3.1.2. Experimental Setup

This study focuses on the effect of the proposed SbBDEM enhancement technique applied in the pre-processing to prepare the images for the subsequent stages. The performance was measured by comparing the performance with the system trained using original images and two established histogram-based enhancement techniques. The final classification stage used only the handcrafted learning features to reduce the overall computation, as the mass was already accurately detected and segmented from prior stages. To compare the breast density-wise performance, the initially randomized labeled image numbering was saved from the detection phase onto the following stages to make an unbiased comparison among the same test images. Additionally, a 5-fold cross-validation was performed on the classification stage to ensure the average of using all learning features to compare performance. These experiments were visualized and executed on a workstation equipped with CPU Intel(R) CoreTM i7-10870H 2.3 GHz with single GPU graphic card NVIDIA GeForce RTX2060 6GB, 16 GB RAM, and trained and tested on MATLAB (Natick, MS, USA).
