Please use this identifier to cite or link to this item:
Title: Redes completamente convolucionales en la segmentación semántica de lesiones melanocíticas
Author: Moral Algaba, Fernando
Director/Tutor: Balocco, Simone
Keywords: Melanoma
Càncer de pell
Treballs de fi de grau
Diagnòstic per la imatge
Aprenentatge automàtic
Xarxes neuronals (Informàtica)
Reconeixement de formes (Informàtica)
Skin cancer
Computer software
Bachelor's thesis
Diagnostic imaging
Machine learning
Neural networks (Computer science)
Pattern recognition systems
Issue Date: 21-Jun-2017
Abstract: [en] Skin cancer is the more common type of cancer. Melanoma, that begins at melanocytes, is the most aggressive type of skin cancer and responsible of about 90 % of total deaths caused by this disease. Early diagnosis is the best way to defeat melanoma and can increase survival rate to near 100 %. Studies on Automated image detection of skin lesion has evolved achieving high rates of accuracy on melanoma detection and classification. Deep learning and Fully Convolutional Networks has become and useful tool on image analysis. This project explores the application of FCNs on semantic segmentation over combinations of two major datasets, images from dermatologic databases and skin mole images captured by cellular phone camera. Trained nets has been tested over another two datasets of unseen images of skin moles and dermatologic images. Data generated at this study evidence high accuracy, precision, sensitivity and speci city rates despite the small database size, which is composed by only a few hundreds images.
Note: Treballs Finals de Grau d'Enginyeria Informàtica, Facultat de Matemàtiques, Universitat de Barcelona, Any: 2017, Director: Simone Balocco
Appears in Collections:Treballs Finals de Grau (TFG) - Enginyeria Informàtica
Programari - Treballs de l'alumnat

Files in This Item:
File Description SizeFormat 
codi_font.zipCodi font9.61 MBzipView/Open
memoria.pdfMemòria11.76 MBAdobe PDFView/Open

This item is licensed under a Creative Commons License Creative Commons