Claim Missing Document
Check
Articles

Found 2 Documents
Search

Adagrad Optimizer with Compact Parameter Design for Endoscopy Image Classification Sofyan Pariyasto; Suryani; Vicky Arfeni Warongan; Arini Vika Sari; Wahyu Wijaya Widiyanto
ULTIMATICS Vol 17 No 2 (2025): Ultimatics : Jurnal Teknik Informatika
Publisher : Faculty of Engineering and Informatics, Universitas Multimedia Nusantara

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.31937/ti.v17i2.4225

Abstract

Research on CNN Model and Adagrad Optimizer is expected to help identify diseases in the medical world. Especially in the field of image classification in Gastrointestinal endoscopic procedures . The research is specifically for the process of classifying medical images of Diverticulosis, Neoplasm, Peritonitis and Ureters . Previously, there have been quite a lot of studies on CNN and its various optimizers. However, those who have studied the Adagrad optimizer are not too many, especially those discussing the use of minimum parameters. The use of minimum parameters is expected to be one of the contributions of researchers in the fields of computing and medicine. The research was conducted to determine the use of the best parameters and obtain the highest level of accuracy. The research was conducted using minimum epochs starting from epoch 1, epoch 5, and epoch 10. Then the combination process between epoch and the number of convolution layers between 1 and 5 was carried out, resulting in 15 combinations. The test was carried out using 4000 images with 1000 images in each class. From the results of the test, the highest accuracy value was obtained, namely 82.875%. Then the highest average accuracy value was 81.625%.
OPTIMALISASI KLASIFIKASI CITRA MEDIS MENGGUNAKAN CNN DAN ADAM OPTIMIZER DENGAN PARAMATER MINIMUM Sofyan Pariyasto; Vicky Arfeni Warongan; Suryani Suryani
JOURNAL OF SCIENCE AND SOCIAL RESEARCH Vol. 8 No. 1 (2025): February 2025
Publisher : Smart Education

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.54314/jssr.v8i1.2515

Abstract

Abstract: Research in the field of imaging, especially in medical terms, is expected to have a positive impact on the treatment and diagnosis of diseases in the medical world. Medical image classification is a topic that is often researched, this is indicated by the many national and international journals that discuss this topic. Research on medical image classification using Convolutional Neural Network (CNN) usually focuses on the use of maximum parameters (hyper parameters) to get the best results. However, those that use minimal parameters and the smallest resources are still lacking. Based on the existing problems, it is carried out to obtain optimization in the medical image classification process. The classification of medical images in this study focuses on brain tumor images consisting of three classes, namely meningioma, glioma and pituitary tumor. The approach taken in this study is to use the CNN model and Adaptive Moment Estimation (Adam) Optimizer. The study was conducted by combining the smallest parameters from the Adam Optimizer. The parameters combined are Epoch and Convolution Layer. Where 3 Epoch categories (1,5,10) and 5 convolution layers (1,2,3,4,5) are used. From the tests carried out, the highest accuracy results obtained were 92.8% with epoch parameters of 10 and three convolution layers. Meanwhile, the highest average accuracy was recorded at 90.7% with epoch parameters of 10. The fastest computation time required for model creation was 24.83 seconds, and the lowest CPU resource usage during the model creation process was 16.45%. Keywords: Image Classification, CNN Optimization, Adam Optimizer, Brain Tumor, Minimum Parameters Abstrak: Penelitian dibidang citra khususnya dalam hal medis diharapkan dapat membawa dampak baik bagi penanganan dan diagnosis penyakit dalam dunia medis. Klasifikasi citra medis menjadi topik yang cukup sering diteliti, hal ini ditandai dengan banyaknya jurnal baik nasional maupun internasional yang membahas mengenai topik ini. Penelitian mengenai klasifikasi citra medis menggunakan Convolutional Neural Network (CNN) biasanya berfokus pada penggunaan paramater maksimal (hyper parameter) untuk mendapatkan hasil terbaik. Namun yang menggunakan paramater minimal dan sumber daya terkecil masih belum ada. Berdasarkan permasalahan yang ada maka dilakukan untuk mendapatkan optimalisasi dalam proses kalsifikasi citra medis. Klasifikasi citra medis dalam penelitian ini difokuskan pada citra tumor otak yang terdiri dari tiga kelas yaitu, meningioma, glioma dan pituitary tumor. Pendekatan yang dilakukan dalam penelitian ini adalah menggunakan model CNN dan Adaptive Moment Estimation (Adam) Optimizer. Penelitian dilakukan dengan melakukakn kombinasi paramater terkecil dari Optimizer Adam. Paramater yang dikombinasikan yaitu Epoch dan Lapisan konvolusi. Dimana digunakan 3 kategori Epoch (1,5,10) serta 5 lapisan konvolusi (1,2,3,4,5). Dari pengujian yang dilkaukan didapatkan hasil  Akurasi tertinggi yang diperoleh adalah 92,8% dengan parameter epoch 10 dan tiga lapisan konvolusi. Sementara itu, akurasi rata-rata tertinggi tercatat sebesar 90,7% dengan parameter epoch 10. Waktu komputasi tercepat yang diperlukan untuk pembuatan model adalah 24,83 detik, dan penggunaan sumber daya CPU terendah selama proses pembuatan model adalah 16,45%.Kata kunci: Klasifikasi Citra, Optimalisasi CNN, Adam Optimizer, Tumor Otak, Paramater Minimum