An empirical study of object detection models for the detection of iron deficiency anemia using peripheral blood smear images.

Abstract

Iron Deficiency Anemia (IDA) is the nutritional disorder that occurs when the body does not contain enough iron, an essential component of hemoglobin (Hb). The World Health Organization (WHO) estimated that IDA is the main cause of anemia in 1.62 billion cases worldwide [1]. Although IDA rarely results in death, it has significant adverse impacts on human health. During diagnosis, the hemoglobin indices show low mean corpuscular hemoglobin and mean corpuscular hemoglobin volume. On Peripheral Blood Smear (PBS) images viewed under a microscope by hematologists, IDA shows hypochromic and microcytic red cells. The purpose of the proposed research is to develop a computer-aided system that will allow hematologists to diagnose and detect diseases more accurately and quickly, therefore saving them time and effort. In order to diagnose or detect clinical disorders, non-invasive techniques like machine learning algorithms are employed. This work aims to identify IDA by utilizing the RetinaNet-Disentangled Dense Object Detector (DDOD) to localize hypochromic microcytes in PBS images. To the best of our knowledge, this is the first work using the object detection technique to detect IDA based on the Red Blood Cell (RBC) morphology. We carried out an extensive quantitative and qualitative evaluation of the model. Additionally, a comparison was made between the performance of our model and other object detection models. The results showed that our approach outperformed state-of-the-art techniques, with a mean average precision that was more than 8% higher.