Thesis Clasificación de imágenes de mamografías mediante modelos de inteligencia artificial basados en vision transformer para la detección de tejido canceroso
Loading...
Date
2025-10
Authors
Journal Title
Journal ISSN
Volume Title
Program
Ingeniería Civil Telemática
Departament
Campus
Campus Casa Central Valparaíso
Abstract
El cáncer de mama es la principal causa de muerte por cáncer en la población femenina mundial y lograr una detección temprana es crucial para un tratamiento efectivo. El desarrollo de sistemas CAD (Diagnóstico asistido por computador) basados en CNN (Redes neuronales convolucionales) han demostrado ser de gran ayuda para los médicos, sin embargo, este tipo de modelos trabajan sobre una porción de la imagen, dejando fuera características relevantes y además, presentan una gran complejidad computacional debido a las múltiples convoluciones que se realizan. En este trabajo se presenta un modelo de deep learning basado en el paradigma de vision transformer, el cual emplea transfer-learning al tomar como base modelos pre-entrenados y que son adaptados mediante fine-tuning. Como resultado, se obtiene una arquitectura compuesta por tres clasificadores entrenados de forma independiente, que colectivamente se encargan de entregar como resultado si hay presencia de lesiones, junto con su tipo y nivel BI-RADS correspondiente.
Breast cancer is the leading cause of cancer-related death among the female population worldwide, and achieving early detection is crucial for effective treatment. The development of CAD (computer-aided diagnosis) systems based on Convolutional Neural Networks (CNNs) has proven to be highly beneficial for physicians. However, these models typically operate on a portion of the image, potentially excluding relevant features, and they also involve high computational complexity due to the numerous convolutions performed. This work presents a deep learning model based on the Vision Transformer para-digm, which employs transfer learning by using pre-trained models that are adapted through fine-tuning. As a result, an architecture composed of three independently trained classifiers is obtained; collectively, they are responsible for determining the presence of lesions, their type, and the corresponding BI-RADS level.
Breast cancer is the leading cause of cancer-related death among the female population worldwide, and achieving early detection is crucial for effective treatment. The development of CAD (computer-aided diagnosis) systems based on Convolutional Neural Networks (CNNs) has proven to be highly beneficial for physicians. However, these models typically operate on a portion of the image, potentially excluding relevant features, and they also involve high computational complexity due to the numerous convolutions performed. This work presents a deep learning model based on the Vision Transformer para-digm, which employs transfer learning by using pre-trained models that are adapted through fine-tuning. As a result, an architecture composed of three independently trained classifiers is obtained; collectively, they are responsible for determining the presence of lesions, their type, and the corresponding BI-RADS level.
Description
Keywords
Vision transformer, Inteligencia artificial, Cáncer de mama, Mamografía, Artificial intelligence, Breast cancer, Mammography
