Introduction to the JMI Special Issue on Advances in Breast Imaging

Abstract.

The editorial introduces the JMI Special Issue on Advances in Breast Imaging, reflecting on the current forefront of breast imaging research.

This special issue of the Journal of Medical Imaging (JMI) brings together 15 studies reflecting the current forefront of breast imaging research. While inspired by the 17th International Workshop on Breast Imaging (IWBI), held in Chicago in June 2024,¹ contributions were welcomed from investigators worldwide regardless of conference participation. This special issue demonstrates that IWBI remains central to fostering innovation and community in breast imaging, bridging technical and data science advances to improve cancer detection and outcomes.

Artificial intelligence (AI) and deep learning are major threads across this special issue. Olinder et al.² leverage AI to monitor mammographic density changes across screening rounds, linking density evolution to future cancer risk. Similarly, Kuling et al.³ examine longitudinal changes in breast density and parenchymal enhancement in high-risk women, showing the influence of age, menopause, and hormonal factors through deep learning-based segmentation of MRI data. Workflow and diagnostic accuracy are explored by Timberg et al.,⁴ who demonstrate that adding prior mammograms to digital breast tomosynthesis (DBT) can reduce false-positive recalls while affecting sensitivity. Yang et al.⁵ perform texture analysis across DBT vendors, exposing significant image differences and underscoring the need for vendor-neutral AI algorithm development. Dahlblom et al.⁶ propose AI-driven triage to target DBT readings to high-risk cases, thus maintaining accuracy and expanding cancer detection without saturating radiologist workload.

Synthetic data and simulation are recurring motifs. Van Camp et al.⁷ find that augmenting model training with simulated microcalcification clusters improves detection in contrast-enhanced mammography when real data is limited. Chen and Martel⁸ combine self-supervised Swin Transformers and convolutional neural networks, reporting improved breast cancer discrimination with limited labeled data. Multimodal and radiomics approaches further diversify the field. Muhtadi and Gallippi⁹ show that using both ultrasound B-mode and Nakagami images in deep learning improves breast tumor classification. Dong et al.¹⁰ (featured on the issue cover) present robust radiomic feature selection for classifying breast density grades, with good external generalizability using their RFE-SHAP methodology. Reader agreement is vital in clinical translation. Romanov et al.¹¹ compare percent density estimates from AI and expert readers, demonstrating higher inter-observer agreement with automated AI tools, and equivalent performance in risk assessment.

Reliability in image-based classification is the focus of Whitney et al.,¹² who introduce sureness metrics to assess CADx performance for distinguishing pure DCIS from mixed IDC/DCIS using MRI features. This sureness-based evaluation provides nuanced lesion-level reliability insights that can strengthen decisions made with AI tools. Image restoration and simulation address technical challenges. Brandão et al.¹³ show that denoising and image restoration in low-dose mammography can simulate higher-dose image quality, enhancing microcalcification detectability without added radiation. Osuala et al.¹⁴ apply generative adversarial networks to simulate dynamic contrast enhancement in breast MRI, enabling virtual contrast imaging and robust tumor segmentation even when real data are limited. Lyu et al.¹⁵ use hybrid simulation for breast CT to assess microcalcification detectability, finding that deep learning model observers outperform humans for small calcification clusters. Lee et al.¹⁶ propose adaptive DBT protocols using scout scans to personalize scan angular range according to lesion type and breast tissue parameters, potentially improving diagnostic yield.

The assembled studies confirm that breast imaging research is progressing through integration—of multimodal data,⁹ simulation and synthetic data,^7,13–15 workflow optimization,⁶ robust radiomics and explainable AI,¹⁰ and standardized assessment protocols.¹¹ These contributions collectively aim to improve cancer risk assessment, detection sensitivity, and clinical workflow efficiency, while reinforcing reproducibility and interoperability.⁵ Moving forward, focus on classification reliability,¹² optimal use of synthetic data, and adaptive, patient-relevant imaging¹⁶ will enable personalized screening and diagnosis with greater accuracy and robustness.

We thank the authors who contributed to this special issue for their excellent work, and we invite interested researchers, clinicians, and innovators to join us at the next International Workshop on Breast Imaging scheduled for June 2026 in Thessaloniki, Greece.¹⁷