基于组合BP神经网络的皮肤肿瘤目标识别  

Recognition of skin tumor object based on BP neural network ensemble

作　　者：谢凤英[1] 李阳[1] 姜志国[1] 孟如松[2] 

机构地区：[1]北京航空航天大学图像处理中心,北京100191 [2]中国人民解放军空军总医院,北京100142

出　　处：《中国体视学与图像分析》2015年第1期16-21,共6页Chinese Journal of Stereology and Image Analysis

基　　金：国家自然科学基金(61471016;61371134;61271436)

摘　　要：本文对皮肤肿瘤目标识别技术进行研究。首先利用阈值分割方法对皮损区域进行分割;然后,依据皮肤肿瘤早期诊断ABCD准则,对皮损区域提取了颜色、纹理和形状等特征,并基于相关性分析对所提取的特征进行优选;最后采用组合BP神经网络模型实现了皮肤肿瘤目标的分类识别。本文方法在黄色人种皮肤镜图像上进行实验,结果表明,该方法具有更高的分类精度,敏感度和特异度分别达到了93.3%和96.7%,识别结果令人满意。Recognition of skin tumor object is studied in this paper. Firstly, dermoscopic images are seg- mented by threshold method. Secondly, the features including color, texture and shape are extracted based on ABCD rule which differentiates the benign from the malignant through analyzing the Asymmetry, Border irregularity, Color variation and Different tained based on correlation analysis. Finally, the structures of a lesion. Then the optimal features are oh- neural network ensemble. The proposed method is lesion object is classified using back propagation （BP） tested on the images from yellow race, and experimen-tal results show that the recognition method has a higher accuracy, and the sensitivity and specificity may come up to 93.3% and 96.7%, respectively.

关 键 词：皮肤镜图像 BP神经网络 皮肤肿瘤 识别分类 

分 类 号：R739.5[医药卫生—肿瘤]