A Convolutional Neuro-Fuzzy Network Using Fuzzy Image Segmentation for Acute Leukemia Classification

Acute lymphoblastic leukemia (ALL) is a kind of blood cell malignancy that is distinguished by the presence of many lymphoid blasts in circulation. ALL accounts for over 80% of juvenile leukemia and is most common in children aged 3 to 7. The conventional approach to ALL classification requires meticulous examination of cell images by expert pathologists, which is time-consuming and results in non-standardized reports. Automation in ALL prognoses is an essential but challenging procedure. Fuzzy reasoning rules are integrated into connectionist networks when fuzzy systems and neural networks are combined to generate neuro-fuzzy systems. Current neuro-fuzzy systems, on the other hand, are created with shallow structures that have poor generalization capacity. This article automatically detects ALL from microscopic cell images, employing a deep convolutional neuro-fuzzy network. In order to reduce overfitting, data augmentations were utilized, and the number of samples got increased before training the model. Image preprocessing includes a two-stage fuzzy color segmentation technique is used to separate leukocytes from the rest of the blood components. After applying image preprocessing, the size of images decreased to $64 \times64$ such that containing a single nucleus per image. We propose a convolutional neuro-fuzzy network based on Takagi-SugenoKang (TSK) fuzzy model for acute lymphoblastic leukemia diagnosis. Our suggested model achieved an accuracy of 97.31% on average for acute lymphoblastic leukemia detection.