With the recent progress of artificial intelligence (AI) technology, deep learning-based approaches in the medical field have increased lately. This paper proposes a deep learning network using Ajou University Hospital’s 10-year breast cancer patient dataset to predict the recurrence year of cancer. The proposed network analyzes the whole prognostic factors of the patient. In addition, the influence of each prognostic factor was analyzed by excluding the factors in the training respectively. The network showed high performance by achieving 0.91 area under the receiver operating characteristic (ROC) curve (AUC). For AI hardware accelerator implementation, the proposed fixed 16-bit integer quantization method was performed to compress the parameter of the proposed network. The proposed quantization method enabled 37.41% parameter compression of the proposed network. The accelerator showed higher throughput and lower power consumption than the graphic processing unit (GPU). The proposed hardware accelerator architecture is implemented on the Xilinx Kintex UltraScale+ field programmable gate array (FPGA).
