| Abstrakt: |
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. [ABSTRACT FROM AUTHOR] |
