https://repositorio.ufba.br/handle/ri/34554 Fri, 15 May 2026 13:02:07 GMT 2026-05-15T13:02:07Z https://repositorio.ufba.br/handle/ri/40970 Título: Um estudo abrangente sobre segmentação de glomérulos com dados de treinamento limitados em imagens histopatológicas de alta resolução Autor(es): Souza Júnior, Luiz Otávio de Oliveira Primeiro Orientador: Oliveira, Luciano Rebouças de Abstract: The growing availability of scanned whole slide images (WSIs) has expanded digital pathology, enabling medical decision-making and computational analysis directly from high-resolution images. Kidney disease diagnoses using WSIs rely on the analysis of specific tissue structures, and automatic analysis depends on accurately segmenting key components such as glomeruli, tubules, interstitium, and vessels. This thesis focuses on glomeruli, which are essential in assessing WSIs after kidney biopsies. These structures are impacted by lesions related to various diseases. In machine learning-based WSI analysis, glomeruli are often the first regions segmented to guide subsequent tasks. The Bowman’s capsule (BC) is crucial, marking the boundary between glomerular components and surrounding interstitial tissue. This work proposes two studies aimed at addressing the segmentation of glomeruli in high-resolution kidney histopathological images. In the first study, we investigate the feasibility of segmenting glomeruli in human WSIs using deep-learning models trained exclusively on mouse data. Mice and humans share several biological similarities, including genetic, physiological, and structural characteristics, making mice a common model for studying human diseases. While this cross-species knowledge transfer is well-established in medicine, it remains underexplored in computational pathology, where WSIs serve as primary research objects. To address this gap, we evaluated five semantic segmentation models: U-Net, U-Net 3+, Res-U-Net, DeepLabV3+, and MA-Net, using datasets consisting of 18 mouse WSIs and 42 human WSIs. Among these, U-Net 3+ delivered the best performance in intra-dataset evaluation, achieving an average DICE score of 0.930 on HE-stained mouse images. On human data, U-Net 3+ also excelled, attaining DICE scores of 0.772, 0.824, and 0.791 on HE, PAS, and PAMS stains, respectively. Moreover, U-Net 3+ proved promising generalization when trained solely on mouse data and tested across the entire human dataset, achieving a DICE score of 0.798 on HE-stained images. However, while these models performed well on images within the same staining technique, their performance declined when applied across different stains, highlighting a limitation in cross-stain generalization. The second study focuses on the segmentation challenges posed by borderless glomeruli affected by global sclerosis. We developed an automated framework for patch cropping and stitching, eliminating manual intervention to streamline the segmentation process. Our experiments show that while standard segmentation models can achieve state-of-the-art results for normal and partially sclerotic glomeruli, their performance deteriorates significantly for globally sclerotic glomeruli. Notably, segmentation accuracy for these cases was highly dependent on the staining type and generally remained poor across models. We compared non-foundation models (U-Net, U-Net 3+, and SwinTransformer + U-Net) with and without fine-tuning against the SegGPT foundation model. Non-foundation models, trained exclusively on normal glomeruli with HE, PAS, and PAMS stains, achieved high performance on normal glomeruli (mDice > 0.92) and moderate performance on partially sclerotic glomeruli (mDice > 0.72). However, their performance dropped sharply to mDice > 0.02 for globally sclerotic glomeruli, with minimal improvements even after fine-tuning. In contrast, SegGPT demonstrated substantial improvement, achieving a significantly higher mDice score (> 0.37) for globally sclerotic glomeruli by leveraging only a few query samples. This result highlights the potential of foundation models in addressing segmentation challenges for glomeruli affected by severe lesions. In summary, the studies presented in this thesis represent a significant step forward in the segmentation of glomeruli in WSIs. Our findings offer a comprehensive analysis of glomerulus segmentation with limited training data, demonstrating the potential of mouse-to-human transfer learning, as well as the use of foundation models to improve segmentation accuracy for glomeruli affected by sclerosis. Editora / Evento / Instituição: Universidade Federal da Bahia Tipo: Tese Fri, 06 Dec 2024 00:00:00 GMT https://repositorio.ufba.br/handle/ri/40970 2024-12-06T00:00:00Z https://repositorio.ufba.br/handle/ri/40805 Título: Modeling and uncertainty assessment of dynamical systems and digital twins Autor(es): Costa, Erbet Almeida Primeiro Orientador: Schnitman, Leizer Abstract: This thesis presents an innovative methodology aimed to meet the growing demands and challenges in Scientific Machine Learning (SciML) applied to electrical submersible pumps (ESP) and pressure swing adsorption (PSA) processes through robust learning, parameter estimation, deep learning, and digital twins. The proposed methodology comprehensively evaluates multiple facets of uncertainty inherent to identifying the SciML model, considering the literature base, data sensitivity, and computational effort. The methodology identifies and validates deep learning models, using non-linear models to generate and overcome experimental data limitations. The methodology uses an integrated Bayesian method as a methodological step to estimate parameters, assess uncertainty, and validate phenomenological and data-driven models. The method is treated in steps that successfully align the model with experimental data, both dynamically and in a steady state, showing the methodology’s potential to represent the system’s behavior within existing uncertainties. This development enables the construction of reliable and computationally efficient dynamic AI models for planning, building, controlling, and optimizing digital twins. The methodology is put under test in several case studies. The results are composed of models validated against synthetic and experimental data that are compatible with the dynamic behavior of the nonlinear model and its uncertainties. The first validation of the method is carried out through a case study involving the development of a soft sensor for a polymerization reactor, which demonstrates robustness and consistency in the treatment of uncertainties in the SciML field. Two case studies are performed for ESP-based artificial lift systems. In those case studies, the technique showed promise for the characterization and representation of the system and paves the way for applications in oil production fields, particularly in production control, optimization, and assistance. In the context of PSA processes, a new approach is presented for developing a digital twin with uncertainty assessment capable of mapping PSA systems’ cyclical and complex behavior. Through continuous online learning and integrating a new feedback tracker, the digital twin accurately represents and adapts to the complexities of the PSA system, addressing challenges such as adsorbent degradation. This methodology offers answers about the applications of AI and digital twins in optimizing industrial processes and supporting sustainable development in various sectors. Together, these works contribute valuable results and methodologies to their respective fields, demonstrating the potential of advanced technologies to improve system representation, address uncertainties, and pave the way for future developments in industrial applications. Editora / Evento / Instituição: Universidade Federal da Bahia Tipo: Tese Thu, 13 Jun 2024 00:00:00 GMT https://repositorio.ufba.br/handle/ri/40805 2024-06-13T00:00:00Z https://repositorio.ufba.br/handle/ri/40721 Título: Implementable MPC strategies for oilfield with electric submersible pump installations Autor(es): Santana, Bruno Aguiar Primeiro Orientador: Martins, Márcio André Fernandes Abstract: This work addresses the development of model predictive control (MPC) strategies applied to artificial lift systems based on electric submersible pumping (ESP), aiming to overcome challenges related to process stabilization in wide operating ranges, compliance with typical ESP constraints, and the pursuit of energy efficiency goals. Although the topic has been widely explored in the literature, the proposed solutions are mostly based on linear approaches that have degraded performance outside the nominal condition. Strategies based on nonlinear models and adaptive MPC have been suggested, but they face practical challenges, such as implementation in embedded systems and ensuring closed-loop stability, factors that are associated with the safety and reliability required by the industry. The main contribution of this thesis is the creation of MPC schemes that stabilize the process in a wide operating range, considering economic goals and operational constraints, in addition to allowing implementation in embedded systems with automatic generation of C/C++ code. This thesis proposes four MPC strategies to overcome the limitations of the approaches found in the literature: (i) an adaptive MPC control with successive linearization of the ESP model, ensuring the feasibility of the control law and operational stability over a wide range; (ii) a nonlinear MPC control coupled with an extended Kalman filter, which extends the operating range and estimates difficult-to-measure variables, such as the average flow rate; (iii) a robust extension of the infinite-horizon MPC, which considers operational uncertainties and constraints and ensures robust stabilization; and (iv) an improved version of the nonlinear MPC with an infinite prediction horizon to ensure nominal stability and computational feasibility in practical scenarios. Simulations and hardware-in-the-loop tests demonstrate the feasibility of these MPC solutions for real-time operation, highlighting the potential of nonlinear and robust approaches to optimize the operation of ESP systems in oil production fields. Editora / Evento / Instituição: Universidade Federal da Bahia Tipo: Tese Tue, 20 Aug 2024 00:00:00 GMT https://repositorio.ufba.br/handle/ri/40721 2024-08-20T00:00:00Z https://repositorio.ufba.br/handle/ri/39987 Título: Análise da eficiência computacional para solução do problema da cinemática inversa de robôs antropomórficos utilizando a teoria de bases de Gröbner. Autor(es): Silva, Sérgio Ricardo Xavier da Primeiro Orientador: Schnitman, Leizer Abstract: The Denavit-Hartenberg algorithm is a method used for decades to solve one of the classic problems in the kinematics of robotic manipulators, the inverse kinematic problem. When this method is used, there is a need for additional algorithms to solve the problem, such as the Paul method. The Gröbner Bases Theory for the solution of inverse kinematics, as a supplementary method to the Denavit-Hartenberg algorithm, will be presented in this work. To familiarize the reader with each method, the manipulator robots Stäubli TS20, a SCARA type robot, and Unimation PUMA 560, an anthropomorphic manipulator with six rotating joints, will be used as case studies applying the Paul method and the method proposed in this work, where the computational efficiency data will be used for comparison. The main goal of this work is to analyze the computational efficiency in solving the problem of the inverse kinematics of anthropomorphic manipulating robots using both methods. With each approach, the problem of inverse kinematics for the two serial robots will be solved. When comparing each method, this work will demonstrate that the method using Gröbner Bases Theory is more computationally efficient for the solution of the inverse kinematic problem of anthropomorphic robots. Editora / Evento / Instituição: Universidade Federal da Bahia Tipo: Tese Mon, 13 Jul 2020 00:00:00 GMT https://repositorio.ufba.br/handle/ri/39987 2020-07-13T00:00:00Z