António Cunha

Abstract
In modern industrial environments, ensuring the quality of manufactured components is critical, particularly when dealing with reflective surfaces that hinder conventional inspection techniques. Although deep learning-based methods offer robust solutions for visual defect detection, their performance often hinges on the availability of substantial annotated datasets. In industrial scenarios, labeling such datasets is costly and time-consuming. This study investigates applying sample selection techniques to reduce annotation efforts for porosity detection on machined aluminium parts. Several selection strategies were evaluated using a real-world dataset composed of high-resolution images, including uncertainty, diversity, random-based criteria, and hybrid combinations. The best-performing strategy, which combined entropy-based uncertainty, spatial diversity, and random-based, achieved an F1-score of 86.70% and a recall of 82.99% after ten iterations using only 2,400 annotated images, corresponding to 66.67% of the active learning pool. Although the fully supervised model achieved an F1-score of 88.84% and a recall of 86.30%, the proposed approach proved a competitive alternative. These results demonstrate that selective data annotation can significantly reduce labeling effort while maintaining reliable performance in defect detection, even under the challenging conditions posed by reflective industrial parts.

Abstract
Early detection of gastrointestinal lesions such as intestinal metaplasia (IM), dysplasia, and polyps remains challenging due to their subtle appearance and the scarcity of well-annotated medical image datasets. To address this limitation, we introduce Cut Instance Mixing (CIM), a domain-specific data augmentation method designed to generate anatomically plausible lesion-containing images through the identification of biologically relevant regions of interest and seamless lesion blending using Poisson image editing and gradient-based mixing. CIM was evaluated across three distinct endoscopic datasets (IM, dysplasia, and polyps) using a ResNet50 classifier and five-fold cross-validation. The proposed method consistently outperformed state-of-the-art augmentation techniques. In IM classification, CIM with alpha = 0.8 achieved the highest performance (AUC: 0.879, Accuracy: 0.823), surpassing MixUp, CutMix and random copy-paste. In dysplasia detection, CIM reached near-perfect results (AUC: 0.997, Accuracy: 0.966), and demonstrated strong generalization on an external polyp dataset (AUC: 0.830, Accuracy: 0.769). Grad-CAM analyses further confirmed that CIM preserves clinically relevant features, improving model attention on lesion regions. These findings demonstrate that CIM enables the generation of realistic and biologically coherent synthetic samples, effectively mitigating data imbalance and enhancing classification robustness. The method is architecture-agnostic and broadly applicable to tasks requiring anatomically consistent augmentation, providing a promising direction for improving deep learning systems in gastrointestinal imaging.

Abstract
Disinformation is an ancient social phenomenon that has found a favourable environment for dissemination in internet-based social networks. While the scientific community seeks to address the problem by creating specific tools to detect and classify the various types of false information, we argue that systems thinking is necessary to understand and holistically address this major threat. The works that directly cite Disinformation Systems treat this term as a grouping of concepts, mechanisms, objectives and institutions in a large multidisciplinary repository that finds a self-explanation in the term systems. Through a qualitative and theoretical basis, this research proposes that the generation of disinformation can be defined as a system model, theorizing that the entire process of creating, producing and disseminating disinformation can be defined systematically. Thus, we define an initial descriptive model and affirm that the generation of disinformation can be characterized in terms of a sociotechnical work system. We tested the model in historical disinformation scenarios showing that it fits the components and flows of the system. Although initial, this work has the potential to enable the development of new systemic insights and research in the area of disinformation.

Abstract
Disinformation, although an ancient phenomenon, has gained unprecedented reach and speed with the rise of the internet and social media platforms. While traditional fact-checking approaches focus on the semantic content of information, this paper proposes a quantitative analysis based on metadata and formal textual features to investigate disinformation from a quality dimension perspective, assuming that false or misleading information often fails to meet informational quality criteria. Using an experimental approach, we analyzed two datasets of news from reliable and unreliable sources and applied statistical methods, including the Mann-Whitney U test, Cliff’s Delta, and Rosenthal’s r, to measure differences and effect size in the quality dimensions of accuracy, currency, readability, consistency and reliability. The results show that lexical cohesion and lexical diversity are the strongest discriminators of source reliability, followed by structural error rates, while currency and readability display only weak discriminative power. The proposed News Reliability Index (NRI) emerges as a moderate but complementary indicator. Overall, reliable sources consistently demonstrate higher information quality, but structural differences alone are insufficient to detect disinformation, especially considering the capacity of generative AI to produce syntactically coherent texts. We conclude that semantic content analysis remains essential for identifying disinformation, with structural features best applied as supporting signals in detection models. Finally, we highlight future challenges, such as the growing use of artificial intelligence in generating high-quality disinformation, which may reduce the effectiveness of structural metrics and complicate automation in verification processes.

Abstract
Diabetic retinopathy (DR) is a vision-threatening complication of diabetes and one of the leading causes of blindness worldwide. It is characterized by the appearance of lesions on the retina, such as microaneurysms, hemorrhages, hard exudates, and soft exudates, which are crucial for staging the disease. Diagnosis is typically performed through analysis of fundus images, a manual process that is time-consuming and prone to subjectivity. To address this, this study explores the automatic segmentation of DRrelated lesions using deep learning techniques. Four convolutional neural network architectures were evaluated: U-Net, FPN, DeepLabV3+, and Attention U-Net. The IDRiD dataset was used for training and validation The DeepLabV3+ model with ResNet50 achieved the highest overall performance, while FPN was the only model capable of detecting microaneurysms in the multiclass task. These findings underscore the importance of architecture selection, loss function design, and preprocessing choices. Future work may explore new datasets, enhanced data augmentation, and the impact of optic disc removal on segmentation accuracy. © 2025 The Authors. Published by Elsevier B.V.