Tratamiento informático para la segmentación de tejido pulmonar en imágenes de tomografía computarizada, mediante el desarrollo de un código en python.

Abstract

In this project, we developed two tools that will serve as a preliminary step for generating future Computer Aided Detection (CAD) systems for the automated detection of numerous lung diseases. The first tool was made with the aim of extracting information from files with DICOM format in an easy and fast way without needing to use other types of external programs, but only by using Python and the Jupyter Notebook programming environment, which are open source and freely available. The implemented process makes it possible to generate images in different anatomical planes, such as: the coronal plane and the sagittal plane, by means of the pixel matrix that represents an axial section of the computed tomography process. We used techniques that are independent of the image generated in grayscale. In addition, they do not depend on the type of window used, since the pixel matrix in Hounsfield units is used to obtain the image. On the other hand, the second tool consists of a lung segmentation technique for chest computed tomography images. The segmentation process uses the pixel matrix in Hounsfield units to initially identify lung tissue through its degree of linear X-ray attenuation. Then, the marker-based Watershed algorithm is used, which is the main basis of the technique. In addition, different morphological operations and transformations are used on the axial sections for obtaining the lung segmentation. In this project, 35 tests in computed tomography (CT) modality with different protocols and with an average of 97 images per test were used to evaluate the first tool. Similarly, the lung segmentation tool was tested against 17 chest CT tests, demonstrating that the tool is accurate and fast