Optimized Deep Learning-Based Fully Resolution Convolution Neural Network for Breast Tumour Segmentation on Field Programmable Gate Array

Deep learning (DL) approaches have been highly interesting in segmentation and classification in recent years. During breast cancer detection, a convolutional neural network (CNN) requires several up-sampling operations to recover the original image from the feature map. This research introduces an optimised fully resolution-CNN (FR-CNN) based breast tumour segmentation in the field programmable gate array (FPGA) platform. The FPGA implementation of FR-CNN considers both fixed and floating point operations to find the best trade-off between accuracy and hardware complexity. The FR-CNN network model usually requires several adder and multiplier units that consume more power and area. Hence, an optimised Vedic multiplier based on a carry select adder with Simplified Sum-Carry Generation Logic (VCSA-SSCGL) is introduced. In addition, the particle swarm optimisation algorithm (PSO) is introduced for tuning the parameters in the network model. In the experimental scenario, the proposed model achieved an accuracy of 96.89%, precision of 95.84%, F-score of 96.08%, specificity of 96.73%, mean absolute error (MAE) of 0.87, dice similarity coefficient (DSC) of 0.93, and Jaccard coefficient (JC) of 0.9. Also, the FPGA design of a proposed model consumed only 0.6124W power and a LUT of 12,167. The experimental results prove the efficiency of a proposed method.