The Prostate_cISH_Epithelium_Segmentation Dataset
Corresponding author: Henrik Sahlin Pettersen (henrik.s.pettersen@ntnu.no)
Consultant Pathologist / Associate Professor, St. Olav's Hospital / NTNU, Trondheim, Norway
Short Webpage Description
This dataset provides high-resolution histopathological images and corresponding expert-annotated segmentation masks, specifically designed for developing AI models for prostate epithelium segmentation. The images feature Chromogenic In Situ Hybridization (cISH) staining for various miRNAs, alongside controls and standard Hematoxylin & Eosin (HE) stains. Data originates from 70 patients: 30 with prostate cancer (PCa) and 40 with benign prostatic hyperplasia (BPH).
Sample Collection
Prostate Cancer (PCa): 30 patients. Samples include normal glandular epithelium (core type 'a'), Gleason 3 pattern (core type 'b'), and Gleason 4 pattern (core type 'c'), ideally in triplicate for each marker.
Benign Prostatic Hyperplasia (BPH): 40 patients. Samples consist of triplicate normal glandular epithelium for each marker.
Markers and Controls
Data is provided for the following stains, each organized into its own top-level folder:
miRNAs: miR‑550A, miR‑1246, miR‑3614, miR‑4326, miR‑4632, miR‑4742, miR‑4754, miR‑7850
Controls: U6 (Positive), Scr (Negative)
Standard Stain: Hematoxylin & Eosin (HE)
Data Format & Organization
Each high-resolution image (.jpg) has a corresponding pixel-level segmentation mask (.png) delineating the prostate epithelium, suitable for training deep learning models. Segmentation masks are single-channel images where pixel value 0 indicates background and pixel value 255 indicates epithelium.
The data is organized first by marker, then by tissue type/origin. Within each marker's top-level folder (e.g., HE/, 550A/), the structure is:
[Marker]/
├── Normal/
│ ├── Normal_TURP_BPH/ (Images/masks from BPH patients)
│ └── Normal_Prostatectomy/ (Normal core 'a' images/masks from PCa patients)
└── Cancer/
└── Cancer_Prostatectomy/ (Gleason 3/4 core 'b'/'c' images/masks from PCa patients)
All image and mask files are located directly within the innermost folders (Normal_TURP_BPH, Normal_Prostatectomy, Cancer_Prostatectomy). Filenames encode marker, patient type, anonymized ID, sample number, core type, and optional experimental details.
Terms of Use
Distributed under a CC0 license for open research and development.