Leila Weitzel

Segmentação e classificação de padrões nucleares de Ki-67 usando Deep Learning

The Ki-67 proliferation index is a cornerstone biomarker in diagnostic pathology, yet its manual assessment suffers from significant inter-observer variability and lacks functional granularity. Recent evidence suggests that the morphological patterns of Ki-67 nuclear staining (Nuclear Patterns - NPs), as described by Dias et al. (2021), provide deeper insights into the cell cycle phase and tumor biology. This post-doctoral research project aims to develop and validate an automated deep learning pipeline for the precise segmentation and classification of Ki-67 NPs. The methodology will comprise three core stages: 1) Nuclear instance segmentation from whole-slide images using state-of-the-art models (e.g., HoverNet, Mask R-CNN); 2) Construction of a curated dataset of nuclear patches annotated according to the NP classes (NP1-NP4 and Mitosis); 3) Training and benchmarking of deep learning classifiers (including CNNs, Vision Transformers, and Graph Neural Networks) for NP classification.