In this work, an algorithm of automatic segmentation of the lung, the wrapping bones and the heart in three-dimensional computerized tomography images will be defined. The segmentation algorithm will be based on thresholds, labeling and watershed. This is the first step in the creation of an anatomical atlas, which is used to regularize the inverse problem of electrical impedance tomography (TIE). Two thousand three-dimensional images of computed tomography of the thoracic region will be segmented. To this end, the DICOM files will be read, the relevant metadata will be determined, the images will be resampled to isotropic resolution. Visualization routines will be created to evaluate the results. This set of results will be used in another scientific initiation project that will segment the lung using supervised learning. It will also be possible to verify if incorrectly segmented cases will be correctly segmented by the technique that makes use of supervised learning.
News published in Agência FAPESP Newsletter about the scholarship: