Banca de QUALIFICAÇÃO: VICTOR GABRIEL PEREIRA DE SOUSA

Uma banca de QUALIFICAÇÃO de MESTRADO foi cadastrada pelo programa.
DISCENTE: VICTOR GABRIEL PEREIRA DE SOUSA
DATA: 03/09/2026
HORA: 09:00
LOCAL: meet.google.com/fec-rtnt-mno
TÍTULO: Optimizing Channel Selection for U-Net Segmentation in MIBI-TOF of Triple-Negative Breast Cancer
PALAVRAS-CHAVES: Multiplexed Biomedical Imaging; Biomarker Selection; Semantic Segmentation; Computational Pathology; Dimensionality Reduction.
PÁGINAS: 12
GRANDE ÁREA: Ciências Exatas e da Terra
ÁREA: Ciência da Computação
RESUMO:

Multiplexed imaging techniques provide detailed representations of the tumor microenvironment by simultaneously capturing multiple molecular and structural markers. However, the high dimensionality of these images may introduce redundancy and increase the computational complexity of deep learning-based analysis. This work investigates the impact of channel selection on semantic segmentation of MIBI-TOF images from triple-negative breast cancer samples using a U-Net architecture. Individual channels are systematically evaluated and ranked according to their segmentation performance, and incremental multi-channel configurations are compared with a marker panel recommended in the literature. The results show that segmentation performance depends more on channel informativeness than on the number of channels used. In particular, the dsDNA channel alone achieves a Dice score of 89.36%, comparable to the 89.41% obtained with the complete seven-channel reference panel. These findings demonstrate that appropriate channel selection can substantially reduce input dimensionality while preserving segmentation accuracy, supporting more efficient deep learning pipelines for multiplexed biomedical image analysis.

MEMBROS DA BANCA:
Presidente - 2025885 - FLÁVIO HENRIQUE DUARTE DE ARAÚJO
Interno - 3420215 - LEONARDO PIO VASCONCELOS
Interno - 2025063 - ROMUERE RODRIGUES VELOSO E SILVA