Roberta Hlavata

Background: Aneurysms are life-threatening vascular conditions that require early and accurate detection to prevent fatal outcomes. Methods: Advances in deep learning have demonstrated significant potential in medical image analysis, particularly for automated aneurysm detection. This review paper provides an overview of selected deep-learning approaches for aneurysm detection, with an emphasis on YOLO (You Only Look Once) models and their architectural characteristics, comparative performance, and applicability. Results: Existing YOLO-based studies are examined and compared, highlighting their strengths, limitations, dataset usage, performance metrics, and clinical relevance. To complement this overview, we include test using an annotated dataset of 1342 angiograms. Of these, 1074 angiograms were used for model training and 10% (approximately 107 angiograms) were reserved for validation; the remaining 268 were used for evaluation, with all annotations provided by a radiology specialist. Conclusions: These tests were conducted solely to provide practical examples and a limited comparative demonstration of the selected YOLO variants, rather than to constitute a full original experimental study. The findings underscore the potential of integrating detection models to improve accuracy and robustness in aneurysm identification, paving the way for more reliable and validated computer-aided diagnostic systems.

Deep learning models are increasingly evaluated not only for predictive accuracy but also for their robustness, interpretability, and data quality dependencies. However, current benchmarks largely isolate these dimensions, lacking a unified evaluation protocol that integrates data-centric and model-centric properties. To bridge the gap between data quality assessment and eXplainable Artificial Intelligence (XAI), we introduce DERI1000—the Dataset Explainability Readiness Index—a benchmark that quantifies how suitable and well-prepared a dataset is for explainable and trustworthy deep learning. DERI1000 combines eleven measurable factors—sharpness, noise artifacts, exposure, resolution, duplicates, diversity, separation, imbalance, label noise proxy, XAI overlay, and XAI stability—into a single normalized score calibrated around a reference baseline of 1000. Using five MedMNIST datasets (PathMNIST, ChestMNIST, BloodMNIST, OCTMNIST, OrganCMNIST) and five convolutional neural architectures (DenseNet121, ResNet50, ResNet18, VGG16, EfficientNet-B0), we fitted factor weights through multi-dataset impact analysis. The results indicate that imbalance (0.3319), separation (0.1377), and label noise proxy (0.2161) are the dominant contributors to explainability readiness. Experiments demonstrate that DERI1000 effectively distinguishes models with superficially high accuracy (ACC) but poor interpretability or robustness. The framework thereby enables cross-domain, reproducible evaluation of model performance and data quality under unified metrics. We conclude that DERI1000 provides a scalable, interpretable, and extensible foundation for benchmarking deep learning systems across both data-centric and explainability-driven dimensions.

This study explores the extraction of remote Photoplethysmography (rPPG) signals from images using various neural network architectures, addressing the challenge of accurate signal estimation in biomedical contexts. The objective is to evaluate the effectiveness of different models in capturing rPPG signals from dataset snapshots. Two training strategies were investigated: pre-training models with only the fully connected layer being fine-tuned and training the entire network from scratch. The analysis reveals that models trained from scratch consistently outperform their pre-trained counterparts in extracting rPPG signals. Among the architectures assessed, DenseNet121 demonstrated superior performance, offering the most reliable results in this context. These findings underscore the potential of neural networks in advancing rPPG signal extraction, which has promising applications in fields such as clinical monitoring and personalized medical care. This study contributes to the integration of advanced imaging techniques and neural network-based analysis in biomedical engineering, paving the way for more robust and efficient methodologies.

There are multiple uses for single-channel images, such as infrared imagery, depth maps, and others. To automatically classify objects in such images, an algorithm suited for single-channel image processing is required. This study explores the application of deep learning techniques for the recognition of wild animals using infrared images. Traditional methods of wildlife monitoring often rely on visible light imaging, which can be hindered by various environmental factors such as darkness, fog, and dense foliage. In contrast, infrared imaging captures the thermal signatures of animals, providing a robust alternative for wildlife detection and identification. We test a Convolutional Neural Network (CNN) model specifically designed to analyze infrared images, leveraging the unique thermal patterns emitted by different animal species. The model is trained and tested on a diverse dataset of infrared images, demonstrating high accuracy in distinguishing between multiple species. In this paper, we also present a comparison of several well-known artificial neural networks on this data. To ensure accurate testing, we introduce a new dataset containing infrared photos of Slovak wildlife, specifically including classes such as bear, deer, boar, and fox. To complement this dataset, the Fashion MNIST dataset was also used. Our results indicate that deep learning approaches significantly enhance the capability of infrared imaging for wildlife monitoring, offering a reliable and efficient tool for conservation efforts and ecological studies.

Intracranial aneurysm (IA) is now a common term closely associated with subarachnoid hemorrhage. IA is the bulging of a blood vessel caused by a weakening of its wall. This bulge can rupture and, in most cases, cause internal bleeding. In most cases, internal bleeding leads to death or other fatal consequences. Therefore, the development of an automated system for detecting IA is needed to help physicians make more accurate diagnoses. For this reason, we have focused on this problem. In this paper, we propose a 2D Convolutional Neural Network (CNN) based on a network commonly used for data classification in medicine. In addition to our proposed network, we also tested ResNet 50, ResNet 101 and ResNet 152 on a publicly available dataset. In this case, ResNet 152 achieved better results than our proposed network, but our network was significantly smaller and the classifications took significantly less time. Our proposed network achieved an overall accuracy of 98%. This result was achieved on a dataset consisting of 611 images. In addition to the mentioned networks, we also experimented with the VGG network, but it was not suitable for this type of data and achieved only 20%. We compare the results in this work with neural networks that have been verified by the scientific community, and we believe that the results obtained by us can help in the creation of an automated system for the detection of IA.

Currently, three-dimensional convolutional neural networks (3DCNNs) are a popular approach in the field of human activity recognition. However, due to the variety of methods used for human activity recognition, we propose a new deep-learning model in this paper. The main objective of our work is to optimize the traditional 3DCNN and propose a new model that combines 3DCNN with Convolutional Long Short-Term Memory (ConvLSTM) layers. Our experimental results, which were obtained using the LoDVP Abnormal Activities dataset, UCF50 dataset, and MOD20 dataset, demonstrate the superiority of the 3DCNN + ConvLSTM combination for recognizing human activities. Furthermore, our proposed model is well-suited for real-time human activity recognition applications and can be further enhanced by incorporating additional sensor data. To provide a comprehensive comparison of our proposed 3DCNN + ConvLSTM architecture, we compared our experimental results on these datasets. We achieved a precision of 89.12% when using the LoDVP Abnormal Activities dataset. Meanwhile, the precision we obtained using the modified UCF50 dataset (UCF50mini) and MOD20 dataset was 83.89% and 87.76%, respectively. Overall, our work demonstrates that the combination of 3DCNN and ConvLSTM layers can improve the accuracy of human activity recognition tasks, and our proposed model shows promise for real-time applications.

This work deals with contactless detection of heart rate from facial region using the principles of photoplethysmography imaging (PPGI) and low-cost web camera. Our proposed algorithm was realized in MATLAB environment. Using suitable face trackers, the face is tracked, then the signal is obtained from skin and subsequently the signal processing was performed. After obtaining heart rate, it is compared with heart rate signal obtained from traditional contact photoplethysmography (PPG). Obtained heart rate (BPM) can be used in stress evaluation in educational process.

Recognizing various abnormal activities of human behavior from video is very challenging. The overall results are affected by the available data-sets. The available data-sets contain various abnormal activities, but few of them focus mainly on non-standard human behavior. In data-sets such as KTH, they focus on abnormal activities such as a sudden change in behavior or an object occurrence where it should not occur but also various changes in human interaction. The UCF-crime data-set focuses on data that are more interesting to us, such as fight, abuse, explosions or robbery etc. However, the data-set is very demanding due to the videos length, which contains a given event in just a few seconds. This may affect the overall results of the algorithm used to detect the incident. In this paper, we create a data-set dealing with abnormal activities such as robbery, fight, hijack, harassment and a normal videos. We use the created data-set when training and testing the neural network ConvLSTM (Convolutional Long Short-Term Memory). We have obtained a classification accuracy of 97.64 % on the created data-set and used architecture of neural network.

Recognizing various abnormal human activities from video is very challenging. This problem is also greatly influenced by the lack of datasets containing various abnormal human activities. The available datasets contain various human activities, but only a few of them contain non-standard human behavior such as theft, harassment, etc. There are datasets such as KTH that focus on abnormal activities such as sudden behavioral changes, as well as on various changes in interpersonal interactions. The UCF-crime dataset contains categories such as fighting, abuse, explosions, robberies, etc. However, this dataset is very time consuming. The events in the videos occur in a few seconds. This may affect the overall results of the neural networks that are used to detect the incident. In this article, we create a dataset that deals with abnormal activities, containing categories such as Begging, Drunkenness, Fight, Harassment, Hijack, Knife Hazard, Normal Videos, Pollution, Property Damage, Robbery, and Terrorism. We use the created dataset for the training and testing of the ConvLSTM (convolutional long short-term memory) neural network, which we designed. However, we also test the created dataset using other architectures. We use ConvLSTM architectures and 3D Resnet50, 3D Resnet101, and 3D Resnet152. With the created dataset and the architecture we designed, we obtained an accuracy of classification of 96.19% and a precision of 96.50%.