Integrative Feature Learning from Mammographic Images Using a Hybridized Diagnostic Model

Abstract

Traditional Conventional machine learning approaches to breast cancer diagnosis tend to rely on manual feature extraction, which is fraught with variability and time-consuming. These constraints limit the scalability and consistency of diagnostic platforms. As the need for precise and efficient diagnostic technologies escalates, the need for automated frameworks that can effectively interpret mammographic data and facilitate clinical decision-making arises. This paper introduces the Feature-driven Breast Cancer Classification Model (F-BCC-ML), which is intended to optimize diagnostic precision and efficiency in the detection of breast cancer. The goal is to create a hybridized model that can automate feature extraction and classification from mammogram images and eliminate the dependency on manual techniques to improve clinical results. The F-BCC-ML model combines an Adaptive Classifier Engine (ACE) with a strong preprocessing and segmentation pipeline. First, preprocessing of mammogram images is done for noise reduction using the Improved Bilateral Filtering Technique (IBFT), which maintains important anatomical information. Segmentation is achieved through SegNet, a deep learning architecture optimized for semantic segmentation. Feature extraction merges texture descriptors Weber Local Descriptor-harmonized Local Gabor XOR Pattern (GTE) and Grid Feature Encoding (GFE) with color and deep features from region-segmented regions. The features are subsequently classified as normal or cancerous via the ACE architecture in combination with a Deep Vision Network (DVN). The F-BCC-ML model showcases strong clinical promise through the automation of the diagnostic process and feature representation improvement.