Breast Cancer Detection in Mammogram Images by MapReduce Based Deep Convolutional Neural Networks

Abstract

Imaging techniques using mammographic pictures are the most efficient and straightforward way to diagnose Breast Cancer (BC). The accurate discovery can significantly lower the mortality rate caused by BC. Machine Learning (ML) techniques were utilized for the prediction of BC in images. The ML approaches offered a unique way of making decisions, and they were able to aid the medical expert in providing a second perspective on more precise nodule detection. However, the over-fitting problem in ML degraded the performance of BC detection. To solve this issue, Deep Learning (DL) algorithms were utilized to detect BC early using mammographic images, which is a huge help in detecting BC at an early stage. The Deep Convolutional Neural Networks (DCNN) are being used in DL to help with an accurate BC diagnosis and better BC image outcome detection. However, the time complexity of DCNN is high for a large-scale dataset. In this paper, The DCNN is parallelized in the Mappers of MapReduce (MR) programming model in which the network weights can be iteratively adjusted via determining their fractional gradients following all collections of learning images are propagated throughout the system. Accordingly, the Mapper will parallelize the training phase by distributing the normal, benign, and malignant images that have been preprocessed using the K-Means (KM) algorithm. Each image can be provided to various DCNNs and each DCNN is trained independently in parallel. The output weight of the training stage of DCNNs is aggregated in Reducer and then the weights are then updated for the next iteration. After the completion of the training process, the updated weight is utilized in the all distributed DCNN for classifying the test images of BC. The whole process is termed MapReduce-KM-DCNN (MR-KM-DCNN). Finally, the experimental outcome demonstrates that the proposed MR-KM-DCNN method achieves 88.35 percent accuracy, which is 13.48%, 8.02 %, and 4.85 % higher than the existing SSL-DCNN, AN-DCNN, and KM-DCNN methods. In addition, the running time is reduced by 60% on average compared to previous methods. As a result, it has been verified that the MR-KM-DCNN attained the highest BC detection rate while requiring the smallest amount of computational time.

Effective Feature Selection Using Multi-Objective Improved Ant Colony Optimization for Breast Cancer Classification

“…7. The classification accuracy of the MIACO-SAE is 99.36% whereas the TTS-ML [17] has 98.13%, IFS-GA [18] has 97.43% and DCNN [21] has 88.35%. The MIACO method selects the optimal features with better correlation that helps to improve the classification accuracy using SAE.…”

“…Table 3 shows the comparison of MIACO-SAE with TTS-ML [17], IFS-GA [18] and DCNN [21]. From Table 3, it is known that the classification using MIACO-SAE is improved than the TTS-ML [17], IFS-GA [18] and DCNN [21]. The graphical comparison of classification accuracy is shown in Fig.…”

“…The comparative analysis of MIACO-SAE with existing research are shown in this section. Existing research such as TTS-ML [17], IFS-GA [18] and DCNN [21] are used to evaluate the efficiency of the MIACO-SAE method. Table 3 shows the comparison of MIACO-SAE with TTS-ML [17], IFS-GA [18] and DCNN [21].…”

“…Existing research such as TTS-ML [17], IFS-GA [18] and DCNN [21] are used to evaluate the efficiency of the MIACO-SAE method. Table 3 shows the comparison of MIACO-SAE with TTS-ML [17], IFS-GA [18] and DCNN [21]. From Table 3, it is known that the classification using MIACO-SAE is improved than the TTS-ML [17], IFS-GA [18] and DCNN [21].…”

“…Thirunavukkarasu [21] presented DCNN for classifying the breast cancer. The DCNN was parallelized at the mappers of map reduce (MR) in that network weights were altered through identification of fractional gradients.…”