https://repositorio.ufu.br/handle/123456789/18919 2026-08-24T05:28:32Z 2026-08-24T05:28:32Z https://repositorio.ufu.br/handle/123456789/49646 2026-08-19T06:23:06Z 2026-07-21T00:00:00Z

Title: SQL para ciência de dados aplicada em dados de saúde pública no Brasil Abstract: Although traditional Data Science pipelines rely on exporting large volumes of information to external imperative environments, relational Database Management Systems (DBMSs) have evolved to support complex analytical logic. The In-database analytics paradigm emerges as a solution to eliminate data movement and serialization costs across the network by leveraging native logical optimizers. This work aims to demonstrate the viability of executing a unified analytical lifecycle using exclusively native SQL commands and analytical resources of PostgreSQL. To empirically validate the language’s robustness and expressiveness, a case study was implemented on dengue notifications in Brazil in 2025 (SINAN), integrating them with demographic indicators (IBGE) and occupational classification (CBO). The work methodology comprised designing a resilient architecture divided into a fault-tolerant Staging Area, key cleansing, Third Normal Form (3FN) normalization, and the transposition of sparse matrices using the CROSS JOIN LATERAL operator. The data infrastructure was accelerated through physical indexing (B-Tree) and consolidated into a Materialized View. The results prove the practical efficiency of the relational ecosystem, which processed a volume exceeding 1.65 million records with response times on the order of one second, computing incidence rates, cumulative time series via Window Functions, age attribute engineering, and Pearson statistical correlations. The work contributes to demystifying the use of the relational DBMS as a passive repository, presenting a reproducible and performant roadmap focused on Data Science purely structured in SQL.

2026-07-21T00:00:00Z https://repositorio.ufu.br/handle/123456789/49570 2026-08-15T06:27:52Z 2026-08-05T00:00:00Z

Title: Classificação de lesões cutâneas em imagens infravermelhas com redes neurais convolucionais Abstract: This work investigates the use of convolutional neural networks for the classification of benign and malignant skin lesions from dynamic infrared thermography images, aiming to support skin cancer diagnosis. Based on the hypothesis that the discriminative signal is associated with the lesion's reheating dynamics rather than with an isolated static frame, each thermal sequence was treated as the primary unit of evaluation. Characterized as applied experimental research, the study used biopsy-confirmed labels, five-fold patientgrouped cross-validation and, as the main metric, the pooled AUC with a 95% confidence interval obtained by stratified bootstrap resampling (also verified with a patient-grouped bootstrap). A custom convolutional network trained from scratch, transfer learning (ResNet18) and classical models (SVM and Random Forest) built on representations extracted from the fine-tuned ResNet18 together with thermal-dynamics features were compared. The best models reached an AUC of approximately 0.66, with no detectable difference from the Random Forest under paired bootstrap analysis, whereas the network trained from scratch showed performance compatible with chance level. None of the tested interventions (multichannel temporal input, feature enrichment, more expressive pooling and model ensembling) surpassed this ceiling. The results suggest that sample size is one of the main constraints on this dataset, although acquisition, radiometric representation and temporal modeling limitations may also have contributed to the observed performance ceiling. The work contributes a reproducible pipeline and a careful assessment of the feasibility of dynamic thermography combined with machine learning as a complementary tool for dermatological diagnosis.

2026-08-05T00:00:00Z https://repositorio.ufu.br/handle/123456789/49522 2026-08-14T06:19:50Z 2026-08-05T00:00:00Z

Title: Aprimoramento da resolução espacial do EEG: um modelo computacional baseado no algoritmo eLORETA Abstract: Electroencephalography (EEG) presents excellent temporal resolution, but suffers from spatial resolution limitations due to the nature of the brain volume conductor and the distance between the electrodes and the cortex. To circumvent this limitation, this work proposes and validates a model based on the eLORETA (Exact Low Resolution Brain Electromagnetic Tomography) algorithm to solve the Inverse Problem, focusing on the three-dimensional tracking of neural sources under electrotactile somatosensory stimulation. The pipeline developed was structured into three main stages: two-dimensional statistical inference, computational simulation of the Forward Problem, and spatial reconstruction on real data. Preliminary statistical analysis (ANOVA) confirmed that high-frequency tactile stimuli (125 Hz) elicit the P300 complex with greater amplitude and signal-to-noise ratio, optimizing the input matrices. During synthetic algorithmic validation, the model demonstrated a high capacity for multiple source separability, maintaining low-magnitude localization errors (restricted to distances of 1 to 2 voxels) even under acute 50\% white noise degradation. In the application on real biological data extracted from 10 volunteers using a 60-channel array, the software successfully isolated the anatomophysiological macroregion in the somatosensory cortex, suppressing the spatial scattering typical of surface EEG. Performance analysis further revealed that the adopted parametric mesh (5 mm and 12,213 voxels) established an optimal balance between spatial detail and memory processing viability. It is concluded that the consolidated system establishes a robust computational analytical tool, ensuring high precision in feature extraction for advanced electrophysiological data processing.

2026-08-05T00:00:00Z https://repositorio.ufu.br/handle/123456789/49517 2026-08-14T06:19:43Z 2026-07-31T00:00:00Z

Title: Definição de parâmetros de iluminação de cenas virtuais 3D a partir de imagens

2026-07-31T00:00:00Z