Mestrado em Direcção Estratégica em Tecnologias de Informação

<a class="ep_document_link" href="/27825/1/s41598-026-39196-x_reference.pdf"><img class="ep_doc_icon" alt="[img]" src="/style/images/fileicons/text.png" border="0"/></a>

en

open

Benchmarking multiple instance learning architectures from patches to pathology for prostate cancer detection and grading using attention-based weak supervision

Histopathological evaluation is necessary for the diagnosis and grading of prostate cancer, which is still one of the most common cancers in men globally. Traditional evaluation is time-consuming, prone to inter-observer variability, and challenging to scale. The clinical usefulness of current AI systems is limited by the need for comprehensive pixel-level annotations. The objective of this research is to develop and evaluate a large-scale benchmarking study on a weakly supervised deep learning framework that minimizes the need for annotation and ensures interpretability for automated prostate cancer diagnosis and International Society of Urological Pathology (ISUP) grading using whole slide images (WSIs). This study rigorously tested six cutting-edge multiple instance learning (MIL) architectures (CLAM-MB, CLAM-SB, ILRA-MIL, AC-MIL, AMD-MIL, WiKG-MIL), three feature encoders (ResNet50, CTransPath, UNI2), and four patch extraction techniques (varying sizes and overlap) using the PANDA dataset (10,616 WSIs), yielding 72 experimental configurations. The methodology used distributed cloud computing to process over 31 million tissue patches, implementing advanced attention mechanisms to ensure clinical interpretability through Grad-CAM visualizations. The optimum configuration (UNI2 encoder with ILRA-MIL, 256 256 patches, 50% overlap) achieved 78.75% accuracy and 90.12% quadratic weighted kappa (QWK), outperforming traditional methods and approaching expert pathologist-level diagnostic capability. Overlapping smaller patches offered the best balance of spatial resolution and contextual information, while domain-specific foundation models performed noticeably better than generic encoders. This work is the first large-scale, comprehensive comparison of weekly supervised MIL methods for prostate cancer diagnosis and grading. The proposed approach has excellent clinical diagnostic performance, scalability, practical feasibility through cloud computing, and interpretability using visualization tools.

Producción Científica

Naveed Anwer Butt mail , Dilawaiz Sarwat mail , Irene Delgado Noya mail irene.delgado@uneatlantico.es, Kilian Tutusaus mail kilian.tutusaus@uneatlantico.es, Nagwan Abdel Samee mail , Imran Ashraf mail ,

Butt

<a href="/27915/1/csbj.0023.pdf" class="ep_document_link"><img class="ep_doc_icon" alt="[img]" src="/style/images/fileicons/text.png" border="0"/></a>

en

open

A Systematic Literature Review on Integrated Deep Learning and Multi-Agent Vision-Language Frameworks for Pathology Image Analysis and Report Generation

This systematic literature review (SLR) investigates the integration of deep learning (DL), vision-language models(VLMs), and multi-agent systems in the analysis of pathology images and automated report generation. The rapidadvancement of whole-slide imaging (WSI) technologies has posed new challenges in pathology, especially due to thescale and complexity of the data. DL techniques in general and convolutional neural networks (CNNs) and transform-ers in particular have significantly enhanced image analysis tasks including segmentation, classification, and detection.However, these models often lack generalizability to generate coherent, clinically relevant text, thus necessitating theintegration of VLMs and large language models (LLMs). This review examines the effectiveness of VLMs and LLMsin bridging the gap between visual data and clinical text, focusing on their potential for automating the generationof pathology reports. Additionally, multi-agent systems, which leverage specialized artificial intelligence (AI) agentsto collaboratively perform diagnostic tasks, are explored for their contributions to improving diagnostic accuracy andscalability. Through a synthesis of recent studies, this review highlights the successes, challenges, and future direc-tions of these AI technologies in pathology diagnostics, offering a comprehensive foundation for the development ofintegrated, AI-driven diagnostic workflows.

Producción Científica

Usama Ali mail , Imran Shafi mail , Jamil Ahmad mail , Arlette Zárate Cáceres mail , Thania Chio Montero mail , Hafiz Muhammad Raza ur Rehman mail , Imran Ashraf mail ,

Ali

<a href="/27970/1/s11357-026-02188-w.pdf" class="ep_document_link"><img class="ep_doc_icon" alt="[img]" src="/style/images/fileicons/text.png" border="0"/></a>

en

open

Fish consumption and cognitive function in aging: a systematic review of observational studies

Epidemiological studies consistently link higher fish intake with slower rates of cognitive decline and lower dementia incidence. The aim of the present study was to systematically review existing observational studies investigating the association between fish consumption and cognitive function in older adults. A total of 25 studies (8 cross-sectional and 17 prospective including mainly healthy older adults, age range of participants ranging from 18 to 30 years at baseline in prospective studies to 65 to 91 years, representing the upper limit of the age spectrum) were reviewed. Cognitive functions currently investigated in most published studies included various domains, such as global cognition, memory (episodic, working), executive function (planning, inhibition, flexibility), attention and processing speed. Existing studies greatly vary in terms of design (cross-sectional and prospective), geographical area, number of participants involved, and tools used to assess the outcomes of interest. The main findings across studies are not univocal, with some studies reporting stronger evidence of association between fish consumption and various cognitive domains, while others addressed rather null findings. The most consistently responsive domains were processing speed, executive functioning, semantic memory, and global cognitive ability among individuals consuming fish at least weekly, which are highly relevant to both neurodegenerative and vascular forms of cognitive impairment. Positive associations were also observed for verbal memory and general memory, though these were less uniform and often attenuated after multivariable adjustment. In contrast, associations with reaction time, verbal-numerical reasoning, and broad composite scores were inconsistent, and several fully adjusted models showed null results. In conclusion, the evidence suggests that regular fish intake (typically ≥1–2 servings per week) is linked to preserved cognitive performance, although some inconsistent findings require further investigations.

Producción Científica

Justyna Godos mail , Giuseppe Caruso mail , Agnieszka Micek mail , Alberto Dolci mail , Carmen Lilí Rodríguez Velasco mail carmen.rodriguez@uneatlantico.es, Evelyn Frias-Toral mail , Jason Di Giorgio mail , Nicola Veronese mail , Andrea Lehoczki mail , Mario Siervo mail , Zoltan Ungvari mail , Giuseppe Grosso mail ,

Godos

<a class="ep_document_link" href="/27554/1/s41598-026-37541-8_reference.pdf"><img class="ep_doc_icon" alt="[img]" src="/style/images/fileicons/text.png" border="0"/></a>

en

open

A scalable and secure federated learning authentication scheme for IoT

Secure and scalable authentication remains a fundamental challenge in Internet of Things (IoT) networks due to constrained device resources, dynamic topology, and the absence of centralized trust infrastructures. Conventional password-based and certificate-driven authentication schemes incur high computation, storage, and communication overhead, limiting their suitability for large-scale deployments. To address these limitations, this paper proposes ScLBS, a federated learning (FL)–based self-certified authentication scheme for distributed and sustainable IoT environments. ScLBS integrates self-certified public key cryptography with FL-driven trust adaptation, enabling decentralized public key derivation without reliance on third-party certificate authorities or exposure of private credentials. A zero-knowledge mechanism combined with location-aware authentication strengthens resistance to impersonation, Sybil, and replay attacks. Hierarchical key management supported by a -tree enables efficient group rekeying and preserves forward and backward secrecy under dynamic membership. Formal security verification is conducted under the Dolev–Yao adversary model using ProVerif, confirming secrecy of private and session keys (SKs) and correctness of authentication. Extensive NS-3 simulations and ablation analysis demonstrate that ScLBS achieves lower authentication delay, reduced message overhead, improved network utilization, and decreased energy consumption compared to representative IoT authentication schemes, while maintaining bounded FL overhead. These results indicate that ScLBS provides a balanced trade-off between security strength, scalability, and resource efficiency for constrained IoT networks.

Producción Científica

Premkumar Chithaluru mail , B. Veera Jyothi mail , Fahd S. Alharithi mail , Wojciech Ksiazek mail , M. Ramchander mail , Aman Singh mail aman.singh@uneatlantico.es, Ravi Kumar Rachavaram mail ,

Chithaluru

<a href="/27968/1/sensors-26-01516-v2.pdf" class="ep_document_link"><img class="ep_doc_icon" alt="[img]" src="/style/images/fileicons/text.png" border="0"/></a>

en

open

Human Activity Recognition in Domestic Settings Based on Optical Techniques and Ensemble Models

Human activity recognition (HAR) is essential in many applications, such as smart homes, assisted living, healthcare monitoring, rehabilitation, physiotherapy, and geriatric care. Conventional methods of HAR use wearable sensors, e.g., acceleration sensors and gyroscopes. However, they are limited by issues such as sensitivity to position, user inconvenience, and potential health risks with long-term use. Optical camera systems that are vision-based provide an alternative that is not intrusive; however, they are susceptible to variations in lighting, intrusions, and privacy issues. The paper uses an optical method of recognizing human domestic activities based on pose estimation and deep learning ensemble models. The skeletal keypoint features proposed in the current methodology are extracted from video data using PoseNet to generate a privacy-preserving representation that captures key motion dynamics without being sensitive to changes in appearance. A total of 30 subjects (15 male and 15 female) were sampled across 2734 activity samples, including nine daily domestic activities. There were six deep learning architectures, namely, the Transformer (Transformer), Long Short-Term Memory (LSTM), Gated Recurrent Unit (GRU), Multilayer Perceptron (MLP), One-Dimensional Convolutional Neural Network (1D CNN), and a hybrid Convolutional Neural Network–Long Short-Term Memory (CNN–LSTM) architecture. The results on the hold-out test set show that the CNN–LSTM architecture achieves an accuracy of 98.78% within our experimental setting. Leave-One-Subject-Out cross-validation further confirms robust generalization across unseen individuals, with CNN–LSTM achieving a mean accuracy of 97.21% ± 1.84% across 30 subjects. The results demonstrate that vision-based pose estimation with deep learning is a useful, precise, and non-intrusive approach to HAR in smart healthcare and home automation systems.

Producción Científica

Muhammad Amjad Raza mail , Nasir Mehmood mail , Hafeez Ur Rehman Siddiqui mail , Adil Ali Saleem mail , Roberto Marcelo Álvarez mail roberto.alvarez@uneatlantico.es, Yini Airet Miró Vera mail yini.miro@uneatlantico.es, Isabel de la Torre Díez mail ,

Raza