Application of Artificial Neural Networks in a Two-step Classification for Acute Lymphocytic Leukemia Diagnosis by Blood Lamella Images

Introduction: This study aimed to present a system based on intelligent models that can enhance the accuracy of diagnostic systems for acute leukemia. The three parts including preprocessing, feature extraction, and classification network are considered as associated series of actions. Therefore, any dysfunction or poor accuracy in each part might lead in general dysfunction of the whole system.   Materials and Methods: In the current study, rgb2hsv code and two-dimensional Wiener were used for the preprocessing part. In addition, fuzzy C-means method was applied for the segmentation step and nervous networks-based techniques as well as support vector machines were utilized in the classifying networks.   Results: The results of the proposed method were compared with other training methods; demonstrating that 91.4% as the lowest and 95.7% as the highest mean accuracies belonged to Gradient Descent with "Adaptive Thresholding" and "Resilient Back propagation", respectively. Moreover, the results revealed that regarding the outputs accuracy, 48% as the lowest and 95.7% as the highest mean test accuracies were related to the MPN and proposed networks, respectively.   Conclusion: The application of the proposed network in this study is that eliminate the weak points of all the networks in addition to presenting the advantages of these network. Combining the networks improved the accuracy of output up to 98% and considerably reduced the time required for calculations. It could be concluded that we can reach a more accurate network with less hardware facilities.