A common cancer subtype found in women with high mortality and
occurrence rates is Breast Cancer (BC). BC ranks second among the causes of high
mortality rates in women. The annual death rate due to breast cancer surpasses that of
any other cancer type. The global survival rate for patients with breast cancer remains
suboptimal. To enhance this survival rate, it is essential to implement intervention
techniques for early detection and treatment. Screening using the Medio-Latera-
-Oblique (MLO) view and the Cranio-Caudal (CC) view improved the detection of
cancer signs in small lesions. This motivated the radiologist to use both mammographic
views for screening and subsequently to acquire additional information. To automate
this sequential screening process, Image Processing, and Artificial Intelligence (AI)
techniques are incorporated into these views individually and their results were fused.
Further, feature fusion from both views is analyzed by researchers to enhance the
overall performance of the system. The proposed model is more concentrated on the
extraction and fusion of deep features from the two views to improve screening
efficacy. The effectiveness of the proposed workflow is assessed on mammogram
images taken from the MLO view and CC views of the DDSM dataset. Medical
imaging data in conjunction with Machine Learning (ML) methods are employed for
breast cancer (BC) detection and classification, but they tend to be time-intensive.
Leveraging Deep Learning (DL) algorithms has the potential to further enhance the
detection accuracy.
This work focuses on improving the detection performance by using a fusion of texture
and Resnet 50 deep feature of MLO and CC view mammograms followed by Support
Vector Machine (SVM) classification. An improved accuracy of 98.1% is achieved
when compared to existing works. Henceforth, this work can be employed for the early
BC diagnosis.
Keywords: Breast cancer, Deep learning, Feature fusion, Detection, Classification, Accuracy, Mammogram, Texture, Deep features, Resnet 50.
