A Novel Approach For Breast Cancer Detection Using Neural Networks

Abstract

In the breast tissue, growth of malignant cells started from breast lobules or milk ducts inner lining is known as Breast cancer. This malignant growth will spread to other organs. Most widely spread cancer in women is breast and it takes second place among widespread disease across the world. Thus, it is vital for checking the number of lives lost because of premature breast cancer in positive management and the drop. The fast development in machine learning and exclusively deep learning endures the interest of medical imaging society in spread over these methods to enhance the accuracy of breast cancer diagnosis, Machine-learning approaches, with a focus on deep learning algorithms, have particularly shown a promising applicability in medical image analysis in the area of nuclear medicine. But, breast cancer s accuracy of classification is evaluated by the examination of approaches in machine-learning, linking Convolution Neural Networks (CNN) systems, was not still established. In view of this, the entire work is organised in three phases. In the first phase, Deep Belief Network (DBN) is constructed by including the fuzzy c-means segmentation process with the help of masking with feature extraction Scale Invariant Feature Transform (SIFT) and Speeded Up Robust Transform (SURF). In the second phase, segmentation of the image is done using Mask RCNN (Region-based convolution neural networks) with the assistance of ensemble classifier of random forest and decision tree in V3 and ResNet 152. In the third phase, a new deep learning model established based on a CNN with the combination of encoder and Unet. Moreover, to improve the classification this work used Efficientnet B5 and B6 with the ensemble classifiers such as K-Means Nearest Neighbor (KNN), Decision Tree (DT) and Random Forest (RF).The performance of all these three proposed phases is examined using parameters such as accuracy, sensitivity, specificity, f1-score, ROC. The analysis of ii first phase are as follows: SIFT feature using adaboost classifier achieves 88.39% of accuracy,88.16% of precision, 88.32% of recall and 88.23% of F1 score. SIFT feature using gradient boosting classifier achieves 83.93% of accuracy,88.89% of precision, 81.63% of recall and 82.5% of F1 score. SIFT feature using MultiLayer Perceptron (MLP)classifier achieves 99.11% of accuracy,99.22% of precision, 98.98% of recall and 99.09% of F1 score. The SURF feature using adaboost classifier achieves 95.54% of accuracy,95.57% of precision, 95.35% of recall and 95.45% of F1 score. SURF feature using gradient boosting achieves 91.96% of accuracy,92.19% of precision, 91.5% of recall and 91.77% of F1 score. SURF feature using MLP classifier achieves 78.57% of accuracy,78.3% of precision, 78% of recall and 78.12% of F1 score. The analyses of second phase are as follows: The proposed Deep IncepRC achieves 97.3% accuracy, 97.37%precision, 97.15% recall, 97.25%f1-score and 97.34% AUC. The analysis of third phase are as follows: Proposed Effecientnet5 Ens(DT-KNN) achieves 98% of accuracy, 98% of AUC, 98.2% of f1-score, 98.5% of precision, 97.8% of sensitivity and 98.5% of specificity. Proposed Effecientnet5 Ens(RF-KNN) achieves 98.5% of accuracy, 98.5% of AUC, 98.5% of f1-score, 97.5% of precision, 100% of sensitivity and 96.5% of specificity. Proposed Effecientnet6 Ens(DT-KNN) achieves 99% of accuracy, 99% of AUC, 98.7% of f1-score, 99% of precision, 98.5% of sensitivity and 99.5% of specificity. Proposed Effecientnet6 Ens(RF-KNN) achieves 99.5% of accuracy, 99.5% of AUC, 99.3% of f1-score, 99.8% of precision, 99% of sensitivity and 99.8% of specificity. newline