The advancement of deep learning methods across various applications has forced the creation of enormous training datasets. However, obtaining suitable real-world datasets is often challenging for various reasons. Consequently, numerous studies have emerged focusing on the generation and utilization of synthetic data in the training process. Hence, there is no universal formula for preparing synthetic data and leveraging it in network training to maximize the effectiveness of various detection methods. This work provides a comprehensive overview of several synthetic data generation techniques, followed by a thorough investigation into the impact of training methods and the selection of synthetic data quantities. The outcomes of this research enable the formulation of conclusions regarding the recipe for developing synthetic data with high efficacy in enhancing detection methods. The main conclusion for the synthetic data generation methods is to ensure maximum diversity at a high level of photorealism, which allows improving the classification quality by more than 5% to even 19% for different detection metrics.

The application of image analysis methods to calculate the distance from the camera to the object allows the replacement of specialized hardware devices for distance estimation. In the case of public transport, estimation of the exact position of the passenger gives also the option to determine whether the passenger is inside or outside the vehicle. In the presented work, several distance estimation methods based on typical analytical models and machine learning (ML) methods were tested using recordings from three cameras located in the minibus model. Human head detection was used instead of the entire passenger body to avoid occlusion problems. The analytical method showed worse performance than ML methods in all cases. The difference in the performance of ML models between cameras was negligible and there was no best method found. The computational time for ML models ranges from 0.35 to 100.57 ms, which should result in successful real-world applications. The developed approach can be used not only in public transport but also in all closed areas for the calculation of people or crowd density.

Chronic obstructive pulmonary disease (COPD) is the third leading cause of death worldwide, and emphysema is present in the majority of affected patients and can be identified on computed tomography (CT). This study investigated whether radiomic features derived from automatically and adaptively segmented low-attenuation lung regions can capture distinct imaging characteristics of COPD beyond conventional emphysema measures. Radiomic features were extracted from 6078 chest CT scans of 2243 participants from the COPDGene cohort. Emphysematous regions were segmented using the MimSeg method based on Gaussian mixture modelling with patient-adjusted thresholding, and radiomic features were computed for individual lesion clusters and aggregated per patient using summary statistics, yielding 780 features per subject. Uniform Manifold Approximation and Projection (UMAP) was used to generate a low-dimensional embedding, and feature contributions were evaluated using SHAP analysis and statistical testing. The resulting embedding demonstrated structured patterns broadly aligned with Global Initiative for Chronic Obstructive Lung Disease (GOLD) stages, with greater overlap among GOLD 0–2 and more consolidated groupings for GOLD 3 and 4, reflecting differences in disease severity. The most influential features were predominantly derived from Grey Level Run Length Matrix measures, capturing textural heterogeneity and spatial organisation of emphysematous changes that are not directly described by standard density-based metrics. These findings suggest that radiomic analysis of adaptively segmented CT data may provide complementary and structurally distinct information relative to conventional emphysema measures, supporting a more nuanced characterisation of emphysema patterns in COPD.

This study proposes a methodology for the automated localisation of lower limb bones on T1 weighted MRI scans, employing a deep learning (DL) approach. The primary objective is to facilitate precise identification of skeletal structures, thereby supporting radiomics based diagnostics of neuromuscular disorders. The developed framework is not confined to the recognition of lower limb bones. A dataset of 1,243 MRI scans was used, with a subset of 29 manually labelled bone segmentations of six key lower limb bone classes. Axial slices were divided into training (2,283), validation (300), and hold-out (378) sets. A two part segmentation pipeline was developed using a combination of U-Net and ResNet architectures, with a custom cost function to handle variable label presence across slices. A novel method for obtaining precise bone-related slice location on the MRI volume was developed. Segmentation quality for Tibia and Femur was high, achieving 86.04% and 86.97% Dice score on the hold-out subset. The ResNet classifier correctly identified the defined regions on the volume, achieving AUCs over 97% on the hold-out subset for most leg fragments except for the knee label. This work introduces a new method for anatomical spatial localisation estimation in MRI scans. Unlike previous studies, which could only recognise body parts, the proposed method also estimates the precise bone-related slice location within the scanned volume, providing an added layer of anatomical context. The developed pipeline can be retrained for other body fragments. This solution exhibits a strong potential for use in clinical workflows, especially for studies involving musculoskeletal diseases.