Fecha
2025-09-14
Derechos de acceso
info:eu-repo/semantics/openAccess
Título de la revista
ISSN de la revista
Título del volumen
Editorial

Citas

plumx
0 citas en WOS
0 citas en
Proyectos de investigación
Unidades organizativas
Número de la revista
Resumen
Computer-aided diagnosis of breast histological images can be used to support identification of different types of tumors which is of high prognostic value for breast cancer patients. However, this task remains challenging due to stain variability, limited annotated data, and class imbalance. In this context, Transformer-based architectures have emerged as promising models for digital pathology and computer vision tasks. We propose a novel framework of cascaded deep learning classifiers which simplifies multi-label classification of images by creating a pipeline consisting of three binary classifiers and a final multi-class classifier. With the aim of maximizing balance performance, robustness, and computational efficiency of each model, we designed a comprehensive hyperparameter tuning framework that evaluates different optimization strategies, learning rate schedules, and cross-validation schemes. We further applied temperature calibration and optimized decision thresholds to improve probability reliability and class separation. After this process, the models consistently achieved strong performance with mean balanced accuracies of 0.9532 for Stage 1, 0.9750 for Stage 2 and 0.9352 for Stage 3. When integrated into the cascaded pipeline, the approach reached a mean balanced accuracy of 0.9323, demonstrating both accuracy and stability across folds.
Descripción
Categorías UNESCO
Palabras clave
breast cancer, deep learning, histology, cascaded classifiers, transformers
Citación
López Abad, Emilia. Trabajo Fin de Máster: Introducing Transformer-Based Cascaded Classifiers for Breast Cancer Histopathology Classification. Universidad Nacional de Educación a Distancia (España) 2025
Centro
E.T.S. de Ingeniería Informática
Departamento
Inteligencia Artificial
Grupo de investigación
Grupo de innovación
Programa de doctorado
Cátedra
Datos de investigación relacionados
DOI