Image Enhancement for Breast Cancer Detection on Screening Mammography Using Deep Learning

Abstract

Mammography offers the most efficient approach for detecting breast illnesses early. Nevertheless, Image enhancement to improve breast cancer detection is required since mammograms are low-contrast and noisy images, and typical diagnostic markers such as microcalcifications and masses are challenging to identify. Due to this issue, this paper evaluates the impact of image enhancement on the performance of the deep learning approach and conducts qualitative and quantitative testing of various deep learning models in breast cancer classification. This study uses Mini Digital Database for Screening Mammography (Mini-DDSM) breast dataset containing cancer and normal images. The mammography images are then improved using morphological erosion and enhanced using two image enhancement algorithms which are Unsharp Masking (UM) and High-Frequency Emphasis Filtering (HEF). Deep learning classification algorithms such as ResNet, DenseNet, and EfficientNet are employed to classify breast cancer. Each architecture is compared and analyzed regarding the effect of the image enhancement techniques and achieves the highest 76.08% accuracy score on breast cancer classification in the mammography dataset using the HEF technique.