Published August 29, 2022 | Version 0.1
Dataset Open

HER2 overexpression in gastroesophageal adenocarcinoma from immunohistochemstry imaging.

  • 1. Data science of Bioimages Lab, University of Cologne
  • 2. Department of General, Visceral, Cancer and Transplantation Surgery, University of Cologne
  • 3. Faculty of Mathematics and Computer Science, Friedrich Schiller University Jena
  • 4. Institute of Pathology, University of Cologne
  • 5. Department of Radiology, University of Cologne
  • 6. Research Unit Analytical Pathology, Helmholtz Zentrum München
  • 7. Data science of Bioimages Lab, University of Cologne; gCologne Excellence Cluster on Cellular Stress Responses in Aging-Associated Diseases (CECAD), University of Cologne

Description

HER2 overexpression in gastroesophageal adenocarcinoma from immunohistochemstry imaging.

Primary dataset used in Predicting the HER2 status in esophageal cancer from tissue microarrays using convolutional neural networks [1].

This dataset is composed by HER2 immunohistochemistry images of gastroesophageal adenocarcinoma patients. We provide a train-test split of .jpg image files of individual tissue cores, which were labeled with an immunohistochemstry score (ranging from 0 to 3) and a HER2 overexpression status (positive or negative).

 

Detailed information

Train split comes from a multi-spot tissue microarray (TMA) with 165 tumor cases and a single-spot TMA with 428 tumor cases, as described elsewhere [2]. Test split comes from an independent single-spot TMA with 307 tumor cases as the test dataset. The test set consisted of tumor cases that occurred at a later time point compared to the training set cases. This dataset construction strategy mimics how such a model would be developed and deployed in a clinical routine. Coincidentally, the test set does not include tumor cases with an IHC score of 1. The multi-spot TMA was composed of eight tissue cores (1.2 mm diameter) of each tumor - four cores punched on the tumor margin and four in the tumor center. To construct the single-spot TMA, one tissue core per patient from the tumor center was punched. The cores were transferred to TMA receiver blocks. Each TMA block contained 72 tissue cores. Subsequently, 4 µm-thick sections from the TMA blocks were prepared and transferred to an adhesive-coated slide system (Instrumedics Inc., Hackensack, NJ).

We used a HER2 antibody (Ventana clone 4B5, Roche Diagnostics, Rotkreuz, Switzerland) on the automated Ventana/Roche slide stainer to perform immunohistochemistry (IHC) on the TMA slides. HER2 expression in carcinoma cells was assessed according to staining criteria listed in the Supplemental Table 1 of [1]. Scores 0 and 1 indicated negative HER2 status, and score 3 indicated positive HER2 status. Immunohistochemical expression evaluation was assessed manually by two pathologists according to [3]. Discrepant results, which occurred only in a small number of samples, were resolved by consensus review. Spots with a score of 2 were analyzed by fluorescence in situ hybridization (ISH) to resolve the HER2 status. The ISH analysis evaluated the HER2 gene amplification status using the Zytolight SPEC ERBB2/CEN 17 Dual Probe Kit (Zytomed Systems GmbH, Germany) according to the manufacturer's protocol. A fluorescence microscope (DM5500, Leica, Wetzlar, Germany) with a 63× objective was used for scanning the tumor tissue for amplification hotspots. We counted the signals in randomly chosen areas of homogeneously distributed signals. Twenty tumor cells were evaluated by counting green HER2 and orange centromere-17 (CEN17) signals. The reading strategy followed the recommendations of HER2/CEN17 ratio ≥ 2.0 or HER2 signals ≥ 6.0 for HER2 positive and a HER2/CEN17 ratio < 2.0 for HER2 negative samples.

Slides were digitised with a slide scanner (NanoZoomer S360, Hamamatsu Photonics, Japan) with 40-times magnification and used QuPath's [4] TMA dearrayer to slice the digitized slides into individual images.

 

[1] Pisula JI, Datta RR, Boerner-Valdez L, Avemarg JR, Jung JO, Plum P, Loeser H, Lohneis P, Meuschke M, dos Santos DP, Gebauer F, et al. Predicting the HER2 status in esophageal cancer from tissue microarrays using convolutional neural networks. bioRxiv: https://www.biorxiv.org/content/10.1101/2022.05.13.491769v1

[2] Plum PS, Gebauer F, Krämer M, Alakus H, Berlth F, Chon SH, Schiffmann L, Zander T, Büttner R, Hölscher AH, Bruns CJ. HER2/neu (ERBB2) expression and gene amplification correlates with better survival in esophageal adenocarcinoma. BMC cancer. 2019 Dec;19(1):1-9.

[3] Lordick F, Al-Batran SE, Dietel M, Gaiser T, Hofheinz RD, Kirchner T, Kreipe HH, Lorenzen S, Möhler M, Quaas A, Röcken C. HER2 testing in gastric cancer: results of a German expert meeting. Journal of cancer research and clinical oncology. 2017 May;143(5):835-41.

[4] Bankhead P, Loughrey MB, Fernández JA, Dombrowski Y, McArt DG, Dunne PD, McQuaid S, Gray RT, Murray LJ, Coleman HG, James JA. QuPath: Open source software for digital pathology image analysis. Scientific reports. 2017 Dec 4;7(1):1-7.

Files

test.csv

Files (7.6 GB)

Name Size Download all
md5:96c89103376fa4b5291629ead6c70f22
6.8 kB Preview Download
md5:0c7b5ddbf226352f8876f2bf14feae88
2.0 GB Preview Download
md5:bcd9c8217c3ed97fee91afdeceb692a7
75.1 kB Preview Download
md5:9e8dc5401ce9a16ac6db97536384206c
5.6 GB Preview Download