2025-12-022025-12-022025-09-15MIRANDA, Arthur Monteiro. Avaliação postural em crianças e adolescentes com paralisia cerebral: validação clínica de uma abordagem quantitativa via estimativa de pose 2D. Orientadora: Adriana Rosa Garcez Castro. 2025. 80 f. Trabalho de Curso (Bacharelado em Engenharia Elétrica e Biomédica) – Faculdade de Engenharia Elétrica e Biomédica, Instituto de Tecnologia, Universidade Federal do Pará, Belém, 2025. Disponível em: https://bdm.ufpa.br/handle/prefix/8924. Acesso em:.https://bdm.ufpa.br/handle/prefix/8924Cerebral palsy (CP) is one of the most common neurological conditions in childhood, affecting approximately 2 to 2.5 per 1,000 live births in developed countries, with prevalence potentially higher in developing countries due to limited access to adequate medical resources. Early diagnosis is essential to ensure access to appropriate treatments, improving the quality of life for affected children and their families. However, the assessment process can be complex, involving subjective clinical evaluations and specialized imaging exams, often limited by the availability of trained professionals and advanced technological resources. Artificial Intelligence, particularly deep neural networks, has revolutionized medical image analysis, offering possibilities for automating and objectifying processes traditionally dependent on human expertise. In this work, AI is applied to 2D pose estimation, automatically identifying body keypoints and generating quantitative postural metrics, enabling a more objective and standardized assessment of postural alterations. In this context, this thesis aims to investigate and validate automated quantitative postural metrics for the objective evaluation of the severity of postural alterations in children and adolescents with cerebral palsy via 2D pose estimation using deep neural networks, providing objective data to support clinical documentation and monitoring of therapeutic progress. Method: Development of a mobile application using the BlazePose algorithm to extract 33 body keypoints from static images in three views (frontal, right lateral, and left lateral) of 40 participants (24 controls, 8 with spastic diplegic CP, and 8 with spastic hemiplegic CP). The methodology includes automated calculation of 31 quantitative postural metrics derived from the keypoints, including joint angles, segmental imbalances, body inclinations, and asymmetry indices. Statistical analysis of the metrics was performed using ANOVA, Tukey post-hoc tests, and effect size calculation (Cohen’s d) to quantify group differences.Acesso AbertoParalisia cerebralImagemRede neuralAnálise posturalCerebral palsyImageNeural networkPostural assessmentCNPQ::ENGENHARIAS::ENGENHARIA BIOMEDICATrabalho de Curso - Graduação - MonografiaAttribution-NonCommercial-NoDerivs 3.0 Brazil