Vol 7, No 2 (2026)

BACKGROUND: In the era of low-dose computed tomography replacing chest radiography, attention is increasingly directed not only to lung pathology but also to other organ systems. Early detection of cardiovascular changes and prompt treatment response minimize damage to patient health, potentially reducing mortality and improving both survival and quality of life. The introduction of artificial intelligence into the radiology workflow is considered a significant supportive tool to reduce the omission of clinically important findings and improve the quality of radiology reports.

AIM: To compare the diagnostic accuracy of the xAID Chest CT Module artificial intelligence model and radiologists in detecting peripheral solid pulmonary nodules and clinically significant findings, including coronary artery calcification, and pulmonary artery and ascending aortic dilation on routine chest computed tomography.

METHODS: This is a retrospective, cross-sectional, single-center study involving 300 chest computed tomography studies were processed by the artificial intelligence model. The reference standard was established by an expert radiologist. Additionally, the respective radiology reports were reviewed to compare the diagnostic accuracy of the artificial intelligence model and the radiologists in detecting peripheral pulmonary nodules, coronary calcification, and pathological dilation of the ascending aorta and pulmonary trunk.

RESULTS: For peripheral pulmonary nodules ≥ 6 mm, the xAID Chest CT Module demonstrated a sensitivity of 81.6% and a specificity of 91.0%, whereas routine radiology reports demonstrated 100% diagnostic accuracy in detecting pulmonary nodules. For coronary artery calcification, the artificial intelligence demonstrated a sensitivity of 97.0% and a specificity of 50.0%, compared with a radiologist sensitivity of 79.8% and a specificity of 100%. For main pulmonary artery dilation ≥ 29 mm, artificial intelligence sensitivity and specificity were 95.8% and 78.1%, respectively, whereas the radiologist achieved a sensitivity of 56.9% and a specificity of 100%. For ascending aortic dilation ≥ 40 mm, artificial intelligence reached sensitivity of 100% and specificity of 98.9%, compared with a radiologist sensitivity of 66.7% and a specificity of 100%.

CONCLUSION: The xAID Chest CT Module cannot be considered a standalone tool for diagnosing peripheral pulmonary nodules and coronary calcification; rather, it should be regarded as a support tool in the radiologist workflow to draw attention to small nodules and manifestations of coronary calcification. However, the model's high diagnostic accuracy in assessing the diameters of the ascending aorta and pulmonary trunk offers the potential of a standalone tool for opportunistic screening of vascular pathology using chest computed tomography data.

BACKGROUND: Suboptimal quality of chest radiographs is a common problem in the diagnostic imaging of thoracic diseases. Quality assessment of routine chest radiographs is performed manually, which is complicated by remote reporting and the high workload of radiologists. We previously developed an artificial intelligence-powered tool for quality control of these images. However, its performance in real-world clinical practice has not been investigated.

AIM: To evaluate the effectiveness of the automated artificial intelligence tool for detecting defects in chest radiographs in routine clinical practice.

METHODS: The tool was validated using data from three Moscow outpatient hospitals over a two-month period. Chest radiographs were assessed using an automated quality assessment tool based on ensemble learning that sequentially processes chest radiographs in DICOM (Digital Imaging and Communications in Medicine) format. The tool is integrated into the routine chest radiography workflow at the Moscow Reference Center for Radiology. Following automated assessment, image quality was analyzed by radiologists (n = 10). Based on the automated assessment, a report was generated that included information on incomplete visualization of the lungs and costodiaphragmatic recess in frontal and lateral view, chest rotation, incorrect image orientation, and incorrect information (metadata) about the anatomical region, view, and photometric interpretation of the study. Radiologists assessed radiographic image defects using similar criteria. Each image was evaluated by a single radiologist. The radiologists were blinded to the automated image evaluation results.

RESULTS: A total of 9642 radiographs were processed. The mean processing time per image was 14 seconds. In total, 3386 distinct patient positioning errors were identified, with each error detected in 14%–45% of images. The accuracy of metadata completion for radiographs ranged from 14% to 100%. The accuracy, expressed as the ROC AUC, for detecting patient positioning errors using the quality control tool ranged from 0.782 to 0.947, while the accuracy for detecting metadata completion errors ranged from 0.985 to 1.0.

CONCLUSION: Automated quality control of radiographs is an accurate and rapid method for detecting patient positioning errors and errors in metadata completion.

BACKGROUND: Cognitive load monitoring plays a key role in personalized neurorehabilitation and evaluation of educational program effectiveness. However, existing electroencephalographic systems with a large number of electrodes are technically complex and unsuitable for everyday use. Insufficient studies have explored the minimum electrode configurations necessary for a reliable assessment of the tasks' complexity under cognitive load.

AIM: To determine the minimum electroencephalographic electrode configurations that ensure reliable identification of cognitive load levels during multitasking (MATB-II) and working memory (N-Back) tasks, using feature importance and attribution methods, as well as to evaluate the applicability of electrode configurations proposed in previous studies.

METHODS: A cross-sectional, single-center study using retrospective data was conducted. Cognitive load was assessed using MATB-II and N-Back tasks from the open COGnitive Brain-Computer Interface dataset. Raw electroencephalographic signals were filtered and subjected to spectral analysis with allocation of power in the θ, α, β and γ bands. The Kolmogorov–Arnold Networks model was used for training and selecting informative features to enable feature importance attribution and determine the minimum electrode configurations.

RESULTS: The sns_att4 configuration with four channels and a value of F1 = 0.738 turned out to be the best for the MATB-II task. The sns_gp8 configuration with eight channels and F1 = 0.695 is the best for the N-Back task. Using the Kolmogorov–Arnold Networks model combined with the attribution method enabled the identification of compact electrode configurations (3–10 channels) that maintain high classification accuracy. The optimal 9-electrode configuration (ours_9) for MATB-II provided the mean F1 of 0.714. The three-electrode configuration (ours_3) for N-Back reached F1 = 0.617. Statistical analysis has confirmed that the proposed configurations are competitive with previously published schemes and, in some cases (ours_3 for N-Back), demonstrate an advantage with a minimum number of channels.

CONCLUSION: The identified configurations with a limited number of electrodes ensure reliability in determining a cognitive load. They can be used at home setting at the third stage of rehabilitation and integrated into portable neural devices.

BACKGROUND: Developing highly sensitive diagnostic methods for common extrapyramidal disorders such as Parkinson's disease and essential tremor is a pressing research challenge. Despite the rapid development of various diagnostic methods, including functional radioisotope neuroimaging, differential diagnosis of Parkinson's disease and essential tremor is often challenging. Electrooculographic signals provide additional information that can be leveraged to develop new approaches to the differential diagnosis of Parkinson's disease and essential tremor.

AIM: To identify specific quantitative characteristics of saccadic eye movements in Parkinson's disease and essential tremor patients using a new statistical analysis method for biomedical signals.

METHODS: An observational, cross-sectional, single-center study with retrospective data analysis was conducted. Outpatients were divided into two groups: Group 1, patients with Parkinson's disease; and Group 2, patients with essential tremor. During electrooculogram, patients were presented with a visual stimulus on a monitor: a white square on a black background moving from bottom to top within a 15°visual field. The stimulus was presented eight times, each presentation lasting three seconds. Electrooculographic signals were analyzed in their raw form, without prior signal segmentation. Statistical analysis was performed using macrosaccade analysis software and burst electrical activity analysis software with ROC-AUC diagrams.

RESULTS: Group 1 included 17 patients in stage 1–3 Parkinson's disease, and Group 2 included 10 patients with essential tremor. A decrease in electrooculographic macrosaccade latency was observed in both Group 1 and Group 2 in left and right eyes during the presentation of a series of the visual stimuli (Kendall's τ_b, p ≤ 0.031). The analysis of electrooculographic micromovements revealed wave trains that differentiated Parkinson's disease and essential tremor patient groups (two-tailed Brunel–Munzel test; p ≤ 0.03 for both eyes). A correlation was found between the number of spikes in electrooculographic of the left and right eye in Parkinson's disease patients (Kendall's τ_b is 0.515; p = 0.011).

CONCLUSION: The new method of statistical signal analysis enabled the identification of new quantitative features in electrooculographic in Parkinson's disease and essential tremor patients. These new features were found in both macrosaccade and eye micromovement parameters. The identified features are offer potential for new methods for differential diagnosis of Parkinson's disease and essential tremor.

This review is devoted to the application of neural network models for the analysis of histological images in non-neoplastic liver diseases, with a focus on the technical aspects of model development and training. The relevance of this work is due to the growing interest in applying neural networks and computer vision for research and diagnostic tasks related to microscopic morphology. Although this disease group does not lose its clinical significance, and the importance of microscopic verification of morphological changes, the application of artificial intelligence in this field remains limited and fragmented.

The review systematizes available annotated datasets of liver histological images, applied neural network architectures, image preprocessing approaches, and training strategies. It also examines the loss functions used and other key technical aspects of neural network development. It is shown that, although the contribution of such models to the automation and standardization of the morphological assessment appears promising, their practical implementation is constrained by the limited publicly available annotated data, the high labor intensity of annotation, and insufficient standardization of methodological approaches. It is noted that multimodal data, integrating histological images with clinical, biochemical, or radiological parameters, is of great clinical interest; however, it is currently rare.

Most studies utilize universal neural network architectures, while models fine-tuned to work with microscopic features and liver morphology are applied much less commonly. The analysis of published studies has shown that weakly supervised learning can be sufficient for model development and, in combination with significantly lower annotation costs, offers substantial potential for further development. At the same time, a considerable proportion of studies describe methodologies with limited reproducibility.

Magnetic resonance spectroscopy is a noninvasive method that enables analysis of the metabolic profile of liver tissue and the identification of biochemical indicators of malignant transformation. In this review we summarize current research data on the use of magnetic resonance spectroscopy in the differential diagnosis of benign and malignant focal liver lesions and to assess the diagnostic value of metabolic indices. A search and selection of publications were performed in the PubMed, Web of Science, and Google Scholar databases using the following keywords: "magnetic resonance spectroscopy," "MRS," "liver lesion," "MRI," "choline/lipid," and "MRS metabolites." The analysis included articles published from inception of the said databases up to December 2024 that focused on the application of magnetic resonance spectroscopy for evaluating the metabolic profile of liver tumors. Numerous studies indicate that magnetic resonance spectroscopy allows for the detection of characteristic changes in the levels of choline, lipids, and other metabolites reflecting membrane and energy exchange processes. These parameters offer additional noninvasive criteria for differentiating benign and malignant liver lesions. Although routine clinical use of magnetic resonance spectroscopy remains limited, the method demonstrates high potential for the early detection of tumors and the assessment of their metabolic characteristics. Further research may contribute to the technique standardization and a deeper understanding of the biochemical mechanisms underlying tumor growth, which is particularly important in complex and diagnostically challenging cases.

Air embolism is considered a rare but potentially fatal complication of invasive thoracic surgery. Computed tomography-guided thoracic biopsy is widely employed in clinical practice as a highly effective and relatively safe method for the morphological profiling of lung tumors. However, some publications indicate the possibility of systemic air embolism leading to subsequent severe cardiovascular and neurological complications. At present, predisposing factors, the pathophysiological mechanisms underlying the systemic embolism and the optimal acute-phase management remain poorly understood, making clinical observations of scientific and practical value.

In this case report, a 75-year-old patient developed systemic air embolism after computed tomography-guided transthoracic needle biopsy of a right upper-lobe tumor, which subsequently resulted in acute myocardial infarction and acute ischemic cerebrovascular accident. The patient received conservative therapy in the intensive care unit and was discharged with a neurological deficit for rehabilitation. Histological and immunohistochemical tests confirmed lung adenocarcinoma. Six months thereafter, radical surgical intervention — a right upper lobectomy — was performed.

This case underscores the need for careful patient selection for invasive diagnostics, mandatory early post-procedure monitoring, and readiness of medical staff to promptly recognize and manage rare complications. It broadens the clinical spectrum of systemic air embolism and emphasizes the importance of further investigation in modern thoracic oncology.