Penentuan Status Gizi Wanita Umur Subur (WUS) pada Gambar 2D Berdasarkan Pengukuran LILA dengan YOLOv10 dan SAM

Abstract

The measurement of Mid-Upper Arm Circumference (MUAC) is crucial in determining nutritional status, but manual methods often have limitations such as subjectivity and variations in results. This research proposes a computer vision-based approach by combining the YOLOv10 detection model and the Segment Anything Model (SAM) segmentation model for automated MUAC measurement. The research process includes digital image data collection and manual measurements, annotation and labeling, data preprocessing, and training the YOLOv10 model to detect the upper arm area. The SAM model is then used for segmentation to enhance measurement precision. The YOLOv10 model was trained using a dataset of 144 images, divided into a 70:20:10 ratio. System evaluation was conducted by comparing automatic measurement results with manual measurements using Precision, Recall, and Mean Average Precision (mAP) metrics for detection model training, as well as Mean Absolute Error (MAE) for accuracy assessment. The test results indicate that the developed model demonstrates strong detection performance, achieving an mAP50 score of 0.994 and an mAP50-95 score of 0.939. However, further analysis reveals variations in the differences between system and manual measurements. Among the 72 individuals tested, 22 had differences in the range of 0–1 cm, 18 in the range of 1–2 cm, and 32 exhibited differences greater than 2 cm. The MAE calculation indicates an average difference of 2.11 cm between system and manual measurements. This variation suggests that while the system has high detection accuracy, discrepancies in size calculation remain. Therefore, further development, such as expanding the dataset and refining image processing techniques, is necessary to improve the system’s precision in supporting digital image-based measurenment of MUAC for nutritional status analysis.