Blood Cells Detection Using Faster-RCNN

There has been an increase in the number of health issues in the recent decade with the evolution of various kinds of microorganisms and increasing pollution. The answer to prevent and protect ourselves from the diseases threatening our life lies at the microscopic level in our own blood. Blood consists of erythrocytes (RBCs), lymphocytes (WBCs) and platelets. Object detection is one of the many technologies which can help determine the different constituents of blood. It is a computer technology which deals with object occurrences of pre-defined classes such as cars, humans, buildings, etc. And has far-reaching applications such as self-driven cars, face recognition, face detection, etc. For object detection to be applicable in real-life situations, it should be fast and efficient in determining various entities in an image. There are many object detection techniques developed to make it more efficacious such as CNN and the family of R-CNN. In our research, we apply Faster R-CNN to the Blood Cell Count and Detection dataset to detect RBCs, WBCs, and Platelets. The number, shape and other meta-information or in other words any anomaly in the different constituents of blood can help in early detection of a wide array of issues and/or diseases such as leukemia, lymphoma, anemia, thrombocytopenia, leukopenia, sickle cell disease, etc. Our model detects a bounding box on the three blood constituents with a justifiable accuracy. It identifies all three constituents in the blood smear slide and forms a bounding box around them with their respective labels.