基于深度学习与迁移学习的中医舌象提取识别研究
|
摘要
目的探讨基于Faster R-CNN算法的深度学习与迁移学习技术在中医齿痕舌与裂纹舌局部特征提取识别中的应用效果。方法将500例受试者舌象图片使用LabelImg V 1. 6软件进行齿痕舌与裂纹舌区域标注,划分训练数据集与测试数据集,通过Faster R-CNN的深度学习技术与迁移学习技术中的微调(finetune)方法构建模型,最后使用准确率、精确率、召回率对模型效果进行评价。结果对于裂纹舌,模型识别真阳性99例,假阴性3例,真阴性45例,假阳性3例,准确率0. 960、精确率0. 971、召回率0. 971。对于齿痕舌,模型识别真阳性60例,假阴性20例,真阴性69例,假阳性1例,准确率0. 860、精确率为0. 983、召回率0. 750。模型宏观准确率0. 910、宏观精确率0. 977、宏观召回率0. 860。图像识别结果显示,模型不受舌象中病理变化所在位置的影响,对舌象局部特征提取方面也具有较强的适应性特点。结论基于深度学习与迁移学习的方法,可以较好地完成中医舌象局部特征辨识任务,具有较好的迁移能力。
Objective To explore the application effect of deep learning and migration learning technology based on Faster R-CNN algorithm in local feature extraction and recognition of teeth-printed tongue and fissured tongue in traditional Chinese medicine( TCM). Methods The tongue images of 500 subjects were labeled in the areas of teeth-printed tongue and fissured tongue with the LabelImg V 1. 6 software,and the training data set and test data set were divided. The deep learning technique of Faster R-CNN and the fine-tune in the migration learning technology were used to construct the model,and finally the precision rate,the accuracy rate,and the recall rate were used to evaluate the model effect. Results For the fissured tongue,the model identified 99 cases of true positive,3 cases of false negative,45 cases of true negative,3 cases of false positive. The precision rate was 0. 960,the accuracy rate was 0. 971,and the recall rate was 0. 971. For the teeth-printed tongue,the model identified true positive 60 cases,and false negative 20 cases,true negative 69 cases,and false positive 1 case. The precision rate was 0. 860,the accuracy rate was 0. 983,and the recall rate was 0. 750. The model has a macro precision rate of 0. 910,a macro accuracy rate of 0. 977,and a macro recall rate of 0. 860. The image recognition results showed that the model was not affected by the location of the pathological changes in the tongue image,and it also had strong adaptability to the local feature extraction of the tongue image. Conclusion Based on the method of deep learning and migration learning,the local feature recognition task of TCM tongue image can be completed well and has good migration ability.
Objective To explore the application effect of deep learning and migration learning technology based on Faster R-CNN algorithm in local feature extraction and recognition of teeth-printed tongue and fissured tongue in traditional Chinese medicine( TCM). Methods The tongue images of 500 subjects were labeled in the areas of teeth-printed tongue and fissured tongue with the LabelImg V 1. 6 software,and the training data set and test data set were divided. The deep learning technique of Faster R-CNN and the fine-tune in the migration learning technology were used to construct the model,and finally the precision rate,the accuracy rate,and the recall rate were used to evaluate the model effect. Results For the fissured tongue,the model identified 99 cases of true positive,3 cases of false negative,45 cases of true negative,3 cases of false positive. The precision rate was 0. 960,the accuracy rate was 0. 971,and the recall rate was 0. 971. For the teeth-printed tongue,the model identified true positive 60 cases,and false negative 20 cases,true negative 69 cases,and false positive 1 case. The precision rate was 0. 860,the accuracy rate was 0. 983,and the recall rate was 0. 750. The model has a macro precision rate of 0. 910,a macro accuracy rate of 0. 977,and a macro recall rate of 0. 860. The image recognition results showed that the model was not affected by the location of the pathological changes in the tongue image,and it also had strong adaptability to the local feature extraction of the tongue image. Conclusion Based on the method of deep learning and migration learning,the local feature recognition task of TCM tongue image can be completed well and has good migration ability.
引文
[1]马旭翔,李勇枝,许家佗.舌诊信息化技术在证候诊断及疗效评价中的应用[J].中国中医基础医学杂志,2018,24(12):1716-1719.
[2]汤一平,王丽冉,何霞,等.基于多任务卷积神经网络的舌象分类研究[J].计算机科学,2018,45(12):255-261.
[3]李晓宇,张新峰,沈兰荪.基于支撑向量机的中医舌色苔色识别算法研究[J].北京生物医学工程,2006,25(1):43-46.
[4]许家佗,周昌乐,方肇勤,等.舌像颜色特征的计算机分析与识别研究[J].上海中医药大学学报,2004,18(3):43-47.
[5]屠立平,许家佗,张志枫,等.一种自然光图像颜色校正方法在舌色图像校正中的应用[J].南京中医药大学学报,2011,27(1):15-18.
[6]OBAFEMI-AJAYI T,KANAWONG R,XU D,et al.Features for automated tongue image shape classification[C]//IEEE International Conference on Bioinformatics&Biomedicine Workshops.IEEE Computer Society,2012:273-279.doi:10.1109/BIBMW.2012.6470316.
[7]黄景斌,许家佗,张志枫,等.基于舌象图像分区指标的糖尿病疗效评价[J].中华中医药杂志,2017,32(3):1148-1151.
[8]OREN ZK,BEN TG,JIMMY B,et al.Automated analysis of high-content microscopy data with deep learning[J].Mol Syst Biol,2017,13(4):924.
[9]KANDASWAMY C,SILVA LM,ALEXANDRE LA,et al.High-content analysis of breast cancer using single-cell deep transfer learning[J].J Biomol Screen,2016,21(3):252-259.
[10]REN S,HE K,GIRSHICK R,et al.Faster R-CNN:Towards real-time object detection with region proposal networks[J].IEEE T PATTERN ANAL,2017,39(6):1137-1149.
[11]ZEILER MD,FERGUS R.Visualizing and understanding convolutional networks[C]//European Conference on Computer Vision.Springer International Publishing,2013:818-833.doi:10.1007/978-3-319-10590-1_53.
[12]JIA Y,SHELHAMER E,DONAHUE J,et al.Caffe:convolutional architecture for fast feature embedding[C]//Proceeding of the 22nd Acm International Conference on Multimedia.ACM,2014:675-678.doi:10.1145/2647868.2654889.
[13]李峰,董昌武.中医诊断学[M].北京:科学出版社,2018:42-68.
[14]朱虹,翟超,吕志,等.基于Faster-RCNN的智能家居行人检测系统设计与实现[J].工业控制计算机,2018,31(4):68-70.
[15]王黎,陆慧娟,叶敏超,等.Faster-RCNN的癌症影像检测方法[J].中国计量大学学报,2018,29(2):136-141.
[16]桑军,郭沛,项志立,等.Faster-RCNN的车型识别分析[J].重庆大学学报,2017,40(7):32-36.
[17]史凯静,鲍泓,徐冰心,等.基于Faster RCNN的智能车道路前方车辆检测方法[J].计算机工程,2018,44(7):36-41.
[18]PAN SJ,YANG Q.A survey on transfer learning[J].IEEE T KNOWL DATA EN,2010,22(10):1345-1359.
[19]周曼,刘志勇,陈梦迟,等.基于AlexNet的迁移学习在流程工业图像识别中的应用[J].工业控制计算机,2018,31(11):80-82.
[20]WANG G,LI W,ZULUAGA MA,et al.Interactive medical image segmentation using deep learning with image-specific fine tuning[J].IEEE Trans Med Imaging,2018,37(7):1562-1573.
[21]王忆勤.中医诊断技术发展及四诊信息融合研究[J].上海中医药大学学报,2019,33(1):1-7.
[2]汤一平,王丽冉,何霞,等.基于多任务卷积神经网络的舌象分类研究[J].计算机科学,2018,45(12):255-261.
[3]李晓宇,张新峰,沈兰荪.基于支撑向量机的中医舌色苔色识别算法研究[J].北京生物医学工程,2006,25(1):43-46.
[4]许家佗,周昌乐,方肇勤,等.舌像颜色特征的计算机分析与识别研究[J].上海中医药大学学报,2004,18(3):43-47.
[5]屠立平,许家佗,张志枫,等.一种自然光图像颜色校正方法在舌色图像校正中的应用[J].南京中医药大学学报,2011,27(1):15-18.
[6]OBAFEMI-AJAYI T,KANAWONG R,XU D,et al.Features for automated tongue image shape classification[C]//IEEE International Conference on Bioinformatics&Biomedicine Workshops.IEEE Computer Society,2012:273-279.doi:10.1109/BIBMW.2012.6470316.
[7]黄景斌,许家佗,张志枫,等.基于舌象图像分区指标的糖尿病疗效评价[J].中华中医药杂志,2017,32(3):1148-1151.
[8]OREN ZK,BEN TG,JIMMY B,et al.Automated analysis of high-content microscopy data with deep learning[J].Mol Syst Biol,2017,13(4):924.
[9]KANDASWAMY C,SILVA LM,ALEXANDRE LA,et al.High-content analysis of breast cancer using single-cell deep transfer learning[J].J Biomol Screen,2016,21(3):252-259.
[10]REN S,HE K,GIRSHICK R,et al.Faster R-CNN:Towards real-time object detection with region proposal networks[J].IEEE T PATTERN ANAL,2017,39(6):1137-1149.
[11]ZEILER MD,FERGUS R.Visualizing and understanding convolutional networks[C]//European Conference on Computer Vision.Springer International Publishing,2013:818-833.doi:10.1007/978-3-319-10590-1_53.
[12]JIA Y,SHELHAMER E,DONAHUE J,et al.Caffe:convolutional architecture for fast feature embedding[C]//Proceeding of the 22nd Acm International Conference on Multimedia.ACM,2014:675-678.doi:10.1145/2647868.2654889.
[13]李峰,董昌武.中医诊断学[M].北京:科学出版社,2018:42-68.
[14]朱虹,翟超,吕志,等.基于Faster-RCNN的智能家居行人检测系统设计与实现[J].工业控制计算机,2018,31(4):68-70.
[15]王黎,陆慧娟,叶敏超,等.Faster-RCNN的癌症影像检测方法[J].中国计量大学学报,2018,29(2):136-141.
[16]桑军,郭沛,项志立,等.Faster-RCNN的车型识别分析[J].重庆大学学报,2017,40(7):32-36.
[17]史凯静,鲍泓,徐冰心,等.基于Faster RCNN的智能车道路前方车辆检测方法[J].计算机工程,2018,44(7):36-41.
[18]PAN SJ,YANG Q.A survey on transfer learning[J].IEEE T KNOWL DATA EN,2010,22(10):1345-1359.
[19]周曼,刘志勇,陈梦迟,等.基于AlexNet的迁移学习在流程工业图像识别中的应用[J].工业控制计算机,2018,31(11):80-82.
[20]WANG G,LI W,ZULUAGA MA,et al.Interactive medical image segmentation using deep learning with image-specific fine tuning[J].IEEE Trans Med Imaging,2018,37(7):1562-1573.
[21]王忆勤.中医诊断技术发展及四诊信息融合研究[J].上海中医药大学学报,2019,33(1):1-7.