Li Panpan, Lv Yan, Shang Haiyan
Department of Pulmonary and Critical Care Medicine, the Sixth Medical Center of PLA General Hospital, Beijing 100048, China.
Department of Pulmonary and Critical Care Medicine, the Fourth Medical Center of PLA General Hospital, Beijing100048, China.
Bioimpacts. 2024 Dec 4;15:30640. doi: 10.34172/bi.30640. eCollection 2025.
In recent years, advancements in information and communication technology (ICT) and the internet of things (IoT) have revolutionized the healthcare industry, enabling the collection, analysis, and utilization of medical data to improved patient care. One critical area of focus is the development of predictive care systems for early diagnosis and treatment of cancer and disease.
Leveraging medical IoT data, this study proposes a novel approach based on transformer model for disease diagnosis. In this paper, features are first extracted from IoT images using a transformer network. The network utilizes a convolutional neural network (CNN) in the encoder part to extract suitable features and employs decoder layers along with attention mechanisms in the decoder part. In the next step, considering that the extracted features have high dimensions and many of these features are irrelevant and redundant, relevant features are selected using the Harris hawk optimization algorithm.
Various classifiers are used to label the input data. The proposed method is evaluated using a dataset consisting of 5 classes for testing and evaluation, and all results are provided into tables and plots.
The experimental results demonstrate that the proposed method acceptable performance compared to other methods.
近年来，信息通信技术（ICT）和物联网（IoT）的进步彻底改变了医疗行业，使得医疗数据的收集、分析和利用能够改善患者护理。一个关键的关注领域是开发用于癌症和疾病早期诊断与治疗的预测护理系统。
本研究利用医疗物联网数据，提出了一种基于变压器模型的疾病诊断新方法。在本文中，首先使用变压器网络从物联网图像中提取特征。该网络在编码器部分利用卷积神经网络（CNN）提取合适的特征，并在解码器部分采用解码器层以及注意力机制。下一步，考虑到提取的特征具有高维度且其中许多特征是不相关和冗余的，使用哈里斯鹰优化算法选择相关特征。
使用各种分类器对输入数据进行标记。所提出的方法使用一个由5个类别组成的数据集进行测试和评估，所有结果都以表格和图表形式呈现。
实验结果表明，与其他方法相比，所提出的方法具有可接受的性能。
