基于EfficientNet的皮肤癌识别与分类
2021-11-04张嘉颖
DOI:10.19850/j.cnki.2096-4706.2021.09.004
摘 要:针对人工识别皮肤癌恶性肿瘤中不可避免的人为因素,以及效率低、设备要求高等问题,提出了一种基于EfficientNet网络的新的皮肤癌识别与分类方法。首先,由于样本数据量过小,通过数据预处理实现数据增强,从而防止训练模型出现过拟合的问题。然后将数据集在EfficientNet网络模型上进行训练,同时采用Adam调整学习率,进而实现皮肤癌图像的识别与分类。实验结果表明,该模型的准确率和查全率可分别达到90.78%和88.23%,在保证了准确率和查全率的前提下,参数量大大减少,可有效提升临床医学诊断的效率。
关键词:EfficientNet模型;Adam;皮肤癌识别
中图分类号:TP391.4 文献标识码:A 文章编号:2096-4706(2021)09-0013-03
Skin Cancer Identification and Classification Based on EfficientNet
ZHANG Jiaying
(College of Computer Science and Technology,Huaqiao University,Xiamen 361021,China)
Abstract:A new method for skin cancer identification and classification based on the EfficientNet network is proposed,aiming at the inevitable human factors,low efficiency and high equipment requirements in the manual identification of skin cancer and malignant tumors. Firstly,since the sample data is too small,data enhancement is realized through data preprocessing,so as to prevent the problem of overfitting in the training model. Then,the data set is trained on its EfficientNet network model,with Adam adjusting the learning rate for skin cancer image identification and classification. The experimental results show that the accuracy and recall of the model can reach 90.78% and 88.23%,respectively. On the premise of ensuring the accuracy and recall,the number of parameters is greatly reduced,which can effectively improve the efficiency of clinical medical diagnosis.
Keywords:EfficientNet model;Adam;skin cancer identification
0 引 言
皮膚癌是最常见的癌症类型,常见的皮肤癌有基底细胞癌(Basal Cell Carcinoma,BCC)、鳞状细胞癌(Squamous Cell Carcinoma,SCC)、恶性黑色素瘤(Melanoma,ML)等[1]。2017年,仅美国新确诊ML患者就有85 686人,其中8 056人因此死亡,ML患病率已在所有癌症类型中排第六[2]。ML虽然只占所有皮肤癌的2%,但却最易导致死亡,BCC是最常见的皮肤癌类型,发展速度慢但具有局部侵袭性,SCC约占20%至30%,仅次于BCC[3]。目前常用的皮肤癌检测方法有皮肤活体组织检查和病理组织学检查,即使借助皮肤镜图像分析、共焦显微镜等可帮助医生更好地诊断,诊断结果仍然会受到主观因素的影响,不能保证诊断的准确率[4]。况且这种诊断只适用于一小部分掌握特征识别技巧的专业人士,不具有普适性,因此不能够帮助人们及时地识别出皮肤癌并且尽早地治疗皮肤癌。
为了减少模型训练参数,同时又能够……
