YOLOv5를 이용한 정상/비정상 적혈구 구별 알고리즘 개발

초록

Blood with cells such as WBC, RBC, and platelets, is often used in health diagnoses using morphological analyses. Among them, RBC is used as an indicator for diagnosing diseases such as diabetes, anemia, and heart failure; however, their accuracy is poor due to subjective analyst interpretations since they change reversibly based on various environmental factors such as viscosity, osmotic pressure, and calcium in the blood. In this study, an algorithm that objectively analyzes the shape of RBC and measures the ratio using YOLOv5, an AI technology, was developed and used to distinguish them as normal or abnormal. Data was augmented using image processing technology due to low data, and the YOLOv5 model was applied after labeling, where it was classified into four classes. The developed algorithm was verified through precision and recall, and subsequently applied to the test data to distinguish normal/abnormal red blood cells to obtain a ratio.

키워드