Classification and Retrieval of Digital Pathology Scans: A New Dataset

TLDR

The authors introduce Kimia Path24, a new dataset of whole‑scan pathology images for classification and retrieval, and propose a compound patch‑and‑scan accuracy metric to challenge high‑accuracy performance. They generate 1,325 test patches from 24 tissue textures, allow training sets of 27,000–50,000 patches, and benchmark LBP, dictionary, and CNN methods on the dataset. CNNs achieve the best performance with a 41.80 % accuracy.

Abstract

In this paper, we introduce a new dataset, Kimia Path24, for image classification and retrieval in digital pathology. We use the whole scan images of 24 different tissue textures to generate 1,325 test patches of size 1000x1000 (0.5mm x 0.5mm). Training data can be generated according to preferences of algorithm designer and can range from approximately 27,000 to over 50,000 patches if the preset parameters are adopted. We propose a compound patch-and-scan accuracy measurement that makes achieving high accuracies quite challenging. In addition, we set the benchmarking line by applying LBP, dictionary approach and convolutional neural nets (CNNs) and report their results. The highest accuracy was 41.80% for CNN.