Detecting distorted and benign blood cells using the Hough transform based on neural networks and decision trees

Sickle-cell anemia is one of the most important types of anemia. This paper presents an algorithm for detecting blood cells characteristic of sickle-cell anemia. First, I discuss the construction of an algorithm that can be used to detect and count benign or distorted red blood cells (RBCs) in a microscopic colored image, even if those cells are hidden or overlapped. Second, I explain the process for checking and analyzing the constructed RBC data by applying two important techniques in data mining: the neural network (NN) and the decision tree. I then review experiments demonstrating that these models show high accuracy when predicting the counts of benign or distorted cells. In these experiments, the algorithm has segmented around 99.98% of all input cells, helping to improve the diagnosis of sickle-cell anemia. The NN has shown a 96.9% agreement with the algorithm’s prediction outcomes, and the classification and regression tree has achieved 92.9%.