Mo Pu-Chun, Hsu Hsiu-Yun, Lin Cheng-Feng, Cheng Yu-Shiuan, Tu I-Te, Kuo Li-Chieh, Su Fong-Chin
Department of Biomedical Engineering, College of Engineering, National Cheng Kung University, Tainan, Taiwan.
Department of Physical Medicine and Rehabilitation, National Cheng Kung University Hospital, College of Medicine, National Cheng Kung University, Tainan, Taiwan.
Front Bioeng Biotechnol. 2024 Feb 29;12:1351485. doi: 10.3389/fbioe.2024.1351485. eCollection 2024.
Diabetes mellitus and chronic kidney disease represent escalating global epidemics with comorbidities akin to neuropathies, resulting in various neuromuscular symptoms that impede daily performance. Interestingly, previous studies indicated differing sensorimotor functions within these conditions. If assessing sensorimotor features can effectively distinguish between diabetes mellitus and chronic kidney disease, it could serve as a valuable and non-invasive indicator for early detection, swift screening, and ongoing monitoring, aiding in the differentiation between these diseases. This study classified diverse diagnoses based on motor performance using a novel pinch-holding-up-activity test and machine learning models based on deep learning. Dataset from 271 participants, encompassing 3263 hand samples across three cohorts (healthy adults, diabetes mellitus, and chronic kidney disease), formed the basis of analysis. Leveraging convolutional neural networks, three deep learning models were employed to classify healthy adults, diabetes mellitus, and chronic kidney disease based on pinch-holding-up-activity data. Notably, the testing set displayed accuracies of 95.3% and 89.8% for the intra- and inter-participant comparisons, respectively. The weighted F1 scores for these conditions reached 0.897 and 0.953, respectively. The study findings underscore the adeptness of the dilation convolutional neural networks model in distinguishing sensorimotor performance among individuals with diabetes mellitus, chronic kidney disease, and healthy adults. These outcomes suggest discernible differences in sensorimotor performance across the diabetes mellitus, chronic kidney disease, and healthy cohorts, pointing towards the potential of rapid screening based on these parameters as an innovative clinical approach.
糖尿病和慢性肾病在全球范围内呈不断上升的流行趋势，常伴有神经病变等合并症，导致各种神经肌肉症状，影响日常活动。有趣的是，先前的研究表明在这些病症中感觉运动功能存在差异。如果评估感觉运动特征能够有效区分糖尿病和慢性肾病，那么它可作为早期检测、快速筛查和持续监测的有价值的非侵入性指标，有助于区分这两种疾病。本研究使用一种新颖的捏举活动测试和基于深度学习的机器学习模型，根据运动表现对不同诊断进行分类。来自271名参与者的数据集，包括三个队列（健康成年人、糖尿病患者和慢性肾病患者）的3263份手部样本，构成了分析的基础。利用卷积神经网络，采用三种深度学习模型根据捏举活动数据对健康成年人、糖尿病患者和慢性肾病患者进行分类。值得注意的是，测试集在参与者内部和参与者之间比较的准确率分别为95.3%和89.8%。这些病症的加权F1分数分别达到0.897和0.953。研究结果强调了扩张卷积神经网络模型在区分糖尿病患者、慢性肾病患者和健康成年人的感觉运动表现方面的能力。这些结果表明，糖尿病、慢性肾病和健康队列之间的感觉运动表现存在明显差异，表明基于这些参数进行快速筛查作为一种创新临床方法具有潜力。
