眼部疾病在初期可能被忽视,导致很少有人及时就医检查,从而延误疾病的诊断与治疗。为应对这一问题,将眼部疾病检测系统部署到智能手机中,使人们随时能够通过拍照来进行眼部检查,从而及早发现病情并防止其恶化。本研究旨在利用眼部特征...眼部疾病在初期可能被忽视,导致很少有人及时就医检查,从而延误疾病的诊断与治疗。为应对这一问题,将眼部疾病检测系统部署到智能手机中,使人们随时能够通过拍照来进行眼部检查,从而及早发现病情并防止其恶化。本研究旨在利用眼部特征数据集训练神经网络模型,并将其部署到智能手机设备上,以实现对白内障、葡萄膜炎、甲亢突眼这三种眼部疾病以及正常眼睛图像的识别分类。本文通过将深度学习技术与智能手机相结合,提出了针对眼部疾病检测的模型MobileEDT。该模型利用卷积神经网络提取病灶特征,同时结合Transformer的自注意力机制,实现了模型的轻量化。实验结果表明,所提出的眼部疾病检测模型在准确性(Accuracy为0.999)和效率方面均表现出较高性能。Eye diseases may be overlooked in their early stages, resulting in few people seeking timely medical checkups, which delays the diagnosis and treatment of the disease. To counteract this problem, eye disease detection systems are deployed into smartphones to enable people to perform eye examinations by taking pictures at any time, thus detecting the condition early and preventing its deterioration. The aim of this study is to train a neural network model using an eye feature dataset and deploy it to a smartphone device for recognizing and classifying three eye diseases, namely cataract, uveitis, and eye protrusion due to hyperthyroidism, as well as normal eye images. In this paper, we propose MobileEDT, a model for eye disease detection, by combining deep learning techniques with smartphones. The model utilizes convolutional neural networks to extract lesion features, while incorporating Transformer’s self-attention mechanism to achieve a lightweight model. Experimental results show that the proposed eye disease detection model exhibits high performance in terms of both accuracy (Accuracy of 0.999) and efficiency.展开更多
文摘眼部疾病在初期可能被忽视,导致很少有人及时就医检查,从而延误疾病的诊断与治疗。为应对这一问题,将眼部疾病检测系统部署到智能手机中,使人们随时能够通过拍照来进行眼部检查,从而及早发现病情并防止其恶化。本研究旨在利用眼部特征数据集训练神经网络模型,并将其部署到智能手机设备上,以实现对白内障、葡萄膜炎、甲亢突眼这三种眼部疾病以及正常眼睛图像的识别分类。本文通过将深度学习技术与智能手机相结合,提出了针对眼部疾病检测的模型MobileEDT。该模型利用卷积神经网络提取病灶特征,同时结合Transformer的自注意力机制,实现了模型的轻量化。实验结果表明,所提出的眼部疾病检测模型在准确性(Accuracy为0.999)和效率方面均表现出较高性能。Eye diseases may be overlooked in their early stages, resulting in few people seeking timely medical checkups, which delays the diagnosis and treatment of the disease. To counteract this problem, eye disease detection systems are deployed into smartphones to enable people to perform eye examinations by taking pictures at any time, thus detecting the condition early and preventing its deterioration. The aim of this study is to train a neural network model using an eye feature dataset and deploy it to a smartphone device for recognizing and classifying three eye diseases, namely cataract, uveitis, and eye protrusion due to hyperthyroidism, as well as normal eye images. In this paper, we propose MobileEDT, a model for eye disease detection, by combining deep learning techniques with smartphones. The model utilizes convolutional neural networks to extract lesion features, while incorporating Transformer’s self-attention mechanism to achieve a lightweight model. Experimental results show that the proposed eye disease detection model exhibits high performance in terms of both accuracy (Accuracy of 0.999) and efficiency.
