The primary objective of this research work is to build a binary classifier for categorizing the input brain magnetic resonanceimaging (MRI) images as either demented or nondemented with high accuracy. A novel hyper-parameter updating method called Randomized Adam (RanAdam) is proposed for enhancing the dementia classification accuracy of elephant herding optimization algorithm and other swarm intelligence (SI) algorithms. Usually, Adam method is widely used in deep learning neural networks for hyper-parameters updating, and it is ingenious to use Adam and its modified version called RanAdam as hyper-parameters updating method for SI algorithms. The proposed RanAdam algorithm tries to find actual optimal values for hyper-parameters near the optimal values given by Adam method through the Controlled Randomness procedure. This research work also compares dementia MRI image classification performance of elephant herding optimization-based transformation technique with the standard clustering approaches and other transformation approaches. In this research work, 117 subjects (65 non-dementia and 52 dementia subjects) acquired from the Open Access Series of Imaging Studies (OASIS) database is used. Two cases are analyzed in all the techniques: with and without statistical features. The highest accuracy of 90.6% is achieved by elephant herding optimization (EHO)-based transformation technique combined with RanAdam for updating hyper-parameters for the case without statistical features. To verify the efficiency of the proposed technique, a popular Pima diabetic dataset is considered in addition to the OASIS dementia dataset and 88% accuracy is earned for EHO-based transformation technique combined with RanAdam.

中文翻译：

---

痴呆 MRI 图像分类使用基于大象群优化和随机亚当方法更新超参数的变换技术

这项研究工作的主要目标是建立一个二元分类器，用于将输入的脑磁共振成像 (MRI) 图像高精度地分类为痴呆或非痴呆。提出了一种称为随机亚当（RanAdam）的新型超参数更新方法，用于提高大象放牧优化算法和其他群体智能（SI）算法的痴呆分类准确性。通常，深度学习神经网络中广泛使用 Adam 方法进行超参数更新，而巧妙地使用 Adam 及其修改版本 RanAdam 作为 SI 算法的超参数更新方法。提出的 RanAdam 算法试图通过受控随机性程序找到接近 Adam 方法给出的最优值的超参数的实际最优值。这项研究工作还比较了基于大象放牧优化的转换技术与标准聚类方法和其他转换方法的痴呆 MRI 图像分类性能。在这项研究工作中，使用了从开放获取系列成像研究 (OASIS) 数据库中获取的 117 名受试者（65 名非痴呆和 52 名痴呆受试者）。在所有技术中都分析了两种情况：有统计特征和没有统计特征。90.6% 的最高准确率是通过基于大象群优化 (EHO) 的转换技术与 RanAdam 相结合来实现的，用于在没有统计特征的情况下更新超参数。为了验证所提出技术的效率，