Feature-driven breast cancer classification via hybrid model using mammogram images

Deep learning's swift development has generated substantial excitement about its application in medical imaging. Machine learning (ML) methods can support radiologists in diagnosing breast cancer (BC) without resorting to invasive procedures. However, traditional ML classifiers require the extraction of detailed hand-crafted features, which is a time-intensive task to achieve accurate results. Hence, this paper proposes a novel Feature-driven Breast Cancer Classification using the Modified Loss and Activation function-assisted LeNet (MLAL) model, named F-BCC-ML. The process of detecting BC using mammogram images comprises several key stages. In the first step, the image undergoes enhancement using the Improved Bilateral Filtering Technique (IBFT), which reduces the noise while conserving critical structural details like edges. Next, the image is subjected to segmentation using SegNet, a deep-learning model designed for semantic segmentation. After segmentation, the next phase is feature extraction, where various features like Weber Local descriptor assisted Local Gabor XOR Pattern (WLD-LGXP) for texture analysis, Median Binary Pattern (MBP), colour features, and deep features are derived from the segmented image. Once the features are extracted, they are fed into the classification stage, where the Modified Loss and Activation function assisted LeNet (MLAL) model, more sophisticated Deep Convolutional Neural Network (DCNN) are used to classify the image as either normal or cancerous. The result is a prediction that indicates whether the breast tissue is benign or shows signs of cancer, helping radiologists make more accurate and informed decisions. The MLAL+DCNN accomplished the maximum accuracy of 0.936, precision of 0.947 and F-measure of 0.942, respectively.