Enhancing diagnostic precision and mammogram classification in breast cancer detection utilizing an optimized Explicit Feature Interaction Aware Graph Neural Network

Breast cancer analysis is vital for women’s health, directly affecting mortality rates. Mammography, the most reliable diagnostic method, can be subject to inaccuracies due to manual interpretation by radiologists. Digital mammograms, along with computer-aided diagnostic (CAD) systems, improve diagnostic accuracy by reducing false positives and increasing survival rates. This research proposes an enhanced method for mammogram classification in breast cancer detection using an Optimized Explicit Feature Interaction Aware Graph Neural Network (EFIAGNN-LWO). Initially, the input image is gathered from the Curated Breast Imaging Subset of Digital Database for Screening Mammography (CBIS-DDSM) dataset. The image undergoes pre-processing with an Adaptive Two-Stage Unscented Kalman Filter (ATUKF) to improve contrast, reduce noise, and improve histogram distribution in low-contrast mammograms. The preprocessed images are fed to the Multi-Hypothesis Fuzzy-Matching Radon Transform (MHFMRT) to extract texture features. These features are then passed to the Explicit Feature Interaction Aware Graph Neural Network (EFIAGNN), which classifies the mammograms as benign or malignant. The method is further optimized using Leaf in Wind Optimization (LWO) to improve classification accuracy. The proposed method is implemented in Python, achieving better performance with a F1 score of 97.33 %, precision of 99.39 %, an accuracy of 99.20 %, an and a sensitivity of 99.80 %. The proposed method shows notable recall improvements over existing methods like Inception-V4, Bidirectional Long Short-Term Memory (BiLSTM), DenseNet121, and attaining 98.23 % for benign instances and 99.01 % for malignant cases. The model also achieves a reduced computational time of 61 s, enabling quicker diagnosis and more efficient decision-making in clinical environments.