基于颜色自相关图的乳腺肿瘤良恶性分类

Classification of benign and malignant breast tumors based on color auto-correlogram

目的:构建一种乳腺肿瘤良恶性分类模型,使医生得到更加客观、准确的诊断结果。方法:借助BreaKHis数据集,提取乳腺肿瘤病理图像颜色自相关图的64维特征,利用k-NN分类器构建乳腺肿瘤良恶性分类模型,并对乳腺肿瘤良恶性进行分类。结果:颜色自相关图中像素空间距离d=1时分类精度最高,准确度平均达到87.01%,灵敏度平均达到88.52%,特异度平均达到85.49%。结论:该模型为乳腺肿瘤良恶性分类提供了一种新型的检测手段,可有效提高乳腺肿瘤良恶性临床诊断的准确率。

Objective To propose a classification model for benign and malignant breast tumors to improve their diagnoses. Methods The 64-dimension color auto-correlogram features of the breast tumor pathological images were extracted with BreaKHis dataset. A classification model was constructed by using k-NN classifier, and then was used for the classification of benign and malignant breast tumors. Results In the color auto-correlogram, the highest classification accuracy was obtained when the pixel space distance d was equal to 1. The mean values of the accuracy, sensitivity and specialty were 87.01%, 88.52% and 85.49% respectively. Conclusion The model provides a new type of detection method for the benign and malignant classification of breast tumors and improves effectively the accuracy of clinical diagnosis.