Analyzing Performance YOLOv8 in Detection, Amount, and Type of Acne Lesions

Acne is one of the most common skin diseases and usually appears on the facial area. The onset of acne is caused by oily skin conditions and excessive pores, resulting in the appearance of small bumps. Acne has 5 types, namely whiteheads, blackheads, nodules, pustules, and papules. Although the five types of acne have different characteristics, there are still dermatologists who make mistakes in making diagnoses, making it difficult for dermatologists to provide the right diagnosis. This research uses the YOLOv8 model. YOLOv8 has broad potential in the field of dermatology, because this model can detect objects of various sizes and has accurate and efficient performance results to reduce diagnostic errors that previously often occurred due to differences in the characteristics of five types of acne. In addition, YOLOv8 is a model that can perform several tasks at once such as quantitative analysis, segmentation, and classification of acne. Therefore, this study performs three detections, namely acne detection, number, and type of acne. The data used for the three detections contains 823 images of people who have acne and labels to help the process of detecting acne and acne type. The accuracy produced by this model is 89% for the detection of acne, 82% for the number of acne, and 49% for the type of acne. However, the detection of acne types resulted in low accuracy due to an imbalance in the number of labels, similarity between acne types, errors in labeling, poor selection of hyperparameters, and a less diverse dataset. Future efforts should be able to correct dataset imbalances, such as the number of datasets, labeling acne type classes evenly to improve detection results by producing better accuracy, using the right hyperparameters, and data augmentation. This study confirms that YOLOv8 can improve the efficiency of dermatologists in diagnosing accurate acne so that patients can receive appropriate treatment.