A Unified Taxonomy of AI Techniques for Breast Cancer Detection: Integrating Learning Paradigms, Imaging Modalities, Evaluation Metrics, and Deployment Challenges

The past decade has seen an increase in the use of artificial intelligence (AI) in breast cancer detection, which can only be called exponential, and it has revolutionized the accuracy and efficiency of breast cancer diagnosis. machine learning (ML) and deep learning (DL) models have been extensively applied in various imaging modalities, including mammography, ultrasound, magnetic resonance imaging (MRI), and even histopathology. It is also against this background that though there has been this kind of advancement, there are no systemic methodologies that have been established to classify and analyse these AI-grounded strategies in a more holistic kind of way. In general, in the present paper, we propose a unified and extended taxonomy in accordance with which AI approaches in the field of breast cancer detection might be classified as having four basic dimensions: learning paradigm (e.g., supervised, unsupervised, reinforcement learning), model architecture (e.g., SVM, CNN, GAN), diagnostic task (e.g., classification, segmentation, detection), and imaging modality. A systematic review of 25 recently published and peer-reviewed studies bought this taxonomy to the ground. This is a review that critically analyses the model performance, dataset usage and evaluation metrics, accuracy, area under curve (AUC), F1-score, precision and recall. In addition, the proposed taxonomy gives the association of specific AI methods with well-known publicly available datasets, such as BreaKHis, DDSM, MIAS, and BUSI, allowing one to fully understand the adequacy of models by imaging type. It displays tendencies, prejudices and not effectively used combinations of models and modalities as well. The paper describes the presence of critical open problems in the area, such as interpretability of models, generalizability, computational efficiency, and, whether they can be used in clinical or low-resource environments or not. The work can be used as a helpful reference to guide researchers, developers, and clinicians in the development of useful, efficient, and explainable AI-based diagnosis systems of breast cancer by providing a systematic taxonomy and highlighting the state-of-the-art plus the ongoing challenges. One may think of the taxonomy not as a summary of the current knowledge as of the paper publication date but as a road map of what is expected to happen in the sphere of AI-powered oncology.