<![CDATA[In deep learning, Convolutional Neural Network (CNN) is an algorithm from Artificial Neural Network (ANN) which is generally used to analyze visual images. This algorithm can automatically extract important features from each image without human assistance, besides that the CNN algorithm is also more efficient than other neural network methods, especially in memory and complexity. In training, the algorithm will be given training data in the form of images that have been labeled so that the algorithm will be able to recognize the important characteristics of each of the labeled images. After the training stage, the trained algorithm will be given data validation in the form of an unlabeled image to be analyzed and classified. The algorithm will analyze the training and validation data for the specified number of epochs and provide information in the form of the level of accuracy of each epoch that is performed. Some that affect the level of accuracy include the type of optimizer, the pixel size of the input image, and the number of epochs. In this study, the CNN algorithm was used with a layer sequence made personally by the author. The research was conducted in a cloud-based Jupyter notebook environment called Google Colab. The dataset used in this study is the Alzheimer's MRI Preprocessed Dataset which can be accessed by the public on the Kaggle website. The dataset consists of 6400 brain MRI scan images which are divided into four classes, namely: Non Demented, Very Mild Demented, Mild Demented, and Moderate Demented. As much as 20% of the dataset is used as data validation. In this study, the dataset will be analyzed by the CNN algorithm with several predetermined scenarios, then the accuracy of the training and validation data will be compared with each other to find the most optimal scenario. There are two input image pixel size scenarios to be compared, namely 128 x 128 pixels and 224 x 224 pixels. There are three types of optimizers that will be compared, namely Stochastic Gradient Descent (SGD), Adam, and RMSprop. From the research results, the most optimal type of optimizer to use with the architecture that has been made and the Alzheimer's MRI Preprocessed Dataset is the Adam optimizer. Architectural training with an input size scenario of 224 x 224 pixels, seven epochs, and using the Adam optimizer achieves the most optimal accuracy rate, namely with a training data accuracy rate of 93.01% and a data validation accuracy rate of 94.45%. Architecture training with an input size scenario of 224 x 224 pixels and using the Adam optimizer achieves the most optimal number of epochs, namely achieving an accuracy level above 90% in just five epochs. Keywords: CNN, Alzheimer's, accuracy, optimizer, optimal. Daftar Pustaka [1] Burns, A., & Iliffe, S. (2009). Alzheimer's disease. Bmj-British Medical Journal, 338. [2] Dementia. (2022, 20 September). https://www.who.int/news-room/factsheets/detail/dementia [3] Goodfellow, I., Bengio, Y., & Courville, A. (2016). Deep learning. MIT press. [4] Khan, S., Barve, K. H., & Kumar, M. S. (2020). Recent advancements in pathogenesis, diagnostics and treatment of Alzheimer’sdisease. Current Neuropharmacology, 18(11), 1106-1125. [5] LeCun, Y., Bengio, Y., & Hinton, G. (2015). Deep learning. nature, 521(7553), 436-444. [6] Mendez, M. F. (2006). The accurate diagnosis of early-onset dementia. The International Journal of Psychiatry in Medicine, 36(4), 401-412. [7] Mortimer, J. A., Borenstein, A. R., Gosche, K. M., & Snowdon, D. A. (2005). Very early detection of Alzheimer neuropathology and the role of brain reserve in modifying its clinical expression. Journal of geriatric psychiatry and neurology, 18(4), 218-223. [8] National Institute for Health and Clinical Excellence. (2006, November). Dementia: Quick Reference Guide. Diambil kembali darihttps://web.archive.org/web/20080227161412/http://www.nice.org.uk/nicemedia/pdf/CG042quickrefguide.pdf. [9] Simon, R. P., Aminoff, M. J., & Greenberg, D. A. (2009). Clinical neurology. Lange Medical Books/McGraw-Hill. [10] Smith, M. A. (1998). Alzheimer disease. International review of neurobiology, 42, 1-54. [11] Valueva, M. V., Nagornov, N. N., Lyakhov, P. A., Valuev, G. V., & Chervyakov, N. I. (2020). Application of the residue number system to reduce hardware costs of the convolutional neural network implementation. Mathematics and computers in simulation, 177, 232-243.]]>
Copyrights © 2023
