Implementasi Arsitektur EfficientNet pada Platform Android untuk Klasifikasi Penyakit Hipertensi Berdasarkan Citra Iris Mata

Abstract

Hypertension is classified as a non-communicable disease and is commonly known as “The Silent Killer” due to its minimal or absent symptoms. This condition is characterized by blood pressure exceeding the normal range. Commonly used methods to determine hypertension conditions currently involve examinations at healthcare facilities or self-monitoring using digital blood pressure monitors. However, both methods have limitations, such as requiring time to visit healthcare centers and the relatively high cost of devices for some segments of the population. As an alternative, the iridology approach offers a more accessible and affordable method for classifying hypertension conditions through visual analysis of iris images. One of the indicators of hypertension in iridology is the appearance of a gray or cloudy ring around the iris, known as the sodium ring. This ring indicates a metabolic imbalance caused by excessive salt consumption. However, further medical examination is necessary to prevent potential complications. This study utilized 300 iris images collected from Prof. Dr. Chairuddin P. Lubis Hospital, USU, and the general public in Padang Bulan Subdistrict, which were evenly divided into three categories comprising normal, prehypertension, and hypertension. Subsequently, the data were divided into training (70%), validation (20%), and testing (10%) subsets. The pre-processing steps included labeling, cropping, resizing, and augmentation. The EfficientNet architecture was chosen for its effectiveness in extracting visual features through compound scaling across depth, width, and resolution. The results confirmed the model’s ability to accurately categorize iris images into three classes, reaching a test accuracy of 93%. This performance was optimized by tuning the hyperparameters with 30 epochs and batch size 32. The developed system has the potential to serve as an alternative solution for hypertension classification and aids in facilitating early detection and intervention.