基于高分辨率卷积神经网络的皮肤常见肿瘤智能诊断模型构建
Construction of Intelligent Diagnosis Model for Common Skin Tumors Based on High Resolution Convolutional Neural Network
周兴雯 1马春驰 2王琳3
作者信息
- 1. 凉山彝族自治州第二人民医院皮肤科,四川 西昌 615000;四川大学华西医院皮肤科,四川 成都 610044
- 2. 成都理工大学环境与土木工程学院,四川 成都 610059
- 3. 四川大学华西医院皮肤科,四川 成都 610044
- 折叠
摘要
目的 探究高分辨率卷积神经网络(HRNetW32)模型在皮肤常见肿瘤临床诊断中的应用.方法 基于高分辨率特征提取皮肤常见肿瘤智能诊断模型,利用HRNetW32 模型,实现统一输入皮肤常见肿瘤皮肤镜图像,自动预测皮肤常见肿瘤类型的诊断结果;同时将构建的模型与VGG16、VGG19、ResNet34 等常见卷积神经网络模型进行对比分析.结果 HRNetW32 模型在训练集和验证集准确率分别为99.72%和95.00%,损失函数值分别为0.15 和0.21,表明所构建的模型能准确高效地提取皮肤常见肿瘤皮肤镜图像的高维特征.同时HRNetW32 模型表现出了优于VGG16、VGG19、ResNet34 模型的精确率、召回率、Micro F1 分数、灵敏度、特异度、真正率和假正率.结论 HRNetW32 模型可用于常见皮肤肿瘤筛检,且诊断准确率较高,具有较高临床诊断价值.
Abstract
Objective To explore the application of high-resolution convolutional neural network(HRNetW32)model in the clinical diagnosis of common skin tumors.Methods Propose an intelligent diagnosis model for common skin tumors based on high-resolution feature extraction,uses the HRNetW32 model to realize the unified input of dermoscopic images of common skin tumors and automatically predict the diagnostic results of common skin tumor types.At the same time,the constructed model is compared with common convolutional neural network models such as VGG16,VGG19,and ResNet34.Results The accuracy of the HRNetW32 model in the training set and the validation set were 99.72%and 95.00%,respectively,and the loss function values were 0.15 and 0.21,respectively,indicating that the constructed model could accurately and efficiently extract the high-dimensional features of dermoscopic images of common skin tumors,and the HRNetW32 model showed better precision,recall,MicroF1 score,sensitivity,specificity,true rate and false positive rate than VGG16,VGG19 and ResNet34 models.Conclusion The HRNetW32 model can be used for the screening of common skin tumors,and has high diagnostic accuracy and high clinical diagnostic value.
关键词
皮肤肿瘤诊断/图像识别/特征融合/高分辨率卷积神经网络Key words
skin tumor diagnosis/image recognition/feature fusion/high resolution convolutional neural network引用本文复制引用
基金项目
凉山州2021年度技术研究开发与推广应用项目(21ZDYF0106)
出版年
2024
国家科技期刊平台
NSTL
万方数据