Metadata record for the manuscript: Computer extracted gland features from H&E predicts prostate cancer recurrence comparably to a genomic companion diagnostic test: a large multi-site study

Summary This metadata record provides details of the data supporting the claims of the related manuscript: “Computer extracted gland features from H&E predicts prostate cancer recurrence comparably to a genomic companion diagnostic test: a large multi-site study”. The related study present a tissue non-destructive method for automated biochemical recurrence (BCR) prognosis, termed ”Histotyping”, that employs computational image analysis of morphologic patterns of prostate tissue from a single, routinely acquired hematoxylin and eosin slide. Type of data: automated BCR prognosis training and validation data; clinical data; ground truth maps & segmentation results Subject of data: Homo sapiens: computational image analysis of morphologic patterns of prostate tissue from a single, routinely acquired hematoxylin and eosin slide Sample size: n=889 patients Population characteristics: patient were digitised on a variety of whole-slide scanners. Recruitment: patients from 6 sources: University of Pennsylvania (UPenn), University Hospitals Cleveland Medical Center (UH), NewYork-Presbyterian Hospital/Weill Cornell Medical Center (WCM), University of Turku (UTurku), The Cancer Genome Atlas (TCGA), and the Icahn School of Medicine at Mount Sinai (MS). Data access The following files are openly available as part of this data record: - ‘training_set_HT_scores.xlsx’ contains the Histotyping scores of each patient in the training set (n=214). - ‘boundary_layer_data.xlsx’ contains the Histotyping scores of each validation set patients during successive boundary layer removals. - ‘HT_UMAP_supporting_data.xlsx’ contains image metrics, Histotyping scores, Histotyping+ scores, and UMAP components for each patient in the validation set. - The folder ‘gland_segmentations.zip’ contains ground truth masks and segmentation results on lumen segmentation model validation set images. Note that output masks are provided before preprocessing. Statistics and features in the related article were derived using postprocessing described in the supplementary information. The patient clinical data are contained in the Excel spreadsheet ‘patient_clinical_info.xlsx’. These data are not publicly available for the following reason: material transfer agreements from source hospitals do not allow public sharing of patient information. However, the data can be made available upon reasonable request to the corresponding author. Corresponding author(s) for this study Dr Anant Madabhushi, Case Western Reserve University, Department of Biomedical Engineering, Cleveland, OH, USA. [e-mail]. Study approval Data collection was approved by institutional review boards at each institution and conducted in accordance with U.S. Common Rule guidelines.