Data record for the article: How Can Artificial Intelligence Models Assist PD-L1 Expression Scoring in Breast Cancer: Results of Multi-institutional Ring Studies

This record provides details of the data supporting the claims of the related article: “How Can Artificial Intelligence Models Assist PD-L1 Expression Scoring in Breast Cancer: Results of Multi-institutional Ring Studies”.The related study aimed to assess the effectiveness of an artificial intelligence-assisted model in the histopathological analysis of Programmed death-ligand 1 (PD-L1) expression of tumor-infiltrating immune cells.<br>Type of data: histology image files; spreadsheet data filesSubject of data: Formalin-fixed, paraffin-embedded tumor resection blocks from patients with invasive breast cancerSample size: 100 samplesPopulation characteristics: Among the 100 patients, 39 patients under 50 years old, 61 patients over 50 years old. 97 patients were diagnosed with Invasive carcinoma of no special type, 2 were diagnosed with invasive lobular carcinoma and 1 was diagnosed with meta-plastic. 3 cases were classified as grade 1, 37 cases as grade 2 and 60 cases as grade 3. carcinoma. 28 cases were clinical stage 1, 57 cases were stage 2 and 15 cases were stage 3.Recruitment: 100 patients were surgically diagnosed as invasive breast cancer in the fourth hospital of Hebei Medical University. All patients did not receive any treatment before operation. One hundred tumor resection samples (formalin-fixed, paraffin-embedded blocks) were collected from the 100 patients.Date of data collection: January to June 2019Geographic location: Hebei<br><b>Data access</b>All data files associated with this study are openly available as part of this figshare data record:<br><i><b>manual immune cell scoring data_fig1.xlsx </b></i>Excel spreadsheet containing full immune cell scoring data from 31 pathologists for 109 images in three ring studies.<br><i><b>data_fig2.zip</b></i>This zip folder contains 3 subfolders, with .png files supporting article figure 2. This folder additionally contains raw images and epithelium mask images from the Roche SP142 assay.<br><i><b>summary statistics of ring studies.xlsx</b></i>Excel spreadsheet with a summary of the raw scoring data for ‘all’ and each of the three levels of pathologists. This data underpins the box plots presented in figure 3 of the article.<br><b><i>scoring accuracy evalutions.zip</i></b>This zip folder contains 6 .xlsx files with the scoring accuracy data for 3 experiments in each test category; 2-class and 4-class. These data tables underpin the box plots presented in figure 4 of the article.<br><i><b>data_fig5.zip</b></i>This zip folder contains .out, .json and .npy files supporting figure 5 of the article. These data files provide the continuous and categorical AI scores, which are the main results being presented in this article.<br><i><b>Supplementary_Information.zip</b></i>This zip folder contains .png, .zip, .pptx and .xlsx files which make up the Supplementary Information PDF file available with the published article.<br><br><b>Corresponding authors for this study</b>Yueping Liu ([e-mail])Jianhua Yao ([e-mail])<br><b><br></b><b>Study approval </b>All tissues and data were retrieved under the permission of the institutional research ethics board of the Fourth Hospital of Hebei Medical University with the declaration number of 2020KY112.<br>