Desarrollo de un clasificador basado en redes neuronales para la detección de escoliosis en imagen RX de columna.

thumbnail.default.alt
Share
Date
2022-01-27
Publisher
Universidad Antonio Nariño
Document type
COAR type
http://purl.org/coar/resource_type/c_7a1f
Citation
Bibliographic Managers
Document Viewer
Select a file to preview:
item.page.resume
Abstract
Medicine has grown considerably with technology in recent years, tools such as Machine Learning in the medical area are some of the most technical that represent a great benefit for the prevention and diagnosis of pathology in patients. In this work, a convolutional neural network consisting of two convolution layers and a feedforward network with 128 neurons and an output layer with three neurons is implemented in Python for the classification of scoliosis type C and S and healthy patients from RX images. of column of a Kaggle database and a proprietary database obtained from the Sagrado Corazón de Jesús Hospital, with the purpose that this algorithm serves as support to physiotherapists and orthopedists in decision making and that it can be carried out in a shorter time. The results obtained during the development of the tests indicate that the tool has an acceptable degree of accuracy compared to other articles such as "Development and Validation of Deep Learning Algorithms For Scoliosis Screening Using Back Images" Which Has An Accuracy Over 80% Using Back Images. Real-time patients or photographs that included a total of 3240 images for training (including 1029 men and 2211 women) and for validation a total of 400 images (300 images with scoliosis and 100 of healthy patients), however the implemented classifier In this work, a limited sample of images was trained and validated (150 images, divided into 75 images for training and 75 images for testing), obtaining a success rate of 94%, which can be a support tool for the diagnosis of the type of scoliosis in an automated way, unlike conventional methods
item.page.coverage.spatial
item.page.coverage.temporal