Organ segmentation with computerized tomography images using neural networks

Currently, the analysis and study of medical images with Artificial Intelligence have played a significant role because health is one of the most important areas in daily life. Usually, doctors give accurate diagnoses to their patients, but giving a diagnosis requires time to analyze each case. For example, computerized images depend greatly on specialized personnel and machines to obtain these computed tomography images. For these reasons, the areas of computer science have different approaches to analyzing medical images. Furthermore, the present work is based mainly on analyzing computerized images using Neural Networks, allowing organ segmentation. For this study, computerized liver images have been chosen since the liver is one of the main organs for medical diagnoses. To achieve the objective and implement two architectures, a primary convolutional neural network encodes and decodes, and another is called U-Net. The most important step for this is manipulating the data, in this case, the manipulation of the images. For this, pre-processing was carried out using image processing algorithms to improve quality. Moreover, the implementations were evaluated by comparing different metrics such as loss, precision, dice coefficient, and mean square error, obtaining better values of similarity with the Dice coefficient, which is the one that dictates the precision of the segmentation. However, the accuracy is not the highest between the two methods. For this reason, a more robust future work with more data is necessary.

• 1. Verfasser: Chuquín Rivera, Shirley Lizeth (author)
• Format: bachelorThesis
• Sprache: eng
• Veröffentlicht: 2021
• Schlagworte: Imágenes tomográficas computarizadas; Segmentación de imágenes; Redes convolucionales; Pre procesamiento de imágenes; CT Images; Image segmentation; Convolutional neural networks (CNN); Image Pre-Processing
• Online Zugang: http://repositorio.yachaytech.edu.ec/handle/123456789/450