SPINAL CORD SEGMENTATION AS OAR IN PLANNING CT FOR RADIOTHERAPY USING HISTOGRAM MATCHING, TEMPLATE MATCHING, AND U-NET

Resumo

Radiotherapy is one of the major option used in cancer management. The treatment involves several steps, one of which is the construction of a computed tomography (CT) model of the patient so that the target tissues and organs at risk (OARs) surrounding that target can be evaluated. With the CT, the responsible physician delimits the OARs slice by slice, as the spinal cord that comprises almost all the tomography becomes more tiring to be segmented and thus susceptible to errors. Thus, this paper presents a method of spinal cord segmentation in planning CT for radiotherapy using template matching, histogram matching and a fully convolutional neural network. The result achieved an accuracy of 99.38\%, specificity of 99.12\%, sensitivity of 93.83\%, and dice index of 81.33\%, without any segmentation refinement