Pasha Akram, Ahmed Syed Thouheed, Painam Ranjith Kumar, Mathivanan Sandeep Kumar, P Karthikeyan, Mallik Saurav, Qin Hong
Department of Computer Science and Engineering, REVA University, Bengaluru, India.
Indian Institute of Technology Hyderabad, India.
Heliyon. 2024 Apr 26;10(9):e30241. doi: 10.1016/j.heliyon.2024.e30241. eCollection 2024 May 15.
Parkinson's disease (PD) is an age-related neurodegenerative disorder characterized by motor deficits, including tremor, rigidity, bradykinesia, and postural instability. According to the World Health Organization, about 1 % of the global population has been diagnosed with PD, and this figure is expected to double by 2040. Early and accurate diagnosis of PD is critical to slowing down the progression of the disease and reducing long-term disability. Due to the complexity of the disease, it is difficult to accurately diagnose it using traditional clinical tests. Therefore, it has become necessary to develop intelligent diagnostic models that can accurately detect PD. This article introduces a novel hybrid approach for accurate prediction of PD using an ANFIS with two optimizers, namely Adam and PSO. ANFIS is a type of fuzzy logic system used for nonlinear function approximation and classification, while Adam optimizer has the ability to adaptively adjust the learning rate of each individual parameter in an ANFIS at each training step, which helps the model find a better solution more quickly. PSO is a metaheuristic approach inspired by the behavior of social animals such as birds. Combining these two methods has potential to provide improved accuracy and robustness in PD diagnosis compared to existing methods. The proposed method utilized the advantages of both optimization techniques and applied them on the developed ANFIS model to maximize its prediction accuracy. This system was developed by using an open access clinical and demographic data. The chosen parameters for the ANFIS were selected through a comparative experimental analysis to optimize the model considering the number of fuzzy membership functions, number of epochs of ANFIS, and number of particles of PSO. The performance of the two ANFIS models: ANFIS (Adam) and ANFIS (PSO) focusing at ANFIS parameters and various evaluation metrics are further analyzed in detail and presented, The experimental results showed that the proposed ANFIS (PSO) shows better results in terms of loss and precision, whereas, the ANFIS (Adam) showed the better results in terms of accuracy, f1-score and recall. Thus, this adaptive neural-fuzzy algorithm provides a promising strategy for the diagnosis of PD, and show that the proposed models show their suitability for many other practical applications.
帕金森病（PD）是一种与年龄相关的神经退行性疾病，其特征为运动功能障碍，包括震颤、僵硬、运动迟缓及姿势不稳。据世界卫生组织统计，全球约1%的人口已被诊断患有帕金森病，预计到2040年这一数字将翻倍。帕金森病的早期准确诊断对于减缓疾病进展和减少长期残疾至关重要。由于该疾病的复杂性，使用传统临床检测方法难以准确诊断。因此，开发能够准确检测帕金森病的智能诊断模型变得十分必要。本文介绍了一种新颖的混合方法，即使用带有Adam和PSO两种优化器的自适应神经模糊推理系统（ANFIS）来准确预测帕金森病。ANFIS是一种用于非线性函数逼近和分类的模糊逻辑系统，而Adam优化器能够在每次训练步骤中自适应调整ANFIS中每个参数的学习率，这有助于模型更快地找到更好的解决方案。粒子群优化算法（PSO）是一种受鸟类等群居动物行为启发的元启发式方法。与现有方法相比，将这两种方法结合起来有可能在帕金森病诊断中提高准确性和鲁棒性。所提出的方法利用了两种优化技术的优势，并将其应用于所开发的ANFIS模型，以最大化其预测准确性。该系统是利用公开获取的临床和人口统计学数据开发的。通过比较实验分析选择ANFIS的参数，以根据模糊隶属函数数量、ANFIS的训练轮数和PSO的粒子数量来优化模型。进一步详细分析并展示了两个关注ANFIS参数和各种评估指标的ANFIS模型：ANFIS（Adam）和ANFIS（PSO）的性能。实验结果表明，所提出的ANFIS（PSO）在损失和精度方面表现更好，而ANFIS（Adam）在准确率、F1分数和召回率方面表现更好。因此，这种自适应神经模糊算法为帕金森病的诊断提供了一种有前景的策略，并表明所提出的模型适用于许多其他实际应用。
