UNITOPATHO

我们介绍了UniToPatho，这是从292的全幻灯片图像中提取的9536苏木精和曙红染色斑块的注释数据集，旨在训练用于结直肠息肉分类和腺瘤分级的深度神经网络。通过Hamamatsu纳米动物学S210扫描仪以20倍放大 (0.4415 μ m/px) 获取载玻片。每张幻灯片属于不同的患者，并由专家病理学家根据以下六类进行注释:

正常组织；
HP增生性息肉；
TA.HG-管状腺瘤，高度不典型增生；
TA.LG-管状腺瘤，低度不典型增生；
TVA。HG-肾小管绒毛状腺瘤，高度不典型增生；
TVA。LG-肾小管绒毛状腺瘤，低度不典型增生。

We present UniToPatho, an annotated dataset comprising 9536 hematoxylin and eosin-stained patches extracted from 292 whole-slide images, designed for training deep neural networks for colorectal polyp classification and adenoma grading. Slides were scanned at 20× magnification (0.4415 μm/px) using a Hamamatsu Nanozoomer S210 scanner. Each slide is from a distinct patient, and was annotated by expert pathologists into the following six categories:
1. Normal tissue
2. Hyperplastic polyp (HP)
3. Tubular adenoma with high-grade dysplasia (TA.HG)
4. Tubular adenoma with low-grade dysplasia (TA.LG)
5. Tubulovillous adenoma with high-grade dysplasia (TVA.HG)
6. Tubulovillous adenoma with low-grade dysplasia (TVA.LG)

搜集汇总

背景与挑战

背景概述

UNITOPATHO是一个医学图像数据集，包含从292张全幻灯片图像中提取的9536个苏木精和曙红染色斑块，专门用于训练深度神经网络进行结直肠息肉分类和腺瘤分级。数据集由专家病理学家根据六类病理类型进行注释，包括正常组织、增生性息肉以及不同级别的管状和肾小管绒毛状腺瘤，适用于医学人工智能研究。