Optimized Attention with Sparse Graph Convolutional Network Based Tumor Segmentation Using Field Programmable Gated Array

In recent years, deep learning (DL) techniques have become increasingly intriguing for segmentation and classification. A convolutional neural network (CNN) needs to do multiple upsampling operations in order to get the input image from the characteristics map during breast Tumor (BT) detection. In this study, a field programmable gate array (FPGA) for BT segmentation using optimized Residual group attention network based Sparse Graph Convolutional Neural Network (RGAN-SGCNN) is introduced. The optimal trade-off between accuracy and hardware complexity is determined by the FPGA implementation of RGAN-SGCNN taking into account both fixed and floating point operations. Several adder and multiplier units, which demand more space and power, are typically needed for the RGAN-SGCNN network model. A carrier select quality-efficient approximate multiplier with low power binary adder (QEAM-LPBA) is therefore used as the foundation for an optimized approximate multiplier. Furthermore, the Running city game optimizer (RCGO) is presented as a means of fine-tuning the network model's parameters. The proposed model demonstrated high efficiency in an experimental scenario, with an accuracy of 98.89. The proposed model's FPGA design, with a power consumption of 0.32W and a LUT of 12,167, demonstrated efficiency through experimental findings.