基于神经网络迁移学习在皮肤病类别检测中的应用

Application of neural network transfer learning in dermatological category detection

目的了解基于神经网络迁移学习应用在皮肤病类别检测中的作用。方法通过卷积神经网络迁移学习ResNet50和InceptionV3模型构建能够有效区分黑色素瘤、黑色素瘤和脂溢性角化病3类皮肤病变的图像识别算法。结果该皮肤病类别检测应用算法在测试数据集上的综合识别准确率达到85%(510/600),黑色素瘤的识别准确率高达95%(111/117)。结论实验表明卷积神经网络模型在医学图像识别领域具备潜力,可有效辅助临床医师诊断识别,具有重要的临床意义。

Objective To investigate the role of network-based on neural in the detection of dermatological categories.Methods An image recognition algorithm capable of effectively distinguishing between melanoma,common moles,and seborrheic keratosis,which are three categories of dermatological conditions,was constructed using convolutional neural network transfer learning with the ResNet50 and InceptionV3 models.Results The comprehensive recognition accuracy of this dermatological category detection algorithm on the test dataset reached 85%(510/600),with a notably high recognition accuracy of 95%(111/117)for melanoma.Conclusion Experimental results demonstrate the substantial potential of convolutional neural network models in the field of medical image recognition.They can effectively assist clinical physicians in diagnosis and recognition,holding significant clinical implications.