BLOOD CELL DETECTION AND CLASSIFICATION WITH AUTOENCODER-DRIVEN R-CNN

Abstract

The domain of hematological studies and diagnostic medicine has been changed by automated blood cell detection (BCD) and classification through deep learning (DL) and object detection (OD).It is the focus of this work to identify and classify red blood cells (RBCs), white blood cells (WBCs), and platelets, all of which are necessary in the diagnosis of infections, anemia, and clotting disorders. In solving issues such as overlapping cells and varying morphologies, the latest advancements have mainly been the replacement of the Region-Based Convolutional Neural Networks (R-CNN) by Faster R-CNN, which has enhanced accuracy, efficiency and scalability. In the given work, we propose a method based on image enhancement and convolutional sparse autoencoder, which effectively reduces noise and preserves the important properties of dependable identity. Another aspect of the Faster R-CNN model, the MobileNet backbone is another lightweight but capable of analyzing blood cell morphology accurately and at scale. To enhance the results of patients, promote hematological diagnoses, and assist in the construction of an individual treatment course, the proposed system will automate the detection process and provide fast, objective, and reliable results.