Abstract
The wireless capsule endoscopy is a non-invasive imaging method that allows observation of the inner lumen of the small intestine, but with the cost of a longer duration to process its resulting videos. Therefore, the scientific community has developed several machine learning strategies to help reduce that duration. Such strategies are typically trained and evaluated on small sets of images, ultimately not proving to be efficient when applied to full videos. Labelling full Capsule Endoscopy videos requires significant effort, leading to a lack of data on this medical area. Active learning strategies allow intelligent selection of datasets from a vast set of unlabelled data, maximizing learning and reducing annotation costs. In this experiment, we have explored active learning methods to reduce capsule endoscopy videos' annotation effort by compiling smaller datasets capable of representing their content.

Abstract
Inherited retinal diseases such as Retinitis Pigmentosa and Stargardt's disease are genetic conditions that cause the photoreceptors in the retina to deteriorate over time. This can lead to vision symptoms such as tubular vision, loss of central vision, and nyctalopia (difficulty seeing in low light) or photophobia (high light). Timely healthcare intervention is critical, as most forms of these conditions are currently untreatable and usually focused on minimizing further vision loss. Machine learning (ML) algorithms can play a crucial role in the detection of retinal diseases, especially considering the recent advancements in retinal imaging devices and the limited availability of public datasets on these diseases. These algorithms have the potential to help researchers gain new insights into disease progression from previous classified eye scans and genetic profiles of patients. In this work, multi-class identification between the retinal diseases Retinitis Pigmentosa, Stargardt Disease, and Cone-Rod Dystrophy was performed using three pretrained models, ResNet101, ResNet50, and VGG19 as baseline models, after shown to be effective in our computer vision task. These models were trained and validated on two datasets of autofluorescent retinal images, the first containing raw data, and the second dataset was improved with cropping to obtain better results. The best results were achieved using the ResNet101 model on the improved dataset with an Accuracy (Acc) of 0.903, an Area under the ROC Curve (AUC) of 0.976, an F1-Score of 0.897, a Recall (REC) of 0.903, and a Precision (PRE) of 0.910. To further assess the reliability of these models for future data, an Explainable AI (XAI) analysis was conducted, employing Grad-Cam. Overall, the study showed promising capabilities of Deep Learning for the diagnosis of retinal diseases using medical imaging.

Abstract
The gastrointestinal tract is part of the digestive system, fundamental to digestion. Digestive problems can be symptoms of chronic illnesses like cancer and should be treated seriously. Endoscopic exams in the tract make detecting these diseases in their initial stages possible, enabling an effective treatment. Modern endoscopy has evolved into the Wireless Capsule Endoscopy procedure, where patients ingest a capsule with a camera. This type of exam usually exports videos up to 8 hours in length. Support systems for specialists to detect and diagnose pathologies in this type of exam are desired. This work uses a rarely used dataset, the ERS dataset, containing 121.399 labelled images, to evaluate three models from the EfficientNet family of architectures for the binary classification of Endoscopic images. The models were evaluated in a 5-fold cross-validation process. In the experiments, the best results were achieved by EfficientNetB0, achieving average accuracy and F1-Score of, respectively, 77.29% and 84.67%. Copyright © 2024 by SCITEPRESS – Science and Technology Publications, Lda.

Abstract
Effective image and video annotation is a fundamental pillar in computer vision and artificial intelligence, crucial for the development of accurate machine learning models. Object tracking and image retrieval techniques are essential in this process, significantly improving the efficiency and accuracy of automatic annotation. This paper systematically investigates object tracking and image acquisition techniques. It explores how these technologies can collectively enhance the efficiency and accuracy of the annotation processes for image and video datasets. Object tracking is examined for its role in automating annotations by tracking objects across video sequences, while image retrieval is evaluated for its ability to suggest annotations for new images based on existing data. The review encompasses diverse methodologies, including advanced neural networks and machine learning techniques, highlighting their effectiveness in various contexts like medical analyses and urban monitoring. Despite notable advancements, challenges such as algorithm robustness and effective human-AI collaboration are identified. This review provides valuable insights into these technologies' current state and future potential in improving image annotation processes, even showing existing applications of these techniques and their full potential when combined.

Abstract
The early and accurate diagnosis of brain tumors is critical for effective treatment planning, with Magnetic Resonance Imaging (MRI) serving as a key tool in the non-invasive examination of such conditions. Despite the advancements in Computer-Aided Diagnosis (CADx) systems powered by deep learning, the challenge of accurately classifying brain tumors from MRI scans persists due to the high variability of tumor appearances and the subtlety of early-stage manifestations. This work introduces a novel adaptation of the EfficientNetv2 architecture, enhanced with Global Attention Mechanism (GAM) and Efficient Channel Attention (ECA), aimed at overcoming these hurdles. This enhancement not only amplifies the model's ability to focus on salient features within complex MRI images but also significantly improves the classification accuracy of brain tumors. Our approach distinguishes itself by meticulously integrating attention mechanisms that systematically enhance feature extraction, thereby achieving superior performance in detecting a broad spectrum of brain tumors. Demonstrated through extensive experiments on a large public dataset, our model achieves an exceptional high-test accuracy of 99.76%, setting a new benchmark in MRI-based brain tumor classification. Moreover, the incorporation of Grad-CAM visualization techniques sheds light on the model's decision-making process, offering transparent and interpretable insights that are invaluable for clinical assessment. By addressing the limitations inherent in previous models, this study not only advances the field of medical imaging analysis but also highlights the pivotal role of attention mechanisms in enhancing the interpretability and accuracy of deep learning models for brain tumor diagnosis. This research sets the stage for advanced CADx systems, enhancing patient care and treatment outcomes.

Abstract
Pathologies in concrete structures can be visually evidenced on the concrete surface, such as by fissures or cracks, fragmentation of part of the concrete, concrete efflorescence, corrosion stains on the concrete surface, or exposed steel bars, the latter two occurring in reinforced concrete. Therefore, these pathologies can be analyzed via the images of concrete structures. This article proposes a methodology for visually inspecting concrete structures using deep neural networks. This method makes it possible to speed up the detection task and increase its effectiveness by saving time in preparing the identifications to be analyzed and eliminating or reducing errors, such as those resulting from human errors caused by the execution of tedious, repetitive analysis tasks. The methodology was tested to analyze its accuracy. The neural network architecture used for detection was YOLO, versions 4 and 8, which was tested to analyze the gain with migration to a more recent version. The dataset for classification was Ozgnel, which was trained with YOLO version 8, and the detection dataset was CODEBRIM. The use of a dedicated classification dataset allows for a better-trained network for this function and results in the elimination of false positives in the detection stage. The classification achieved 99.65% accuracy.