Please use this identifier to cite or link to this item:
http://hdl.handle.net/10609/138407
Title: | Estudio de técnicas de machine learning para el diagnóstico del melanoma y otras lesiones cutáneas a partir de imágenes |
Author: | Barba Sánchez, Agustín Miguel |
Tutor: | Yu, Longlong |
Others: | Ventura, Carles |
Abstract: | Skin cancer is the most common type of cancer and although melanoma accounts for only 1% of skin cancer, it is one of the deadliest, especially if detected in advanced stages. An early diagnosis would allow increasing the options of its treatment and patient survival. This diagnosis is made by visual inspection of the lesions, measuring parameters that are potentially detectable by a computer vision system, such as size, colour and shape. This opens the door to automatic systems for the diagnosis of skin lesions. The objective of this work is the implementation of an automatic skin lesion classifier. For this purpose, we used the HAM 10000 dataset, a set of dermatoscopic images compiled to train this type of systems. To do this, we have started from an initial prototype based on the EfficientNet model and then different techniques have been applied to improve the system's response. In addition, to improve the training of the model, the original dataset has been augmented using generative adversarial networks. The result of this work is that starting from an initial result of 0.34 for the f1-score and 0.63 for the accuracy, it has been improved to 0.75 for the f1-score and 0.86 for the accuracy. This result is similar to that obtained by human experts, so it could be used as an aid to diagnosis and decision-making. |
Keywords: | deep learning medical image GANs |
Document type: | info:eu-repo/semantics/masterThesis |
Issue Date: | 29-Dec-2021 |
Publication license: | http://creativecommons.org/licenses/by-nc-nd/3.0/es/ |
Appears in Collections: | Bachelor thesis, research projects, etc. |
Files in This Item:
File | Description | Size | Format | |
---|---|---|---|---|
abarsanTFM0122memoria.pdf | Memoria del TFM | 4,3 MB | Adobe PDF | View/Open |
Share:
This item is licensed under a Creative Commons License