Cell morphology based classification for red cells in blood smear images

• The proposed a hybrid neural network architecture for red cell classification.
• The separation of overlapping cells have been attempted.
• Utilized all visual features extracted from red cell images with new classifier.
• Improvement in red cell normality classification result.
• Capable of accurately distinguish different blood diseases.

Red blood cells are the most common type of blood cell and are responsible of delivering oxygen to the body tissues. Abnormalities in red blood cell may change the physical properties of the red cell or shorten its life spend, and may lead to stroke or anemia. In this paper, we proposed a hybrid neural network based classifier, which utilize the visual information extracted from the red blood cell images to determine whether a red cell is normal or abnormal. Based on the feature properties, we clustered the visual features into two main groups, namely shape and texture cluster groups. The input feature clusters were processed using parallel and cascade architecture with multiple input layers. Our experimental result has shown significant improvement in classification accuracy in our proposed system as compared to the single input layer classifier with recent feature selection algorithms.