RADIOLOGICAL IDENTIFICATION AND SEGMENTATION OF INTRACRANIAL HEMORRHAGES IN COMPUTED TOMOGRAPHY

Resumo

Introduction: Intracranial hemorrhage is one of the most critical neurological conditions in radiology, being responsible for high morbidity and mortality. Computed tomography (CT) is the preferred initial method for its evaluation; however, manual interpretation of images is a complex, time-consuming process that is prone to errors, especially in subtle cases. In this context, artificial intelligence (AI) techniques, such as deep learning, emerge as promising tools to support diagnosis. Objective: This subproject focuses on the stage of radiological image segmentation, which is essential for training convolutional neural network (CNN) models. Specifically, the goal is the identification and delineation of intracranial hemorrhage areas in CT scans, generating binary reference masks. These masks act as a gold standard for AI algorithms to learn how to differentiate normal regions from hemorrhagic areas, thereby improving automated diagnostic accuracy. Method: The work will be conducted in stages: (i) collection of images from public radiological datasets; (ii) preprocessing with size normalization, artifact removal, and slice standardization, followed by data conversion into three-dimensional format, mask binarization, and the establishment of logical structures to ensure traceability between each image and its corresponding mask; (iii) manual segmentation of intracranial hemorrhage regions using medical image editing software, resulting in the creation of digital masks; (iv) organization of masks and corresponding images into training, validation, and testing sets, intended for use in deep learning algorithms. Results: The results are still preliminary, based on the initial analysis of approximately 800 CT images. So far, the preprocessing stage has been completed, and the creation of binary masks corresponding to intracranial hemorrhage areas has begun. These advances provide the foundation for future neural network training, ensuring traceability and quality in the prepared dataset. Conclusion: The clinical relevance of this project lies in its direct contribution to the development of decision-support systems, reducing radiologists’ analysis time and increasing diagnostic safety. From an academic perspective, the creation of masks represents an essential link between medical expertise and programming techniques, strengthening interdisciplinarity between health and software engineering. Thus, this project is expected to provide a solid foundation for building robust AI models capable of assisting in the rapid and accurate diagnosis of intracranial hemorrhages.