Breast Cancer Detection Based on Texture Analysis and Deep Learning Techniques

Abstract

Breast cancer is one of the leading causes of cancer-related mortality among women worldwide. Early and accurate diagnosis is essential for improving patient survival and reducing treatment complexity. Mammography remains the gold standard imaging modality for breast cancer screening; however, the interpretation of mammographic images is often challenging due to low contrast, dense breast tissue, image noise, and the subtle appearance of suspicious lesions. This thesis presents ONCO-VISION, an intelligent Computer-Aided Detection (CAD) framework designed to improve breast lesion detection and segmentation by integrating handcrafted texture analysis with deep learning. The proposed approach combines Laws’ Texture Energy Measures (TEM) with the YOLOv11m-seg architecture to enhance the visibility of suspicious mammographic structures, including microcalcifications, spiculated masses, and architectural distortions. The preprocessing pipeline consists of DICOM image normalization, Contrast Limited Adaptive Histogram Equalization (CLAHE), and multi-channel texture feature extraction using Laws’ kernels (L5S5, E5L5, E5S5, and R5R5). These texture-enhanced representations are combined into a five-channel input that improves lesion characterization and localization. The proposed framework was evaluated using the publicly available INbreast dataset, which contains 410 full-field digital mammography images with expert annotations. Performance was assessed using standard object detection metrics, including Precision, Recall, F1-score, and mean Average Precision (mAP). Experimental results demonstrate that integrating texture enhancement with deep learning improves the detection and localization of subtle breast lesions. Although challenges remain, particularly in dense breast tissue and under class imbalance, the proposed ONCO-VISION framework highlights the potential of combining handcrafted texture descriptors with modern deep learning techniques to develop more accurate and reliable computer-aided breast cancer diagnosis systems