Abstract
Prognosis after intracranial hemorrhage (ICH) is influenced by a complex interplay between imaging and tabular data. Rapid and reliable prognosis are crucial for effective patient stratification and informed treatment decision-making. In this study, we aim to enhance image-based prognosis by learning a robust feature representation shared between prognosis and the clinical and demographic variables most highly correlated with it. Our approach mimics clinical decision-making by reinforcing the model to learn valuable prognostic data embedded in the image. We propose a 3D multi-task image model to predict prognosis, Glasgow Coma Scale and age, improving accuracy and interpretability. Our method outperforms current state-of-the-art baseline image models, and demonstrates superior performance in ICH prognosis compared to four board-certified neuroradiologists using only CT scans as input. We further validate our model with interpretability saliency maps. Code is available at https://github.com/MiriamCobo/MultitaskLearning_ICH_Prognosis.git.

Abstract

Abstract
Caves and lava tubes offer ideal environments for testing and improving methodological approaches as natural space analogs and living laboratories. These underground environments hold natural records that help us understand the evolution of our planet. This research reflects on the relevance of lava tubes and caves as simulation sites for extraterrestrial exploration. This study will focus on the methodological approach used in Lanzarote (Canary Islands, Spain) and Selvagens Islands (Madeira, Portugal), as two space analog sites associated with astrobiology projects that demonstrated good practice and reliable science and can inspire other space-related programs. Finally, the lava tube system on Terceira Island (Azores) is presented for the first time in Portugal as a promising new experimental site for geoengineering research and space analog activities. The multisectoral and longitudinal investigations related to a geoengineering approach and the 5Gs project will leverage the unique geodiversity and biodiversity of Natal Cave. Lava tube habitats could ultimately enable the establishment of a sustainable human presence on the Moon or Mars. © The Author(s), under exclusive license to Springer Nature Switzerland AG 2025.

Abstract
Background: Aberrant or loss of cell adhesion drives invasion and metastasis, key hallmarks of cancer progression. In this work, we hypothesized that a gene signature related to cell adhesion could predict breast cancer prognosis. Methods: Highly variant genes were tested for association with overall survival using Cox regression. Adhesion-related genes were identified through gene ontology analysis and multivariate Cox regression, with AIC selection, defined the prognostic signature. The AdhesionScore was then calculated as a weighted sum of gene expression, with risk stratification assessed by Kaplan-Meier and log-rank tests. Results: We found that the AdhesionScore was a significant independent predictor of poor survival in three large independent datasets, as it provided a robust stratification of patient prognosis in the Molecular Taxonomy of Breast Cancer International Consortium (METABRIC) (HR: 2.65; 95% CI: 2.33-3.0, p = 2.34 x 10-51), The Cancer Genome Atlas (TCGA) (HR: 3.46; 95% CI: 2.35-5.09, p = 3.50 x 10-10), and the GSE96058 (HR: 2.83; 95% CI: 2.20-3.65, p = 6.29 x 10-16) datasets. The 5-year risk of death in the high-risk group was 32.41% for METABRIC, 27.8% for TCGA, and 17.54% for GSE96058 datasets. Consistently, HER2-enriched and triple-negative breast carcinomas (TNBC) cases showed higher AdhesionScores than luminal subtypes, indicating an association with aggressive tumor biology. Conclusions: We have developed, for the first time, a molecular signature based on cell adhesion, as well as an associated AdhesionScore that can predict patient prognosis in invasive breast cancer, with potential clinical application. We developed a novel adhesion-based molecular signature, the AdhesionScore, that robustly predicts prognosis in breast cancer across independent cohorts, highlighting its potential clinical utility for patient risk stratification.

Abstract
The increasing integration of renewable energy sources and decentralized generation requires demand-side flexibility to improve grid stability and balance local energy flows. Local Flexibility Markets (LFMs) provide a framework for optimizing flexibility transactions within energy communities. This paper presents a model for quantifying and pricing residential resources flexibility, enabling prosumers to submit bids in an LFM managed by the Community Manager. The methodology relies on a linear optimization problem, where a Home Energy Management System first determines optimal consumption baselines. Then an iterative sensitivity analysis estimates upward, and downward flexibility bands and sets offer prices per resource. The market operates as two asymmetric voluntary pools, clearing flexibility offers and requests. Results show that Battery Energy Storage Systems and Electric Vehicles provide the most effective flexibility, significantly reducing energy costs. Future research should improve pricing mechanisms and scalability to support LFM adoption in different residential settings.

Abstract
Neonatal seizures represent a critical medical issue that requires prompt diagnosis and treatment. Typically, at-risk newborns undergo a Magnetic Resonance Imaging (MRI) brain assessment followed by continuous seizure monitoring using multichannel EEG. Visual analysis of multichannel electroencephalogram (EEG) recordings remains the standard modality for seizure detection; however, it is limited by fatigue and delayed seizure identification. Advances in machine and deep learning have led to the development of powerful neonatal seizure detection algorithms that may help address these limitations. Nevertheless, their performance remains relatively low and often disregards the non-stationary attributes of EEG signals, especially when learned from weakly labeled EEG data. In this context, the present paper proposes a novel deep-learning approach for neonatal seizure detection. The method employs rigorous preprocessing to reduce noise and artifacts, along with a recently developed time-frequency distribution (TFD) derived from a separable compact support kernel to capture the fast spectral changes associated with neonatal seizures. The high-resolution TFD diagrams are then converted into RGB images and used as inputs to a pre-trained ResNet-18 model. This is followed by the training of an attention-based multiple-instance learning (MIL) mechanism. The purpose is to perform a spatial time-frequency analysis that can highlight which channels exhibit seizure activity, thereby reducing the time required for secondary evaluation by a doctor. Additionally, per-instance learning (PIL) is performed to further validate the robustness of our TFD and methodology. Tested on the Helsinki public dataset, the PIL model achieved an area under the curve (AUC) of 96.8%, while the MIL model attained an average AUC of 94.1%, surpassing similar attention-based methods.