Detection of Sickle Cell Anemia in Blood Smear using YOLOv3

Object recognition algorithms such as the YOLOv3 algorithm have a lot of possible applications in the field of medicine. One such application is the study's aim of recognizing the presence of sickle cell-shaped red blood cells (sickle cell disease) in blood smear samples. As YOLOv3 uses features learned from a trained model using deep convolutional neural networks, it is possible to use it in automating the process of morphologically recognizing the presence of sickle cells in blood smear samples. Additionally, as blood smear samples may have different staining methods, the images must be processed prior to recognition which will result in a grayscale image frame. Based on the results generated by the system when processing 12 blood smear samples, a confidence level of 50% corresponding to a threshold value of 0.5 provided the highest accuracy rate when using YOLOv3-Tiny custom weights. Further testing by the researchers showed that the detections in unstained test slides have a higher confidence level per detection than in test slides stained using a darker pigment caused by the stain's effect on the illumination of the image. Overall, the design system was able to achieve 100% accuracy in detecting the presence of sickle cell-shaped RBCs in 12 blood smear samples.