Elsa Ferreira Gomes

Abstract

Abstract
Data leakage is a critical issue in deep learning inflating performance and compromise validity, especially in sensitive areas like medical imaging. This study systematically evaluates two common leakage types in oral squamous cell carcinoma classification from histopathology images: (1) preprocessing leakage (global normalization before dataset splitting) and (2) a severe sample-related (patient-related) contamination scenario created by mixing closely related original and augmented images across splits. We trained 11 CNN and Transformer-based models on a public oral cancer histopathology dataset, benchmarking results against published leakage-free baselines. The results obtained show that the configuration with random splitting of original and augmented images (Scenario 2) artificially increased accuracy by up to 18% (mean +14.3%) compared to leakage-free conditions, while the preprocessing-based leakage (Scenario 1) showed smaller deviations (+1.8%). These inflated metrics arise from a combination of cross-split contamination between closely related samples and increased dataset redundancy, rather than genuine gains in generalization ability. Transformers improved leak-free accuracy (+3.9%) but degraded performance in Scenario 2 (-1.4%), revealing sensitivity to sample-specific biases. The observed performance gains under data leakage conditions are methodological artifacts that undermine clinical reliability, with a severe sample-related contamination scenario (Scenario 2) with random splitting of original and augmented images being particularly detrimental due to its promotion of non-generalizable feature learning. The quantitative benchmarks established here-including a mean accuracy gap of 12.5% (Scenario 2 vs. Scenario 1) across 11 models and Transformer architectures’ sensitivity to contamination-reveal fundamental tradeoffs between metric inflation and model trustworthiness. These findings establish quantitative benchmarks for leakage impacts in medical imaging and inform future guidelines for trustworthy AI development in pathology. © The Author(s) 2026.

Abstract
The Informatics Engineering degree at ISEP, aligned with international standards, was the first undergraduate degree in Portugal to be certified with EUR-ACE®. The programme emphasizes project-based learning, in which students, working in teams, develop interdisciplinary projects applying knowledge from all courses in each semester. A specific laboratory-project course coordinates an integrative project that aims to address complex problems. In the 2nd semester, two computer engineering courses (object-oriented programming and software engineering), and two mathematics courses (discrete mathematics and statistics) are involved, besides the laboratory/project course. This paper focuses on the integration of mathematics with informatics courses in this project, addressing real-world-like problems, bridging software engineering with mathematical topics. To assess the adopted PBL, enquiries were carried out among students. This approach fosters active learning and reinforces the relevance of mathematics within engineering, preparing students for job market demands. © 2026, IGI Global Scientific Publishing. All rights reserved.

Abstract
Sound-based uroflowmetry (SU) is a non-invasive technique emerging as an alternative to traditional uroflowmetry (UF) to calculate the voiding flow rate based on the sound generated by the urine impacting the water in a toilet, enabling remote monitoring and reducing the patient burden and clinical costs. This study trains four different machine learning (ML) models (random forest, gradient boosting, support vector machine and convolutional neural network) using both regression and classification approaches to predict and categorize the voiding flow rate from sound events. The models were trained with a dataset that contains sounds from synthetic void events generated with a high precision peristaltic pump and a traditional toilet. Sound was simultaneously recorded with three devices: Ultramic384k, Mi A1 smartphone and Oppo Smartwatch. To extract the audio features, our analysis showed that segmenting the audio signals into 1000 ms segments with frequencies up to 16 kHz provided the best results. Results show that random forest achieved the best performance in both regression and classification tasks, with a mean absolute error (MAE) of 0.9, 0.7 and 0.9 ml/s and quadratic weighted kappa (QWK) of 0.99, 1.0 and 1.0 for the three devices. To evaluate the models in a real environment and assess the effectiveness of training with synthetic data, the best-performing models were retrained and validated using a real voiding sounds dataset. The results reported an MAE below 2.5 ml/s and a QWK above 0.86 for regression and classification tasks, respectively.

Abstract