<p>Breast cancer represents a critical global health challenge, necessitating early and precise detection to enhance patient outcomes. Traditional manual detection methods for breast cancer from medical images are not only time-consuming but also susceptible to human error. Accurate classification of cancer stages is essential for effective treatment planning and post-diagnosis management. This study introduces a novel meta-learning approach aimed at classifying breast cancer images, particularly in scenarios with limited labeled data. Conventional classification techniques often falter when faced with insufficient labeled data; however, meta-learning addresses this limitation by enabling rapid adaptation to new tasks using minimal examples. The proposed methodology incorporates image segmentation to delineate regions of interest, followed by sophisticated feature extraction to capture critical information. During the meta-training phase, a classifier is refined within a metric space utilizing cosine distance and a flexible scale parameter. The effectiveness of the proposed method is evaluated through various performance metrics, achieving an accuracy of approximately 96.10%, precision of 96.40%, recall of 96.37%, and an F1-score of 96.37%. Comparative analyses reveal that the proposed approach outperforms existing methods in classification accuracy and efficiency, particularly under conditions of limited data availability. This innovative method holds significant promise for the early detection of breast cancer and has the potential to enhance diagnostic accuracy in practical medical settings.</p>