Banca de DEFESA: WYSTERLÂNYA KYURY PEREIRA BARROS

Uma banca de DEFESA de DOUTORADO foi cadastrada pelo programa.
STUDENT : WYSTERLÂNYA KYURY PEREIRA BARROS
DATE: 28/08/2026
TIME: 14:30
LOCAL: https://meet.google.com/fhc-gfwk-cqx
TITLE:

Artificial intelligence for whole slide image analysis in breast cancer

KEY WORDS:

Breast cancer, machine learning, computer vision, digital pathology, histopathology.

PAGES: 120
BIG AREA: Engenharias
AREA: Engenharia Elétrica
SUMMARY:

Breast cancer remains a major public health challenge, and histopathological analysis remains essential for the diagnosis and characterization of the disease. However, the increasing demand for examinations and the limited availability of pathologists have motivated the development of computational methods capable of assisting in the analysis of digitized histological slides. In this context, this work proposes a computational framework for the analysis of whole-slide images (WSIs) of breast tissue, using data obtained primarily from the TCGA-BRCA and HCMI-CMDC projects. In the proposed approach, high-resolution images are divided into smaller regions, referred to as tiles, from which features are extracted using the classical convolutional model VGG19 and CONCH, a foundation model specialized in pathology. These features are subsequently aggregated to generate compact slide-level representations through global strategies, tumor probability-guided strategies, and strategies based on the combination of different feature descriptors. The framework was evaluated on the tasks of classifying tumor versus normal tissue, differentiating histological subtypes, morphologically estimating PAM50-associated classes, and characterizing histopathological components, using classical machine learning algorithms and nested cross-validation protocols. The results demonstrated high discriminative ability in the classification of tumor and normal tissue, whereas histological subtyping, morphological estimation of PAM50-associated classes, and tissue characterization showed more moderate and variable performance due to greater morphological heterogeneity, class imbalance, and the complexity of the respective problems. Overall, the combination of representations extracted by different models yielded superior or more consistent performance compared with the isolated use of individual features, indicating complementarity among the evaluated descriptors. These findings indicate that the proposed strategies enable the generation of compact and discriminative WSI representations, contributing to the investigation and development of computational methods for the histopathological analysis of breast tissue.

COMMITTEE MEMBERS:
Presidente - 1837240 - MARCELO AUGUSTO COSTA FERNANDES
Interno - 1153006 - LUIZ AFFONSO HENDERSON GUEDES DE OLIVEIRA
Externo ao Programa - 3884005 - PATRICK CESAR ALVES TERREMATTE - UFRNExterna à Instituição - RAQUEL DE MELO BARBOSA - UGR
Externo à Instituição - RICARDO NEY OLIVEIRA COBUCCI - UnP
Externo à Instituição - SERGIO NATAN SILVA - UFCG