This study presents a fine-tuning of the BERTimbau model for Named Entity Recognition (NER) in Electronic Health Records (EHR), focusing on oncology data in Brazilian Portuguese. The proposed approach employs an annotated dataset formatted in the Inside-Outside-Beginning (IOB) scheme, enabling precise extraction of clinical entities. The model was trained using the BERTimbau Base Tokenizer, with hyperparameters optimized through the Hugging Face library. Experimental results demonstrate robust performance, achieving F1scores between 0.83 and 0.85 across disease, procedure, and medication categories. When applied to a dataset of 125,825 new clinical records, the model achieved a correct identification rate exceeding 92%, underscoring its potential for medical decision support. Key challenges include enhancing classification accuracy for less frequent entities and adapting the model to diverse clinical contexts. Future directions involve dataset expansion, data augmentation techniques, exploration of more advanced transformer-based models, and integration into hospital decision support systems to further improve healthcare analytics.

The increasing volume of electronic health records (EHRs) in oncology presents a challenge for medical professionals who need to quickly and efficiently extract relevant patient information. Large Language Models (LLMs) have demonstrated potential in automating this process, reducing the cognitive load on healthcare providers while improving the quality and speed of clinical decision-making. This study presents an automatic summarization system for oncology patients’ EHRs, leveraging LLaMA 3 to generate concise and coherent medical summaries. By employing prompt engineering techniques, our approach structures the extraction of key clinical insights into an easily interpretable format, such as diagnoses, treatments, and test results. The system was validated on a dataset of anonymized medical records, demonstrating its capability to improve consultation efficiency and reduce redundancy in manual reviews.

In recent years, artificial intelligence (AI) has enabled continuous advances in medicine, playing a crucial role in the evolution of cancer research. One of the AI techniques widely used in this context is Natural Language Processing (NLP) due to its ability to extract information from unstructured documents such as clinical notes and progress reports. These notes can contain valuable data about the type of cancer and information that can be essential for improving the quality of patient treatment, providing data for research, and offering meaningful figures for health service managers, among other uses. This study aims to present a systematic literature review to investigate and synthesize technological advances in applying NLP in medicine, particularly in cancer diagnosis. It identifies gaps, challenges, and potential strategies to accelerate the diagnosis and treatment of cancer disease using AI.

This case study outlines the agile development of a web information system, called Leitos, to manage the ICU and semi-ICU beds assigned to COVID-19 patients. This system aims to aid the State Health Department of Rio Grande do Norte (SESAP/RN)'s actions in response to Brazil's coronavirus outbreak. The purpose of this study is threefold. First, we intend to present the methodology and strategies used to develop and implement the Leitos system. Second, we describe the adaptations we have made to the agile practices to cope with the urgency of pandemic times and the remote teams. Lastly, we present the main characteristics of the Leitos system. The case study describes a real scenario, where the SESAP/RN requested us to rapidly develop a technology to integrate ICU beds and patient data related explicitly to COVID-19 from 61 health units (hospitals and clinics) spread across the state. With a small remote team of volunteers, we heavily relied on the Agile Scrum Methodology, where the first sprint took only one week and covered approximately 70% of the system's requirements. By the end of October 2020, 215 government agents and clinical unit staff have used this system, which suggests the efforts described in this article were crucial to support SESAP/RN's decisions on public health during the earliest stages of coronavirus pandemic in Rio Grande do Norte. As the main contribution, this article provides valuable insights into the impact of COVID-19 outbreak and the remote work in real agile teams.

With the world pandemic generated by Covid-19, many places in the world are not able to rapidly measure the number of intensive care unit (ICU) beds existing and available in a city, state, or country. Knowing precisely the number of ICU beds in real-time is very important to estimate the health system collapse and to create strategies for the government to provide new ICU beds for patients. Thus, at Rio Grande do Norte, a state in Brazil, the State Health Department requested us to rapidly develop a technology to integrate ICU beds and patient data related explicitly to Covid-19 from the 58 health units (hospitals and clinics) in the state. Thus, this article presents the methodology and strategies used for the development and implementation of a web information system, called Leitos, for the management of ICU and semi-ICU beds assigned to Covid-19 patients. Nowadays, more than 200 government agents and clinical unit staff are using this system that presents the real-time situation of the ICU beds in this state.

The proliferation of Internet of Things (IoT) devices demands robust structures for processing and storing trillion gigabytes of data. Part of those data can be used for evaluating the solution's effectiveness and to improve it. Hence, a reliability analysis based on the solution's produced data is important to validate it and find weaknesses to solve. Therefore, in this paper we process the data generated by a sensor-based smart parking system to present a reliability analysis. This system is implemented placing ultrasonic sensors, a web service, the FIWARE platform, and a mobile application to Android devices, in a three-layer architecture. We evaluate how many errors occur in a certain time window, the success and error rates and the most common kind of error produced. Finally, we conclude that even with some instability our sensing approach is effective with a worst case success rate greater than 96%. Moreover, we find out that our system works in an open environment with tropical climate conditions.