医学图像分类论文:Deep Learning for Identifying Metastatic Breast Cancer

医学图像分类论文:Deep Learning for Identifying Metastatic Breast Cancer

Auther: Dayong Wang, Institution: Harvard Medical School, Publish year: 2016

论文地址：https://arxiv.org/abs/1606.05718v1

Key point:

ISBI challenge: 淋巴结活检图像中转移性乳腺癌的自动检测，体现了深度学习在自动诊断中的应用
将原始图像划分为小的patch，使用patch进行分类训练。然后将patch级的预测结果合成肿瘤的概率热图，从而实现肿瘤定位和肿瘤分类

dataset and metrics:

dataset: 400 whole side images, 270 for train and 130 for test
metrics: AUC, 敏感度

method:

1. image pre-processing: 将背景分割，只保留组织