基于HOG特征和滑动窗口的乳腺病理图像细胞检测

Nuclei detection of breast histopathology based on HOG feature and sliding window

提出一种基于方向梯度直方图(histograms of oriented gradient,HOG)特征和滑动窗口的细胞检测方法,能快速、高效、准确地检测高分辨率病理组织图像中的细胞。该检测算法首先对训练集中的细胞样本块和非细胞样本块提取HOG特征,然后运用HOG特征训练分类器。训练好的分类器用于在整幅病理图像中自动检测细胞。先运用滑动窗的方法在整幅高分辨率病理图像中选取相同尺寸的所有可能的细胞块,被滑动窗选定的图像块提取HOG特征后,送到训练好的分类器中判断是否是细胞块。为了验证提出方法的有效性,将此方法运用于17名乳腺患者的共37张H&E(hematoxylin&eosin)染色高分辨率穿刺切片病理图像上自动检测细胞,通过与softmax(SM)分类器、稀疏自编码器+SM、局部二值模式+SM、支持向量机(support vector machine,SVM)、HOG+SVM、以及HOG+SVM多个模型对细胞检测的准确率、召回率以及综合评价指标的对比表明,本研究提出的方法分别为71.5%,82.3%和76.5%,具有更高的准确率。

A new method was presented which integrated histograms of oriented gradient （HOG） feature and sliding window for rapid, efficient and accurate detection of nuclei from high resolution pathological images. HOG feature was extracted from the training samples which include both nuclei and non-nuclei patches. The supervised classifier were trained with HOG features. The trained classifier was employed for automated nuclei detection from input patches that selected from histopathological images. During the detection, sliding window was used to select patches. In order to verify the effectiveness of the method on detecting nuclei from histopathological images, this article compared the pro- posed method with softmax （SM） classifier, sparse autoencoder（SAE） ＋ SM, local binary pattern （LBP） ＋ SM, sup- port vector machine（ SVM）, HOG ＋ SM, and HOG ＋ SVM models. The experiments on 37 pieces of H＆E staining his- topathological images showed that the proposed method achieved highest precision, recall and F1-measure values, which were 71.5%, 82. 3% and 76. 5% respectively.