of pathological findings and their classification; report template prepared by AI in accordance with the target clinical task, user feedback on AI performance.A multi-stage methodology for implementing AI into radiological practice that was developed and advanced during the experiment is described in this report.EssentialsA methodology for the AI deployment for non-academic radiological sites excluded more than half of the offered AI solutions that do not fulfill the diagnostic and functional requirementsQuality control of AI should be supported by not only data scientists, IT specialists or engineers, but also by radiologists at all stages of selection and testing.Radiologists need to understand the capabilities, limitations of AI by getting an additional training.

Del Lama R.S., Candido R.M., Chiari-Correia N.S., Nogueira-Barbosa M.H., de Azevedo-Marques P.M., Tinós R.

Vertebral Compression Fracture (VCF) occurs when the vertebral body partially collapses under the action of compressive forces. Non-traumatic VCFs can be secondary to osteoporosis fragility (benign VCFs) or tumors (malignant VCFs). The investigation of the etiology of non-traumatic VCFs is usually necessary, since treatment and prognosis are dependent on the VCF type. Currently, there has been great interest in using Convolutional Neural Networks (CNNs) for the classification of medical images because these networks allow the automatic extraction of useful features for the classification in a given problem. However, CNNs usually require large datasets that are often not available in medical applications. Besides, these networks generally do not use additional information that may be important for classification. A different approach is to classify the image based on a large number of predefined features, an approach known as radiomics. In this work, we propose a hybrid method for classifying VCFs that uses features from three different sources: i) intermediate layers of CNNs; ii) radiomics; iii) additional clinical and image histogram information. In the hybrid method proposed here, external features are inserted as additional inputs to the first dense layer of a CNN. A Genetic Algorithm is used to: i) select a subset of radiomic, clinical, and histogram features relevant to the classification of VCFs; ii) select hyper-parameters of the CNN. Experiments using different models indicate that combining information is interesting to improve the performance of the classifier. Besides, pre-trained CNNs presents better performance than CNNs trained from scratch on the classification of VCFs.

Morozov S.P., Gavrilov A.V., Arkhipov I.V., Dolotova D.D., Lysenko M.A., Tsarenko S.V., Smorshok V.N., Parshin V.V., Korb T.A., Gonchar A.P., Blokhin I.A., Logunova T.A., Evteeva K.B., Andreychenko A.E., Vladzymyrskyy A.V., et. al.

Неоднократно описана практическая ценность компьютерной томографии (КТ) органов грудной клетки для диагностики пациентов с подозрением на коронавирусную инфекцию (COVID-19). В условиях высокой загруженности во время пандемии врачи-рентгенологи испытывают нехватку времени для интерпретации результатов исследований. Использование технологий искусственного интеллекта (ИИ) может повлиять на длительность формирования протокола описания КТ-исследований. ЦЕЛЬ ИССЛЕДОВАНИЯ Оценить влияние алгоритма ИИ на скорость описания результатов КТ органов грудной клетки при подозрении на COVID-19 в стационарном звене городского здравоохранения. МАТЕРИАЛ И МЕТОДЫ Проведено ретроспективное исследование, протокол зарегистрирован в ClinicalTrials.gov (NCT04489992). Исследование выполнено на основе данных пациентов, прошедших КТ органов грудной клетки в период с 08.04.20 по 01.12.20 в 105 медицинских организациях стационарного звена городского здравоохранения. КТ органов грудной клетки проводили по стандартным протоколам сканирования. Врачи-рентгенологи анализировали исследования с помощью ИИ-сервиса «Гамма Мультивокс Ковирус» и без него. Формирование протоколов медицинских заключений проведено в Едином радиологическом информационном сервисе в составе Единой медицинской информационно-аналитической системы Москвы. РЕЗУЛЬТАТЫ Без применения ИИ проанализированы 3133 КТ-исследований с признаками COVID-19-ассоциированной пневмонии (1-я группа), с использованием ИИ — 63 379 (2-я группа). Медианная длительность описаний в 1-й и 2-й группах составила 103,0 и 46,0 мин соответственно. Анализ длительности интерпретации врачом-рентгенологом результатов КТ органов грудной клетки до и после внедрения ИИ выявил статистически значимые различия (p<0,0001). Средняя длительность описания КТ органов грудной клетки при использовании ИИ уменьшилась на 29,4%. ЗАКЛЮЧЕНИЕ Внедрение технологии ИИ, направленной на диагностику изменений в легких при COVID-19 по данным КТ органов грудной клетки, в практическую работу отделений лучевой диагностики стационарного звена городского здравоохранения сокращает время подготовки протокола описания врачами-рентгенологами.

Cui X., Zheng S., Heuvelmans M.A., Du Y., Sidorenkov G., Fan S., Li Y., Xie Y., Zhu Z., Dorrius M.D., Zhao Y., Veldhuis R.N., de Bock G.H., Oudkerk M., van Ooijen P.M., et. al.

To evaluate the performance of a deep learning-based computer-aided detection (DL-CAD) system in a Chinese low-dose CT (LDCT) lung cancer screening program.One-hundred-and-eighty individuals with a lung nodule on their baseline LDCT lung cancer screening scan were randomly mixed with screenees without nodules in a 1:1 ratio (total: 360 individuals). All scans were assessed by double reading and subsequently processed by an academic DL-CAD system. The findings of double reading and the DL-CAD system were then evaluated by two senior radiologists to derive the reference standard. The detection performance was evaluated by the Free Response Operating Characteristic curve, sensitivity and false-positive (FP) rate. The senior radiologists categorized nodules according to nodule diameter, type (solid, part-solid, non-solid) and Lung-RADS.The reference standard consisted of 262 nodules ≥ 4 mm in 196 individuals; 359 findings were considered false positives. The DL-CAD system achieved a sensitivity of 90.1% with 1.0 FP/scan for detection of lung nodules regardless of size or type, whereas double reading had a sensitivity of 76.0% with 0.04 FP/scan (P = 0.001). The sensitivity for detection of nodules ≥ 4 - ≤ 6 mm was significantly higher with DL-CAD than with double reading (86.3% vs. 58.9% respectively; P = 0.001). Sixty-three nodules were only identified by the DL-CAD system, and 27 nodules only found by double reading. The DL-CAD system reached similar performance compared to double reading in Lung-RADS 3 (94.3% vs. 90.0%, P = 0.549) and Lung-RADS 4 nodules (100.0% vs. 97.0%, P = 1.000), but showed a higher sensitivity in Lung-RADS 2 (86.2% vs. 65.4%, P < 0.001).The DL-CAD system can accurately detect pulmonary nodules on LDCT, with an acceptable false-positive rate of 1 nodule per scan and has higher detection performance than double reading. This DL-CAD system may assist radiologists in nodule detection in LDCT lung cancer screening.

Petraikin A.V., Skripnikova I.A.

In the review we discussed about the method of quantitative computed tomography (QCT, quantitative computed tomography). In QCT, X-ray density (HU) is converted to bone mineral density (BMD mg / ml) using linear relationships obtained by scanning calibration standards (phantoms). When compared with the normative age data, it is possible to diagnose osteoporosis (OP). The review presents various QCT techniques and their diagnostic capabilities in accordance with the positions of ISCD 2019 - (International Society for Clinical Densitometry). The results of comparison of QCT and conventional dual-energy X-ray absorptiometry (DXA) are  considered.  It is noted that in the study of the proximal femur (PF), the results of the methods are well comparable, according to the results of both methods, it is possible to diagnose OP by the T-score. However, when examining the spine QCT, the volume BMD of the trabecular bone of the vertebral bodies is assessed, and with DXA, the projection BMD is assessed. The approaches to the interpretation of the results are also different - diagnosis of OP in DXA of the spine based on the T-score, but in QCT, the ACR (American College of Radiology) criteria are used.We describe the phantoms used in QCT, as well as provide data on radiation exposure during QCT and DXA.The article describes an approach to opportunistic screening of osteoporosis by the QCT based on the results of previously performed CT scans, including its automated work-flow using artificial intelligence technologies. These promising techniques are attractive due to the large number of CT examinations performed and the exclusion of additional examinations.