Background. Urolithiasis (UCD) is a chronic disease associated with metabolic disorders in the body, which is accompanied by the formation of stones in the organs of the urinary system. Today, the problem of detecting stones in the urinary system, despite all the achievements, remains relevant (increase in morbidity, changes in living conditions (ecology, nutrition)). To solve the difficulties that arise in the process of diagnosing stones in the urinary system, as well as optimizing the diagnostic process itself, including the assessment of the information received and the formation of a protocol, it is currently proposed to use software based on artificial intelligence technologies (AI-based software).

Objective. To evaluate the diagnostic accuracy of AI services in the task of primary binary detection of radiopaque urinary system calculi (>100 HU) without stratification by morphometric characteristics in computed tomography (CT) studies.

Materials аnd methods. A data set of 100 CT scans of the abdominal organs and urinary system was taken from Unified Radiological Information Service of the Unified Medical Information and Analytical System (ERIS EMIAS) and sent for processing to two companies with software based on TII.

Results. Based on the results of testing on the prepared data set with signs of urinary tract stones, the software based on the TII achieved high indicators of diagnostic accuracy metrics exceeding the stated indicators: area under the curve (AUC) up to 0.998 (confidence interval (CI) 0.993–1.000) (AI-2 comprehensive service for analysis of abdominal organs (AOR)), specificity up to 1.000 ((CI 0.900–1.000), sensitivity up to 1.000 (CI 1.000–1.000).

Conclusion. High indicators of diagnostic accuracy of the software based on the TII based on the results of the analysis of the test data set showed that the TII algorithms can be used as a tool for optimizing the diagnostic process, including the detection and morphometry of urinary tract stones.

Background. Augmented reality (AR) technologies are being actively integrated into medicine, including neurosurgery, where their application can enhance the accuracy and safety of operations and improve training effectiveness. AR navigation and teleproctoring have the potential to minimize risks, reduce intervention time, and improve access to specialized care in remote regions.

Objective. To evaluate the effectiveness of AR technologies for neuronavigation, teleproctoring, and training inexperienced surgeons using animal models, as well as simulating basic neurosurgical procedures (lumbar puncture).

Materials and methods. A prospective study was conducted on laboratory animals: 10 rats and 1 sheep. Microsoft Hololens 2 AR glasses and the Medgital Vision application were used. Access to the aorta and carotid arteries was performed on 10 rats (5 with AR navigation, 5 without). An extracranial-intracranial anastomosis was performed on the sheep under AR-guided teleproctoring via Google Meet. Additionally, a lumbar puncture simulation was conducted by projecting patient CT data onto a participant’s body.

Results. The use of AR reduced operation time and incision size. Remote teleproctoring proved effective for intraoperative training and consultation. In the lumbar puncture simulation, anatomical landmarks were accurately projected onto the participant’s body.

Conclusion. AR technologies demonstrated high efficacy for neuronavigation, teleproctoring, and training. Their implementation in clinical practice could enhance surgical safety, reduce intervention time, and expand opportunities for remote collaboration among specialists.

In order to select the most appropriate and effective treatment tactics for patients with clear cell renal cancer (ccRC), a multi-phase computed tomography (CT) with intravenous contrast is performed during the preoperative stage. It is possible that textural analysis data obtained from segmentation of CT data can be used to predict the likelihood of high-grade ccRC, which could potentially adjust treatment tactics.

Purpose: to perform a textural analysis of the preoperative multiphase CT data in patients with verified ссRC, to identify texture features that allow predicting ccRC grade

Material and methods: Retrospective analysis of data from 189 pre-operative CT studies was performed on 164 patients who underwent renal resection or nephrectomy, and the ccRC was morphologicaly verified using the WHO/ISUP grade classification system. CcRC grade 1–2 was verified in 130 cases, ccRC grade 3–4 was verified in 59 cases. For the performance of textural analysis, data from the corticomedullar, nephrographic and excretory CT phases were used, segmentation was carried out with 2D and 3D methods. The study was conducted retrospectively.

Results. The most informative in ccRC grade predicting were the textural features obtained from 3D tumor segmentation in the corticomedullar and excretory CT phases.

The largest number of textural features linearly associated with the ccRC grade 3–4 odds logarithm was obtained from a textural analysis of 3D tumor segmentation data in the corticomedullar and nephrographic CT phases.

The smallest number of textural features statistically significantly associated with the ccRC grade was obtained from a textural analysis of 2D segmentation data in the excretory phase of CT.

In conducting a multi-factor analysis of texture features as predictors of the ccRC grade, the most effective way to obtain texture features was to analyze the results of 3D segmentation into the excretory phase using the first 6 main components as predictors, explaining the 80% and more dispersion of the values of textural features.

The highest values of Nagelkerke pseudo-R² (0.49) and AUC (0.87) were found in a multi-factor logistic regression model, which included SHAPE_Sphericity and SHAPE_Compacity as precursors to 3D segmentation in the excretory phase when using the Reflect filter.

The model with one textural indicator included – DISCRETIZED_TLG – in 2D segmentation into the excretory phase using a Reflect filter was characterized by slightly lower Nagelkerke pseudo-R² (0.44) and AUC (0.85).

Based on the data obtained, we have developed two multi-factor models for predicting the probability of a grade 3–4 ccRC, based on 3D and 2D methods of tumor segmentation in CT images. The model using 3D segmentation data was characterized by predictive accuracy, sensitivity and specificity of 80%, 79.7% and 80.2%, respectively. Discriminative characteristics of the model using 3D segmentation data have surpassed those of the model using 2D segmentation data.

Conclusion. In the course of our work, we proved that it is possible to use textural features obtained from the analysis of CT data with contrast enhancement to predict the ccRC grade and developed a high-precision diagnostic model for predicting ccRC grade.

Purpose of the study: The aim of this study is to identify radiomic features that are invariant to segmentation variability, and to evaluate their ability to differentiate between benign and malignant lesions.

Materials and methods. 64 CT studies from the LIDC/IDRI database were selected to search for features invariant to segmentation variability. The existing segmentations of 64 LIDC/IDRI studies was supplemented with segmentations by radiologists from the Moscow Center for Diagnostics and Telemedicine, performed using a semi-automatic method (3D sphere). For the first 10 studies, manual segmentation (contour brush) was also additionally performed to compare the time of segmentation. On an unrelated WORC GIST dataset, segmentation-invariant features were evaluated in terms of “benign lesion-malignant lesion.”

Results. The study identified 7 radiomic features invariant to segmentation variability for computed tomography studies that showed discriminatory ability for the binary classification task “benign formation-malignant formation” on different dataset.

Conclusion. The identified features can be taken into account in the further development of the radiomic model. Semi-automatic annotation can significantly reduce the time required to obtain a mask of the region of interest for further radiomic analysis.