Residual-Shuffle2DConv-Squeeze Network Approach for Enhanced Blood Cell Segmentation and Hematological Disorder Detection.

Nowadays, blood cell segmentation has emerged as a popular solution for diagnosing hematological disorders. For hematological disorder detection, existing techniques face various limitations, including noise, weak edges, and intensity inhomogeneity. To rectify these problems, a novel Residual-Shuffle2DConv-Squeeze Network approach is proposed in this research to enhance the blood cell segmentation for hematological disorder diagnosis. Distinguishing blood cells, overlapped cells segmentation, edge detection, and morphological operations are the different stages of this approach for performing blood cell segmentation. The Residual-Shuffle Global Attention Network is used for morphological feature extraction, and this network integrates the Residual Network and Shuffle Global Attention Network. To retain fine-grained morphological features and capture variations of blood cell structure, the Residual-Shuffle Global Attention Network model is applied. The Shuffle Global Attention Network module has ShuffleNet and the Global Attention Mechanism. The ShuffleNet reduces the computational cost, and the Global Attention Mechanism helps to preserve crucial features through various layers. Finally, the 2DConv-SNN is applied to detect and classify hematological disorders. The comprehensive experiments are conducted on different datasets, including the Sickle Cell Disease Dataset and the Acute Lymphoblastic Leukemia dataset. The experimental results showcased that the Residual-Shuffle2DConv-Squeeze Network approach enhanced the detection of the hematological disorder with an accuracy of 98.69%, a dice coefficient of 97.05% and a Jaccard index of 96.43% respectively.