基于卷积神经网络的皮肤黑色素瘤识别
首发时间:2019-03-20
摘要:皮肤黑色素瘤是一种恶性皮肤癌症,具有极高的致死率,而及早筛查诊断可大幅提升患者的存活率。皮肤黑色素瘤通过皮肤镜成像方式进行诊断,因此,设计准确、可靠的皮肤黑色素瘤计算机辅助诊断方法会极大地减轻皮肤科医生的工作量,提高诊断效率和诊断的客观性,从而帮助患者及早确诊治疗。近年来,深度卷积神经网络因其强大的自学习特征提取能力,在图像识别、分割、检测等领域得到广泛研究与应用。因此,本文研究基于深度卷积神经网络设计皮肤黑色素瘤自动识别算法,利用OHEM方法增强模型对困难样本的学习能力,设计基于多层感知机的多模型集成学习进一步提升识别准确度。最终,本文方法在黑色素瘤识别中取得0.9695的F1-Score值。
关键词: 黑色素瘤自动识别 卷积神经网络 困难样本学习 多模型集成
For information in English, please click here
Automatic recognition of skin melanoma based on convolutional neural network
Abstract:Melanoma is a kind of malignant skin cancer with extremely high mortality rate.Luckily, Screening and treatment can Largely improve patients\' survial rate in early stage. Melanoma is normaly diagnosed by dermascopic imaging. Therefore, A accurate and reliable melanoma computer-aided diagnosis methods can greatly reduce the wordload of dermatologists, improve the efficiency and objectivity of diagnosis. It is beneficial for patients in earlydiagnose.In this paper, the automatic recognition algorithm of melanoma based on convolutional neural network. OHEM is applied for enhancing the capacity of recognition difficult samples. Then, multi-model ensemble leaning based on multi-layer perception improves the recognition accuracy comparing to single model. Finally, we get a 0.9695 F1-Score in recognizing the melanoma.
Keywords: automatic recognition of melanoma convolutional neural network learning of hard example multi-model ensemble learning
基金:
引用
No.****
动态公开评议
共计0人参与
勘误表
基于卷积神经网络的皮肤黑色素瘤识别
评论
全部评论0/1000