Anisocytosis Abnormalities Classification using Spatial Attention Mechanism

Anisocytosis is an abnormal condition where the diameters of mature Red Blood Cells (RBCs) are not in the normal range (6.2 to 8.2 μm). Bodies are more functional when their cells and tissues receive sufficient oxygen (O
2
), and RBCs carry 0
2
from the lungs and deliver it to all the cells and tissues. RBCs with abnormal (anisocytosis) decrease 0
2
carrying capacity, resulting in several diseases such as Anemia and Thalassemia. Under the microscope, hematologists need to spend more time analyzing the characteristics of each RBC. The proposed size-attention convolutional neural network (SACNN) model classifies the anisocytosis abnormalities efficiently and accurately as early as possible to overcome blood-related diseases. Before the abnormal anisocytosis classification, the overlapped RBCs are split via the modified Fast Radial Symmetric Watershed (mFRSW) technique. The SACNN model is validated on the Chula-PIC-Lab dataset and achieved accuracy and F1 score of 97.22%, which is better performed than benchmark models.