Breast Cancer Detection using Image Processing and Machine Learning: A Comprehensive Review and Improved Segmentation Approach

Breast cancer stands as one of the most prevalent health concerns for women. Early detection of breast cancer can significantly improve the chances of survival. In developed countries, more than 19.9% of women will die per year due to breast cancer. Regular breast cancer screening is an important way to detect cancer early. Image processing techniques are highly used for different types of cancer detection applications with medical screening. Segmentation of the breast tumor region is a critical step in image processing related to this manner. Lots of research work can be found on developing ways for detecting breast cancers. However still, there is a need for a standard and robust cancer region segmentation method. Right cancer region segmentation is significant for better feature extraction and better classification. This study presents a comprehensive literature review about technologies used with image enhancement and tumor segmentation. Further, the study proposes a robust approach to image enhancement and cancer region segmentation for breast cancer detection using image processing techniques and machine learning. In this work, mammograms are enhanced using Contrast Limited Adaptive Histogram Equalization and denoised using the Median blurring filter. This study enables an effective image segmentation approach with two stages: removing the background using thresholding, tumor region segmentation using a combination of thresholding, and Canny edge detection. Segmented tumor region’s features are extracted using the Gabor filter. Here the accuracy of the approach is compared with three main machine learning classifiers: Decision Tree, Random Forest, Multinomial Logistic Regression and mammograms are classified into three classes (malignant, benign and normal). Finally, an ensemble approach is proposed using the hard voting mechanism to improve the accuracy. With the dataset of mini-MIAS this proposed approach achieved 78.89% accuracy. Results demonstrated that the proposed breast cancer detection approach improves the performance of segmentation breast tumor regions.