Reproducibility metadata and experimental artifacts from a prostate cancer binary segmentation study using the public DIAGSET database

The study used the public DIAGSET database as the input image source; the original DIAGSET images are not redistributed in this deposit. The deposited materials document how the research data were processed throughout the workflow, including artifact identification, patch and binary-mask extraction, quality control, graph-based cleaning, patient-level partitioning into training, validation, and test sets, color-normalization strategies, smart sampling of training patches, learning-rate selection, supervised segmentation model training, ensemble optimization, and final inference/evaluation.

The collection includes CSV manifests, SQLite provenance and execution databases, GeoJSON-related artifact metadata, patient-level split artifacts, sampling records, model-training metadata, ensemble configuration files, and quantitative evaluation outputs. Its purpose is to support verification, traceability, and reproducibility of the computational methodology used to study prostate cancer segmentation in histopathological images.