João Manuel Pedrosa

Abstract
The link between epicardial adipose tissue (EAT) and cardiovascular risk is well established, with EAT volume being strongly associated with inflammation, coronary artery disease (CAD) risk, and mortality. However, its EAT quantification is hindered by the time-consuming nature of manual EAT segmentation in cardiac computed tomography (CT). 300 non-contrast cardiac CT scans were collected and the pericardium was manually delineated. In a subset of this data (N = 30), manual delineation was repeated by the same operator and by a second operator. Two automatic methods were then used for pericardial segmentation: a commercially available tool, Siemens Cardiac Risk Assessment (CRA) software; and a deep learning solution based on a U-Net architecture trained exclusively with external public datasets (CardiacFat and OSIC). EAT segmentations were obtained through thresholding to [- 150,- 50] Hounsfield units. Pericardial and EAT segmentation performance was evaluated considering the segmentations by the first operator as reference. Statistical significance of differences for all metrics and segmentation methods was tested through Student t-tests. Pericardial segmentation intra-/interobserver variability was excellent, with the U-Net outperforming Siemens CRA (p < 0.0001). The intra- and interobserver agreement for EAT segmentation was lower with Dice Scores (DSC) of 0.862 and 0.775 respectively, while the U-Net and Siemens CRA obtained DSCs of 0.723 and 0.679 respectively. EAT volume quantification showed that the agreement between a human observer and the U-Net was better than that of two human observers (p = 0.0141), with a Pearson Correlation Coefficient (PCC) of 0.896 and a bias of - 2.83 cm(3) (below the interobserver bias of 9.05 cm3). The lower performances of EAT segmentation highlight the difficulty in segmenting this structure. For both pericardial and EAT segmentation, the deep learning method outperformed the commercial solution. While the segmentation performance of the U-Net solution was below interobserver variability, EAT volume quantification performance was competitive with human readers, motivating future use of these tools. Clinical trial number: NCT03280433, registered retrospectively on 2017-09-08.

Abstract
Cardiovascular diseases are one of the main causes of death in the world. The predominant form of cardiovascular disease is coronary artery disease. Coronary artery calcium scanning is a non-contrast computed tomography exam that is considered the most reliable predictor of coronary events. Deep learning models have been developed for the segmentation of coronary artery calcium but the results have limited interpretability due to the black-box nature of these models. This work proposes an image retrieval pipeline based on a supervised contrastive framework that is capable of enhancing this interpretability by providing similar visual examples of coronary calcifications. In the COCA dataset, it is shown that this retrieval presents a label precision of 0.944 ± 0.230 regarding artery labels of retrieved images, with moderate similarity in terms of calcification area and Agatston score. It is also shown that the retrieval can be used to correct a deep CAC segmentation model by passing predictions from a segmentation model through the retrieval system, improving robustness and explainability.Clinical relevance- This study enhances CAC segmentation through image retrieval, improving both explainability and artery-specific labeling. By providing clinicians with more interpretable and anatomically accurate results, our approach aims to increase confidence in AI-assisted diagnostics leading to better-informed clinical decision-making in coronary artery disease diagnosis.

Abstract
Cardiovascular diseases are the leading cause of death in the world, with coronary artery disease being the most prevalent. Coronary artery calcifications are critical biomarkers for cardiovascular disease, and their quantification via non-contrast computed tomography is a widely accepted and heavily employed technique for risk assessment. Manual segmentation of these calcifications is a time-consuming task, subject to variability. State-of-the-art methods often employ convolutional neural networks for an automated approach. However, there is a lack of studies that perform these segmentations with 3D architectures that can gather important and necessary anatomical context to distinguish the different coronary arteries. This paper proposes a novel and automated approach that uses a lightweight three-dimensional convolutional neural network to perform efficient and accurate segmentations and calcium scoring. Results show that this method achieves Dice score coefficients of 0.93 ± 0.02, 0.93 ± 0.03, 0.84 ± 0.02, 0.63 ± 0.06 and 0.89 ± 0.03 for the foreground, left anterior descending artery (LAD), left circumflex artery (LCX), left main artery (LM) and right coronary artery (RCA) calcifications, respectively, outperforming other state-of-the-art architectures. An external cohort validation also showed the generalization of this method's performance and how it can be applied in different clinical scenarios. In conclusion, the proposed lightweight 3D convolutional neural network demonstrates high efficiency and accuracy, outperforming state-of-the-art methods and showcasing robust generalization potential.

Abstract
The increase in complexity of deep learning models demands explanations that can be obtained with methods like Grad-CAM. This method computes an importance map for the last convolutional layer relative to a specific class, which is then upsampled to match the size of the input. However, this final step assumes that there is a spatial correspondence between the last feature map and the input, which may not be the case. We hypothesize that, for models with large receptive fields, the feature spatial organization is not kept during the forward pass, which may render the explanations devoid of meaning. To test this hypothesis, common architectures were applied to a medical scenario on the public VinDr-CXR dataset, to a subset of ImageNet and to datasets derived from MNIST. The results show a significant dispersion of the spatial information, which goes against the assumption of Grad-CAM, and that explainability maps are affected by this dispersion. Furthermore, we discuss several other caveats regarding Grad-CAM, such as feature map rectification, empty maps and the impact of global average pooling or flatten layers. Altogether, this work addresses some key limitations of Grad-CAM which may go unnoticed for common users, taking one step further in the pursuit for more reliable explainability methods.

Abstract
Arterial stiffness (AS) is a well-established predictor of cardiovascular events, including myocardial infarction and stroke. One of the most recognized methods for assessing AS is through arterial pulse wave velocity (aPWV), which provides valuable clinical insights into vascular health. However, its measurement typically requires specialized equipment, making it inaccessible in primary healthcare centers and low-resource settings. In this study, we developed and validated different machine learning models to estimate aPWV using common clinical markers routinely collected in standard medical examinations. Thus, we trained five regression models: Linear Regression, Polynomial Regression (PR), Gradient Boosting Regression, Support Vector Regression, and Neural Networks (NNs) on the EVasCu dataset, a cohort of apparently healthy individuals. A 10-fold cross-validation demonstrated that PR and NN achieved the highest predictive performance, effectively capturing nonlinear relationships in the data. External validation on two independent datasets, VascuNET (a healthy population) and ExIC-FEp (a cohort of cardiopathic patients), confirmed the robustness of PR and NN (R- (2)> 0.90) across different vascular conditions. These results indicate that by using easily accessible clinical variables and AI-driven insights, it is possible to develop a cost-effective tool for aPWV estimation, enabling early cardiovascular risk stratification in underserved and rural areas where specialized AS measurement devices are unavailable.