Reuse of medical images and segmentation tools to coadiuvate medicalclinical decision

The widespread use of medical imaging technologies in clinical practice has enabled the creation of large datasets. However, the potential of these data for developing automated decision-support systems remains underutilized due to the lack of capacity to parse and analyze those data. Automated and accurate medical image segmentation is crucial for developing decision-support tools. These segmentations can be valuable for initializing other algorithms, such as simulations, or for quantitative assessments of characteristics across different pathologies. Deep Learning methods require large, annotated datasets, which are not always available in this context. In this work, I have reused existing tools and datasets to enhance or enable the creation of annotated datasets to develop clinical decision-support tools. This work comprises two main tasks: the automated segmentation of the femoral region and the determination of a local reference system to initialize a simulation to estimate femoral fracture risk. A semi-automated tool was developed and validated for the creation of a large segmented dataset, used to test both the automated segmentation algorithm and the automated reference system identification. The automated segmentation tool combines existing deep learning models with graph-cut techniques, while the automated reference system consists of unsupervised techniques that don't require shape priors. The second task focused on the rapid segmentation of white matter hyperintensities in FLAIR MRI sequences of patients with Myotonic Dys-trophy Type 1 or Sickle Cell Disease. Here, a tool initially developed for multiple sclerosis was adapted to the new tasks. This tool, in conjunction with neuroradiologist expertise, was successfully used to create gold-standard segmented datasets and then adapted for the two target conditions. In summary, using existing tools for large-scale data annotation and integrating machine learning models, the research demonstrates more efficient segmentation workflows, paving the way for further advances in automated diagnostics.