Use this link to cite:
http://hdl.handle.net/2183/39634 Detección y clasificación automática de los niveles de riesgo de las lesiones melanocíticas en imágenes dermatoscópicas
Loading...
Identifiers
Publication date
Authors
Filgueiras Baamonde, Marcos
Other responsabilities
Universidade da Coruña. Facultade de Informática
Journal Title
Bibliographic citation
Type of academic work
Academic degree
Abstract
[Resumen]: El presente proyecto tiene como objetivo el desarrollo de una herramienta para la detección y clasificación automática de lesiones melanocíticas en imágenes dermatoscópicas. El melanoma maligno es uno de los tipos de cáncer de piel más agresivos, con una incidencia creciente en todo el mundo. A nivel global, se diagnostican aproximadamente 324,000 nuevos casos de melanoma cada año, y en España, se estima que la incidencia alcanza los 12-13 casos por cada 100,000 habitantes, siendo más frecuente en personas de piel clara. La mortalidad por melanoma aumenta considerablemente en las fases avanzadas de la enfermedad, lo que subraya la importancia crucial de un diagnóstico precoz para mejorar la supervivencia y reducir la tasa de mortalidad. Con este fin, el proyecto ha desarrollado una herramienta utilizando una combinación de técnicas avanzadas de deep learning y procesamiento de imágenes, entrenando modelos que permiten la segmentación precisa de la zona afectada y la posterior clasificación de las lesiones en diferentes categorías. La detección temprana es clave para el éxito del tratamiento, dado que las tasas de supervivencia a cinco años son superiores al 99% en los casos en los que el melanoma se detecta en sus fases iniciales, pero caen drásticamente cuando el cáncer se ha diseminado. El conjunto de datos utilizado proviene de la ISIC, la mayor base de datos pública de imágenes dermatoscópicas. A partir de estos datos, se ha implementado un modelo de segmentación basado en la arquitectura U-Net, mientras que para la clasificación se ha optado por modelos como DenseNet, aplicando técnicas de transfer learning para mejorar los resultados. Además, se ha hecho uso de data augmentation para incrementar la robustez del conjunto de datos y mitigar el desbalance entre clases. Adicionalmente, se ha desarrollado una aplicación móvil para Android, que permite utilizar los modelos entrenados de forma portable, proporcionando una interfaz simple e intuitiva para facilitar la carga de imágenes, su análisis y la descarga de los resultados. La aplicación ha sido diseñada pensando en su futura expansión, permitiendo la integración de nuevos modelos y funcionalidades. El proyecto ha sentado las bases para investigaciones futuras en la detección temprana del melanoma, y ya se ha comenzado la validación clínica en colaboración con el Servicio de Dermatología del Complejo Hospitalario Universitario A Coruña (CHUAC), lo que permitirá evaluar la efectividad del sistema en un entorno clínico y mejorar su aplicabilidad.
[Abstract]: The aim of this project is to develop a tool for the automatic detection and classification of melanocytic lesions in dermoscopic images. Malignant melanoma is one of the most aggressive types of skin cancer, with an increasing incidence worldwide. Globally, approximately 324,000 new cases of melanoma are diagnosed each year, and in Spain, the incidence is estimated to reach 12-13 cases per 100,000 inhabitants, being more frequent in people with fair skin. Mortality from melanoma increases significantly in the advanced stages of the disease, highlighting the crucial importance of early diagnosis to improve survival and reduce mortality rates. To this end, the project has developed a tool using a combination of advanced deep learning techniques and image processing, training models that allow the precise segmentation of the affected area and the subsequent classification of the lesions into different categories. Early detection is key to the success of treatment, as five-year survival rates are above 99% in cases where melanoma is detected in its early stages, but they drop dramatically once the cancer has spread. The dataset used comes from ISIC, the largest public database of dermoscopic images. From this data, a segmentation model based on the U-Net architecture was implemented, while DenseNet-based models were used for classification, applying transfer learning techniques to improve the results. Additionally, data augmentation was used to increase the robustness of the dataset and mitigate class imbalance. Furthermore, a mobile application for Android was developed, allowing the trained models to be used in a portable way, providing a simple and intuitive interface to facilitate image uploading, analysis, and result downloading. The application was designed with future expansion in mind, allowing for the integration of new models and features. The project has laid the foundation for future research in the early detection of melanoma, and clinical validation has already begun in collaboration with the Dermatology Service at the University Hospital Complex of A Coruña (CHUAC), which will allow for evaluating the system’s effectiveness in a clinical setting and enhancing its applicability.
[Abstract]: The aim of this project is to develop a tool for the automatic detection and classification of melanocytic lesions in dermoscopic images. Malignant melanoma is one of the most aggressive types of skin cancer, with an increasing incidence worldwide. Globally, approximately 324,000 new cases of melanoma are diagnosed each year, and in Spain, the incidence is estimated to reach 12-13 cases per 100,000 inhabitants, being more frequent in people with fair skin. Mortality from melanoma increases significantly in the advanced stages of the disease, highlighting the crucial importance of early diagnosis to improve survival and reduce mortality rates. To this end, the project has developed a tool using a combination of advanced deep learning techniques and image processing, training models that allow the precise segmentation of the affected area and the subsequent classification of the lesions into different categories. Early detection is key to the success of treatment, as five-year survival rates are above 99% in cases where melanoma is detected in its early stages, but they drop dramatically once the cancer has spread. The dataset used comes from ISIC, the largest public database of dermoscopic images. From this data, a segmentation model based on the U-Net architecture was implemented, while DenseNet-based models were used for classification, applying transfer learning techniques to improve the results. Additionally, data augmentation was used to increase the robustness of the dataset and mitigate class imbalance. Furthermore, a mobile application for Android was developed, allowing the trained models to be used in a portable way, providing a simple and intuitive interface to facilitate image uploading, analysis, and result downloading. The application was designed with future expansion in mind, allowing for the integration of new models and features. The project has laid the foundation for future research in the early detection of melanoma, and clinical validation has already begun in collaboration with the Dermatology Service at the University Hospital Complex of A Coruña (CHUAC), which will allow for evaluating the system’s effectiveness in a clinical setting and enhancing its applicability.
Description
Editor version
Rights
Atribución 3.0 España








