Classification of Acute Lymphocytic Leukemia (ALL) using Ventral-Dorsal and bottleneck Attention-based Network(VDBAN)
Mohammad Zolfaghari, Hedieh Sajedi · Research Square · 2022
Abstract One of the diseases with a high mortality rate worldwide is leukemia or blood cancer. Therefore, a timely and reliable diagnosis of leukemia is essential. Today, Artificial Intelligence and Deep Learning methods can automatically, quickly, and cost-effectively identify many diseases. In this paper, the state-of-the-art Convolutional Neural Network (CNN)based on the proposed attention blocks is introduced for the automatic classification of Acute Lymphoblastic Leukemia (ALL) and normal (healthy) cell images. Before sending images to the network, the preprocessing operation, including data augmentation and White Blood Cells (WBCs, or Leuko cytes) segmentation by Circular Hough Transform (CHT), is performed. Then, the proposed network attention blocks are designed and implemented based on the ventral and dorsal pathways of the occipital lobe that process visual information in the brain. The proposed network is ResNet18, in which the VentralDorsal Attention Block (VDAB) is embedded inside its blocks and the Bottleneck Attention Block (BAB) between them. Therefore, it is called Ventral-Dorsal and Bottleneck Attention Network (VDBAN). We designed three networks, including the first model (ResNet18), the second model (ResNet18 + VDAB) and the proposed model ((ResNet18+ VDAB + BAB) or VDBAN), and trained and tested them with similar parameters on the ALL-IDB2 database. The simulation results showed that the proposed model had obtained the highest value in all performance metrics compared to the other two models in the test step.