This paper presents a semi-supervised learning framework for breast cancer classification based on features originated from MammoWave, a microwave imaging system that illuminates the breast with electromagnetic fields and measures the resulting scattered signals. To address the possibility of using both labeled and unlabeled data during the machine learning process, we propose an Augmented Laplacian Support Vector Machine (Aug-Lap SVM) that integrates Label Spreading and feature augmentation based on Spectral Embedding and Kernel Principal Component Analysis with a SVM. This hybrid model leverages both labeled and unlabeled data through graph-based Manifold Learning, capturing the underlying data structure to improve classification performance. On the MammoWave dataset, the proposed Aug-Lap SVM achieves 72% accuracy and a weighted F1-score of 0.74, significantly outperforming the traditional SVM baseline (54% accuracy, weighted F1-score: 0.54). It also achieves 92% recall for benign cases (class 1) and reduces false negatives by 47%, demonstrating strong performance in class-imbalanced conditions. These results highlight the effectiveness of graph-based semi-supervised learning for microwave breast cancer imaging and support the clinical potential of MammoWave in scenarios characterized by label scarcity, high class imbalance, or desire to include in the learning process data that is temporarily unlabeled.