Agentic artificial intelligence(AI) represents a structural transition from task-specific tools to autonomous systems capable of goal-oriented, multi-step reasoning. However, translating this potential into radiology requires a fundamental shift from text-dominated architectures to avision-first approachthat engages the specialty’s diagnostic core. In this paper, we characterize the multimodal challenge distinguishing radiology from text-centric domains, where high-dimensional images serve as the primary diagnostic substrate rather than supplementary data. We reviewthree image-text coupling architectures, their clinical capabilities, limitations, and deployment feasibility. These include output-level coupling through sequential image-text pipelines, representation-level coupling through contrastive aligned dual encoders, and token-level coupling through generative vision-language models. To guide development, we introduce afive-level taxonomy of autonomy(Levels 1–5), establishing clear validation checkpoints and regulatory expectations for increasing machine independence, from text-based workflow automation to fully autonomous closed-loop systems. We emphasize that interoperability via standardized protocols, includingDICOMweb, HL7-FHIR, Model Context Protocol (MCP), and Agent2Agent (A2A), is as critical as the AI models themselves. Finally, we address critical implementation guardrails, including mitigatingalgorithmic hallucinations, defending againstadversarial prompt injection, and preserving human diagnostic proficiency amidst the risk of AI-induced skill decay. This comprehensive roadmap provides a structured path for integrating autonomous visual reasoning into modern medical imaging environments.
Objective Breast cancer is the most frequently diagnosed cancer type in women, and early detection and treatment significantly improve survival rates.In conventional imaging methods (mammography and ultrasound), lesions classified as BI-RADS 0 or 4 often lead to a high rate of unnecessary biopsies, resulting in both superfluous invasive procedures for patients and increased costs for the healthcare system.The aim of our study is to investigate the diagnostic validity of the Kaiser score in predicting malignancy in patients who underwent breast MRI following inconclusive or suspicious findings (BI-RADS 0 or 4) on conventional imaging, based on pathology results or clinical follow-up. Materials and methods Between September 2022 and April 2024, 99 patients who underwent breast MRI due to inconclusive results from conventional breast imaging methods (mammography and breast ultrasound) or the detection of suspicious findings (BI-RADS 0 or 4), and who had either histopathological evaluation following biopsy under US or MG guidance or were followed for at least 12 months, were included in this single-center retrospective study.The 99 patients included in the study were divided into two groups based on histopathology or follow-up outcomes (benign lesion group and malignant lesion group).Comparisons were made between these two groups. Results When comparing benign and malignant groups, statistically significant differences were observed for age (p < 0,001), age at menarche (p = 0,004), breastfeeding status (p = 0,009), menopausal status (p = 0,017), and use of hormone replacement therapy (p = 0,036).MRI BI-RADS (p < 0,001) and Kaiser scores (p < 0,001) were notably higher in the malignant group.Kaiser score was identified as the strongest diagnostic predictor of malignancy (AUC = 0,977, 95 % CI: 0.955–0.999; p < 0,001).Using a cut-off value of Kaiser score > 5, the sensitivity and specificity were 97 % (95 % CI: 84.2 %-99.9 %) and 87.9 % (95 % CI: 77.5 %-94.6 %), respectively.The MRI BI-RADS ranked second with a cut-off value > 4 (AUC = 0.878, 95 % CI: 0.805–0.950; p < 0.001); at this threshold, the sensitivity and specificity were 60.6 % (95 % CI: 42.1 %-77.1 %) and 100 % (95 % CI: 94.6 %-100.0 %), respectively.In multivariable logistic regression, higher Kaiser score (OR = 4.25, p < 0.001) and older age (OR = 1.16, p = 0.041) were associated with higher odds of malignancy, whereas older age at menarche was associated with lower odds (OR = 0.347, p = 0.030). Conclusion For breast lesions classified as BI-RADS 0 or 4 on conventional imaging, the Kaiser score provides excellent diagnostic accuracy and may support lesion stratification when used with MRI BI-RADS assessment; biopsy decisions should remain based on integrated clinical and imaging findings.
Adolescent idiopathic scoliosis (AIS) is a complex three-dimensional (3D) spinal deformity. Although standing full-spine radiography remains the cornerstone of clinical diagnosis and longitudinal follow-up, two-dimensional projections cannot fully characterize vertebral rotation, local osseous geometry, or global spatial alignment. In recent years, computed tomography (CT), magnetic resonance imaging (MRI), conventional radiography, EOS/biplanar radiography, 3D ultrasound, and surface imaging have increasingly been integrated with deep learning, statistical shape modeling, and generative approaches, substantially expanding the technical pathways for patient-specific 3D reconstruction in AIS. However, substantial heterogeneity remains across studies with respect to input constraints, target populations, reference standards, and validation endpoints, and the evidentiary boundaries between technical performance, AIS-specific validity, and clinical validity have not been systematically clarified.This review adopted a structured methodological framework and systematically searched the Web of Science Core Collection, PubMed/MEDLINE, and IEEE Xplore, supplemented by forward and backward citation tracking. Study-level evidence was synthesized across imaging inputs, reconstruction tasks, algorithmic paradigms, and validation designs. A total of 115 core studies were included, comprising 32 AIS-specific studies (27.8%), 21 scoliosis-specific or mixed-population studies (18.3%), and 62 studies based on general spinal-image-analysis evidence (53.9%). Based on the reference standards and validation endpoints used, we established a five-level validation framework ranging from image/segmentation-level validation (L1), anatomical-structure-level validation (L2), and clinical-parameter-level validation (L3), to workflow/generalization-level validation (L4) and clinical-translation-level validation (L5). An evidence-gap matrix based on “imaging modality × methodological task” was also constructed.Among the 115 studies, 28 (24.3%) reached a maximum of L1, 49 (42.6%) reached L2, 34 (29.6%) reached L3, and only 4 (3.5%) reached L4; no study met the L5 criteria, and no explicit AIS-specific external validation was identified. Current evidence indicates that CT provides a strong basis for high-fidelity osseous reconstruction, EOS has accumulated relatively continuous AIS-specific evidence for weight-bearing 3D parameter assessment, and 3D ultrasound offers advantages for repeated radiation-free parameter measurement, whereas MRI-based reconstruction and limited-projection 3D recovery remain rapidly evolving fields. The principal challenge has therefore shifted from whether a 3D structure can be generated to whether structural fidelity can be translated into reliable clinical parameters and whether models can generalize consistently across independent AIS populations, imaging systems, and acquisition protocols. Future work should prioritize AIS-specific multicenter validation resources, task-matched 3D reference standards, and a continuous evidence chain linking structural fidelity, clinical-parameter validity, external generalization, and clinical impact, thereby advancing AIS 3D reconstruction from technical feasibility toward trustworthy clinical application.
Introduction Emerging evidence indicates that breast ultrasound can function as an effective stand-alone diagnostic tool for focal breast complaints, with digital breast tomosynthesis (DBT) reserved for inconclusive cases. This study identifies and prioritizes radiologists’ and radiographers’ professional considerations regarding an ultrasound-first approach. Methods Using digital concept mapping, radiologists and radiographers independently generated, categorized, and rated statements related to the implementation of ultrasound as the initial modality. Thematic clusters were derived and prioritized based on participant ratings (7-point scale). Results Radiologists produced 43 statements across seven clusters: 1) Diagnostic accuracy, 2) Logistics and planning, 3) Communication, 4) Guidelines, 5) Patient satisfaction, 6) Age and screening, and 7) Accessibility of approach, prioritizing the clusters Diagnostic accuracy (rating 5.41), Guidelines (rating 5.21), and Logistics and planning (rating 4.88). Radiographers produced 35 statements in four clusters: 1) Communication, 2) Logistics and planning, 3) Organization and responsibilities, and 4) Patient satisfaction, placing greatest emphasis on the clusters Organization and responsibilities (rating 4.91) and Communication (rating 4.74). These patterns indicate that radiologists primarily emphasized diagnostic and clinical feasibility, whereas radiographers focused more on operational and role-related considerations. Conclusion Radiologists and radiographers identified both shared and profession-specific considerations related to an ultrasound-first approach. Differences in prioritized themes reflect distinct professional roles and highlight the complexity of changing established imaging pathways. These findings may inform future research and implementation efforts by drawing attention to profession-specific considerations relevant to the adoption of such pathways.
