The rapid advancement of artificial intelligence (AI) has significantly impacted various aspects of healthcare, particularly in the medical imaging field. This review focuses on recent developments in the application of deep learning (DL) techniques to breast cancer imaging. DL models, a subset of AI algorithms inspired by human brain architecture, have demonstrated remarkable success in analyzing complex medical images, enhancing diagnostic precision, and streamlining workflows. DL models have been applied to breast cancer diagnosis via mammography, ultrasonography, and magnetic resonance imaging. Furthermore, DL-based radiomic approaches may play a role in breast cancer risk assessment, prognosis prediction, and therapeutic response monitoring. Nevertheless, several challenges have limited the widespread adoption of AI techniques in clinical practice, emphasizing the importance of rigorous validation, interpretability, and technical considerations when implementing DL solutions. By examining fundamental concepts in DL techniques applied to medical imaging and synthesizing the latest advancements and trends, this narrative review aims to provide valuable and up-to-date insights for radiologists seeking to harness the power of AI in breast cancer care.

Background: Hormone receptor (HR) status guides breast cancer therapy. Deep learning (DL) applied to contrast-enhanced mammography (CEM) might offer a noninvasive means for HR status prediction, but class imbalance challenges model development and assessment. This preliminary study investigates CEM-based DL for HR status prediction, focusing on class imbalance handling. Materials and methods: In this retrospective study, CEM tumor crops from 105 patients with invasive breast cancer were used. Patients were randomized into training ( n = 68), validation ( n = 16), and independent test ( n = 21) sets. A “Residual Network 18” (ResNet-18), pretrained on ImageNet, was fine-tuned using weighted cross-entropy loss and an Adam optimizer. Model selection used validation area under the precision-recall curve (AUPRC); output probabilities were calibrated via temperature scaling. Performance was reported with accuracy, area under the receiver operating characteristic curve (AUROC), and imbalance-aware metrics (balanced accuracy, Matthews correlation coefficient (MCC)) with 95% confidence intervals (1,000-iteration bootstrap). Results are presented for standard (0.5) and optimized (validation F1-score for HR-negative class) thresholds. Results: Validation AUPRC (model selection metric) was 0.640 (0.304–0.906). On the independent test set (optimized threshold 0.755), the model achieved 91.9% accuracy (86.5–97.3%), AUROC 0.808 (0.648–0.935), balanced accuracy 0.700 (0.550–0.853), and MCC 0.605 (0.296–0.818). Conclusion: A ResNet-18, utilizing patient-level data splitting and imbalance-aware fine-tuning, can capture CEM features for HR status, performing well despite significant class imbalance. Generalizability is limited by dataset characteristics and acquisition specifics, warranting further validation in larger, diverse cohorts to establish clinical applicability. Relevance statement: This work explores whether routinely acquired CEM images contain enough information for DL prediction of HR status. A ResNet-18 was trained with weighted loss and patient-level data splits; performance was quantified with imbalance-aware metrics to provide a realistic assessment in a highly skewed dataset, highlighting both the promise and current constraints of CEM-based molecular imaging. Key Points: A ResNet-18, optimized for class imbalance through weighted training and with calibrated probabilities, predicted HR positivity on CEM with 91.9% accuracy and AUROC 0.81 in an independent test cohort using an F1-tuned threshold. Balanced accuracy (0.70) and MCC (0.60) demonstrate maintained discrimination despite an approximate 85% class imbalance (HR-positive cases). Patient-level splitting was employed to ensure robust evaluation. Limitations related to the dataset’s scope and specific imaging protocols may influence broader generalizability. Graphical

Integrating vision-language models (VLMs) into clinical radiology workflows requires exporting two-dimensional images that preserve diagnostic viewing context, including imaging plane, slice position, window/level, slab parameters, and overlays. Existing approaches typically rely on ad hoc screenshots and external uploads, which discard this context, reduce reproducibility, and introduce workflow friction through task switching. We describe a radiology viewer architecture that maintains model interaction entirely within the diagnostic environment by serializing explicit viewer state into replayable descriptors. Requested views are re-rendered offscreen using shared in-memory voxel data, enabling repeatable, context-faithful evidence generation without disrupting interactive interpretation. The user interface employs a split workspace, combining a diagnostic grid for image review with an integrated chat sidebar for evidence staging and model interaction. Capture actions generate batches of serialized viewer states that are executed by an offscreen renderer; the resulting frames are staged for user curation prior to model submission. Under controlled conditions at a 1,000 × 1,000 pixel output resolution, benchmarks demonstrated perfect reproducibility: repeated captures from identical serialized states produced pixel-identical images, with zero hash mismatches across 50 frames. Capture latency ranged from a median of 4.49 to 16.69 ms per frame, depending on view type, and median output size was 27.9 kB per frame using a high-efficiency image container. These results indicate that the proposed architecture supports reproducible, low-latency, and storage-efficient multislice and multiseries evidence assembly at interactive speeds, enabling seamless integration of VLMs into diagnostic radiology workflows. Relevance statement: Offscreen re-rendering driven by serialized viewing state enables context-preserving and reproducible export of two-dimensional evidence from volumetric radiology viewing, supporting integrated vision-language model workflows with millisecond-scale capture latency from hot cache. Key Points: A split-layout DICOM viewer with an integrated chat interface enables radiologists to capture, stage, and submit visual evidence without leaving the diagnostic environment. Offscreen re-rendering from serialized viewing state produces pixel-identical exports across repeated captures, supporting traceable, auditable, and reproducible evidence generation. Batch capture efficiently assembles multislice, multiplane, and multiseries evidence with minimal manual effort, providing models with richer and more representative study context. Capture latency was millisecond-scale, and high-efficiency image container (HEIC) encoding reduced file size by approximately 2.8-fold compared with the joint photographic experts group (JPEG) format, supporting responsive multi-frame workflows.