Abstract
Cardiac auscultation is still the most cost-effective screening procedure for cardiovascular diseases. The development of computer assisted methods can empower a large variety of health professionals and thus enable mass cardiac health low-cost screening. The procedure for correct cardiac auscultation includes listening to the heart sounds of the four main auscultation spots. Until recently, attempts to develop automatic heart sound analysis methods that explore the multi-channel richness of a real auscultation, were very difficult due to the lack of adequate public datasets. In this work, we use the CirCor Dataset which is characterized by the existence of more than one heart sound per patient (each patient has heart sounds collected at different auscultation spots). Using this dataset, we evaluate and quantify the comparative impact of using a single or a multichannel approach. A single channel approach uses the sound from a single auscultation spot, whereas a multi-channel approach uses four auscultation spots in an asynchronous way. From the different classifiers tested, models that use four auscultation spots achieved a higher overall performance than those that search for abnormalities in a single heart sound spot. Our best result is a multi-channel SVM that analyzes four auscultation spots, with an overall performance of 87,4 %. This opens the path to future research using a multi-channel approach.

Abstract
Physiological signals are often corrupted by noisy sources. Usually, artificial intelligence algorithms analyze the whole signal, regardless of its varying quality. Instead, experienced cardiologists search for a high-quality signal segment, where more accurate conclusions can be draw. We propose a methodology that simultaneously selects the optimal processing region of a physiological signal and determines its decoding into a state sequence of physiologically meaningful events. Our approach comprises two phases. First, the training of a neural network that then enables the estimation of the state probability distribution of a signal sample. Second, the use of the neural network output within an integer program. The latter models the problem of finding a time window by maximizing a likelihood function defined by the user. Our method was tested and validated in two types of signals, the phonocardiogram and the electrocardiogram. In phonocardiogram and electrocardiogram segmentation tasks, the system's sensitivity increased on average from 95.1% to 97.5% and from 78.9% to 83.8%, respectively, when compared to standard approaches found in the literature.

Abstract
Colorectal cancer is the third most common cancer and the second cause of cancer-related deaths in the world. Colonoscopic surveillance is extremely important to find cancer precursors such as adenomas or serrated polyps. Identifying small or flat polyps can be challenging during colonoscopy and highly dependent on the colonoscopist's skills. Deep learning algorithms can enable improvement of polyp detection rate and consequently assist to reduce physician subjectiveness and operation errors. This study aims to compare YOLO object detection architecture with self-attention models. In this study, the Kvasir-SEG polyp dataset, composed of 1000 colonoscopy annotated still images, were used to train (700 images) and validate (300images) the performance of polyp detection algorithms. Well-defined architectures such as YOLOv4 and different YOLOv5 models were compared with more recent algorithms that rely on self-attention mechanisms, namely the DETR model, to understand which technique can be more helpful and reliable in clinical practice. In the end, the YOLOv5 proved to be the model achieving better results for polyp detection with 0.81 mAP, however, the DETR had 0.80 mAP proving to have the potential of reaching similar performances when compared to more well-established architectures. © 2023 by SCITEPRESS - Science and Technology Publications, Lda.

Abstract
Precancerous conditions such as intestinal metaplasia (IM) have a key role in gastric cancer development and can be detected during endoscopy. During upper gastrointestinal endoscopy (UGIE), misdiagnosis can occur due to technical and human factors or by the nature of the lesions, leading to a wrong diagnosis which can result in no surveillance/treatment and impairing the prevention of gastric cancer. Deep learning systems show great potential in detecting precancerous gastric conditions and lesions by using endoscopic images and thus improving and aiding physicians in this task, resulting in higher detection rates and fewer operation errors. This study aims to develop deep learning algorithms capable of detecting IM in UGIE images with a focus on model explainability and interpretability. In this work, white light and narrow-band imaging UGIE images collected in the Portuguese Institute of Oncology of Porto were used to train deep learning models for IM classification. Standard models such as ResNet50, VGG16 and InceptionV3 were compared to more recent algorithms that rely on attention mechanisms, namely the Vision Transformer (ViT), trained in 818 UGIE images (409 normal and 409 IM). All the models were trained using a 5-fold cross-validation technique and for validation, an external dataset will be tested with 100 UGIE images (50 normal and 50 IM). In the end, explainability methods (Grad-CAM and attention rollout) were used for more clear and more interpretable results. The model which performed better was ResNet50 with a sensitivity of 0.75 (±0.05), an accuracy of 0.79 (±0.01), and a specificity of 0.82 (±0.04). This model obtained an AUC of 0.83 (±0.01), where the standard deviation was 0.01, which means that all iterations of the 5-fold cross-validation have a more significant agreement in classifying the samples than the other models. The ViT model showed promising performance, reaching similar results compared to the remaining models. © 2023, ICST Institute for Computer Sciences, Social Informatics and Telecommunications Engineering.

Abstract
The George B. Moody PhysioNet Challenge 2022 explored the detection of abnormal heart function from phonocardiogram (PCG) recordings. Although ultrasound imaging is becoming more common for investigating heart defects, the PCG still has the potential to assist with rapid and low-cost screening, and the automated annotation of PCG recordings has the potential to further improve access. Therefore, for this Challenge, we asked participants to design working, open-source algorithms that use PCG recordings to identify heart murmurs and clinical outcomes. This Challenge makes several innovations. First, we sourced 5272 PCG recordings from 1568 patients in Brazil, providing high-quality data for an underrepresented population. Second, we required the Challenge teams to submit working code for training and running their models, improving the reproducibility and reusability of the algorithms. Third, we devised a cost-based evaluation metric that reflects the costs of screening, treatment, and diagnostic errors, facilitating the development of more clinically relevant algorithms. A total of 87 teams submitted 779 algorithms during the Challenge. These algorithms represent a diversity of approaches from both academia and industry for detecting abnormal cardiac function from PCG recordings. © 2022 Creative Commons.

Abstract
The pericardium is a thin membrane sac that covers the heart. As such, the segmentation of the pericardium in computed tomography (CT) can have several clinical applications, namely as a preprocessing step for extraction of different clinical parameters. However, manual segmentation of the pericardium can be challenging, time-consuming and subject to observer variability, which has motivated the development of automatic pericardial segmentation methods. In this study, a method to automatically segment the pericardium in CT using a U-Net framework is proposed. Two datasets were used in this study: the publicly available Cardiac Fat dataset and a private dataset acquired at the hospital centre of Vila Nova de Gaia e Espinho (CHVNGE). The Cardiac Fat database was used for training with two different input sizes - 512 512 and 256 256. A superior performance was obtained with the 256 256 image size, with a mean Dice similarity score (DCS) of 0.871 ± 0.01 and 0.807 ± 0.06 on the Cardiac Fat test set and the CHVNGE dataset, respectively. Results show that reasonable performance can be achieved with a small number of patients for training and an off-the-shelf framework, with only a small decrease in performance in an external dataset. Nevertheless, additional data will increase the robustness of this approach for difficult cases and future approaches must focus on the integration of 3D information for a more accurate segmentation of the lower pericardium. © 2023, ICST Institute for Computer Sciences, Social Informatics and Telecommunications Engineering.