REPOSITÓRIO P.PORTO
Repositório Científico do Politécnico do Porto
Entradas recentes
An explainable and privacy-preserving machine learning pipeline for early detection of endometriosis leveraging liquid biopsy and minimally-invasive C
Publication . MANESSE, CIRO MIGUEL POÇAS FERREIRA; Martinho, Diogo Emanuel Pereira; Conceição, Luís Manuel Silva
Endometriosis a!ects approximately one in ten women of reproductive age, yet it is diagnosed,
on average, seven to eight years after the onset of symptoms, and in some studies
up to twelve. This diagnostic delay is associated with prolonged symptoms, uncertainty,
repeated healthcare contacts and delayed therapeutic intervention, largely because a definitive
diagnosis still depends on an invasive surgical procedure, namely laparoscopy. Recent
progress in Machine Learning, together with the growing availability of clinical and molecular
data, o!ers a realistic opportunity to shorten this interval. This dissertation investigates
whether a non-invasive, explainable and privacy-preserving Machine Learning pipeline can
support an earlier clinical suspicion of Endometriosis from micro-RNA (miRNA) measured
in a blood sample, that is, a liquid biopsy. Three requirements are addressed: the test must
be non-invasive; its decisions must be explainable, so that a clinician can examine and verify
them; and the training procedure must protect sensitive patient data. To the best of the author’s
knowledge, no previous work brings these three properties together for Endometriosis
detection.
Working exclusively with real, public data, a leakage-safe pipeline was built for the serummiRNA
cohort GSE279435 (127 samples; 67 Endometriosis, 60 benign controls). A central
observation in understanding the data is that the measurements are left-censored at the
assay’s limit of detection, so that a missing value is itself informative. Additionally, every
preprocessing step was fitted strictly inside each cross-validation fold to avoid optimistic
bias.
Five classifiers were compared under nested, repeated, stratified cross-validation. The two
strongest (Random Forest and LightGBM) reached an area under the ROC curve of 0.77–
0.78 with balanced sensitivity and specificity, a performance comparable to that of the
previously published eleven-miRNA model evaluated on the same cohort. Explainability was
provided with SHAP, which makes each prediction inspectable both globally and at the level
of an individual patient, allowing the model to be examined rather than trusted blindly. Five of
the ten most influential miRNAs coincide with the published diagnostic panel, which grounds
the model’s reasoning in established biology; the remaining five are plausible additional
candidates, consistent with the di!erential-expression analysis, that would warrant follow-up
in larger studies.
On privacy, this is, to the best of the author’s knowledge, the first work to apply both
Di!erential Privacy and Federated Learning to non-invasive miRNA-based Endometriosis
detection. In a simulated federated setting, created by partitioning the public cohort into
four virtual sites, a Federated Learning implementation built with NVIDIA FLARE trained a
shared model without moving any raw data and recovered approximately 95% of the accuracy
of a model trained on the pooled data, well above what any single site achieved in isolation.
Di!erential Privacy, in contrast, reduced accuracy to little above chance at the privacy levels
that would meaningfully protect patients, illustrating how costly strong, sample-level privacy
guarantees become when the cohort is this small (n = 127). A cross-platform exploratory
test on an independent plasma cohort indicated that the general signal — that circulating miRNA carries an Endometriosis-related signal — holds across biofluids, even though the
specific serum signature does not transfer directly to plasma.
The contribution of this work is therefore not a higher headline accuracy but a pipeline
that, using real public data, reaches the performance of the previously published model while
adding two properties that model did not have: explanations a clinician can recognise, and
training that never centralises patient data. The proposed pipeline is presented not as a
medical device, but as a reproducible and ethically framed foundation for the larger, multiinstitutional
validation studies needed to support earlier clinical suspicion of Endometriosis.
Design e implementação de um conector seguro e escalável compatível com MCP para acesso de agentes de IA a dados de Business Intelligence
Publication . GRAÇA, DANIEL ALEXANDRE RIBEIRO; Duarte, Fernando Jorge Ferreira
The increasing adoption of Large Language Models (LLMs) and agent-based systems in enterprise
environments has intensified the need for secure, scalable, and standardized mechanisms
that allow Artificial Intelligence (AI) to interact with structured business data. In
the retail domain, Business Intelligence (BI) systems play a central role in supporting operational
and strategic decision-making, yet their integration with autonomous AI agents
remains fragmented and largely ad hoc.
This dissertation addresses this gap by designing, implementing, and evaluating a secure
and scalable integration layer that enables AI agents within Watson (the AI decision-support
platform developed by Omnium:retail) to access enterprise data through declarative tools
and governed enterprise services. The solution is MCP-compatible in contract and architecture:
enterprise capabilities are exposed with JSON Schema contracts and validated before
execution, following the declarative tool-integration model promoted by the Model Context
Protocol (MCP). Watson does not implement an MCP server or client; interoperability is
achieved through an internal registry-based connector and OpenAI function calling rather
than MCP transport.
The work combines a systematic literature review (PICOCS and PRISMA) with architectural
modeling and an empirical consistency study. Across 390 independent executions covering
menu navigation, stock lookup, and product-information scenarios, 98.2% of responses were
semantically correct and grounded in tool outputs. Mean end-to-end latency (∼9 s), however,
remained well above the two-to-three-second interactive target envisaged at project
ideation, reflecting the cost of multi-stage LLM orchestration in a planner–executor pipeline.
The state-of-the-art analysis and evaluation demonstrate that an MCP-compatible integration
layer can support reliable agent access to enterprise data in a multi-tenant setting when
complemented with appropriate architectural patterns, access control, and action validation.
This work contributes to the practical understanding of MCP-aligned integration in
enterprise contexts and provides an empirically grounded foundation for future optimization
within production retail environments.
Dimensionamento de um reator de leito fluidizado borbulhante para produção de bio-óleo a partir da biomassa
Publication . FONSECA, DANIEL CARLOS; Ribeiro, Albina Maria de Sá; Pimenta, Maria Paula Neto
Esta dissertação apresenta o dimensionamento de um reator de leito fluidizado borbulhante
para a produção de bio-óleo a partir da pirólise rápida de biomassa de eucalipto. A motivação
principal do trabalho insere-se no contexto da transição energética e da crescente necessidade
de desenvolver tecnologias de conversão de biomassa capazes de produzir vetores energéticos
renováveis com potencial para substituir, parcial ou totalmente, os combustíveis fósseis em
aplicações de produção de calor e geração de energia.
Num primeiro momento, procedeu-se à caracterização da biomassa, com particular enfoque no
eucalipto, pela sua elevada disponibilidade em Portugal, produtividade florestal e ampla
documentação na literatura. A biomassa é caraterizada em termos da sua composição
elementar, da análise próxima, e das suas principais propriedades termoquímicas relevantes
para processos de pirólise. Em seguida, foi efetuada uma revisão do estado da arte sobre a
pirólise rápida de biomassa, os tipos de reatores mais utilizados e os fatores operacionais que
influenciam os rendimentos em bio-óleo, carbonizado e gás. A parte central do trabalho é
dedicada ao estudo da fluidização e ao dimensionamento de um reator de leito fluidizado
borbulhante. Para tal foram aplicadas correlações empíricas com o objetivo de definir as
condições de operação, nomeadamente a velocidade superficial do gás e de dimensionar os
principais componentes do reator, incluindo a altura da zona do leito e o distribuidor de gás.
Com base em balanços de massa e de energia, e depois de definidas todas as correntes do
processo foi determinada a potência térmica requerida e definiu-se o aproveitamento
energético dos produtos da pirólise.
Os resultados obtidos indicaram que, para as condições de operação de 500 °C e uma
alimentação de 2 kg/h de biomassa seca é necessário um caudal de azoto de 1,13 L/s pré
aquecido a 500 °C. O diâmetro do reator foi fixado em 76 mm e tendo-se obtido uma altura da
zona do leito entre 157 mm e 925 mm fixou-se para altura do reator 500 mm. O distribuidor foi
dimensionado com 13 orifícios. De acordo com o balanço de energia obteve-se uma potência
térmica de 2 kW, para manter as condições de operação, sendo também necessário 0,3 kW para
pré aquecer o azoto.
O estudo demonstra que o dimensionamento proposto é tecnicamente viável e que o reator de
leito fluidizado borbulhante constitui uma solução com potencial para a valorização energética
da biomassa de eucalipto, contribuindo para a diversificação das fontes de energia renovável e
para a redução das emissões associadas ao uso de combustíveis fósseis.
AI-powered framework for automated hardware verification
Publication . SACRAMENTO, DANIEL DA FONSECA; Gericota, Manuel Gradim de Oliveira
Modern eBike systems are increasingly complex, integrating multiple subsystems
that require rigorous verification against formal requirements. This thesis presents
an end-to-end automated test framework that bridges requirements management
(Jama Connect) directly to physical test bench execution through an Artificial Intelligence
(AI)-powered translation layer.
The proposed architecture introduces a three-layer data model: HW_MAP,
TEST_SPEC, and TEST_SEQUENCE. It creates a traceable digital thread from requirements
to measurements. A deterministic AI agent, the JamaTestTranslator, translates
natural-language requirements into executable test sequences without human
interpretation, enforcing a zero-hallucination policy through constrained prompting
and JSON schema validation. A custom multiplexing hardware interface extends
oscilloscope channel availability for multi-signal testing without additional instrumentation.
The framework is validated on two representative test cases: an ORing protection
circuit characterization, which verifies the output voltage levels of a dual-diode
supply arbitration network, and an SPI signal integrity test, which verifies the voltage
levels, rise and fall times, and clock frequency of the communication bus between
a barometer, an Inertial Measurement Unit (IMU), and a microcontroller. In both
cases, the framework substantially reduces the manual effort involved in test script
authoring and instrument configuration, automating the translation from structured
requirements to executable test sequences. The generated reports present pass/fail
verdicts and min/max/measured comparisons directly traceable to the source requirements
in Jama Connect.
Projeto de estabilidade de um edifício em betão armado
Publication . TEIXEIRA, ANDRÉ FILIPE SILVA; Teles, Isabel Maria Alvim
Este documento constitui o relatório final da unidade curricular de Projeto, detalhando o
desenvolvimento do projeto de estabilidade de um edifício em betão armado. O documento
descreve as várias fases do processo: desde a conceção da solução estrutural e prédimensionamento
até à modelação computacional, dimensionamento final e elaboração das
peças escritas e desenhadas. Para a execução deste trabalho, mobilizaram-se competências
transversais adquiridas no percurso académico — com particular enfoque na área das
estruturas — aliando o uso de ferramentas de cálculo automático (Robot Structural Analysis)
aos métodos de verificação analítica tradicionais.