Artificial intelligence (AI) is increasingly shaping radiology, by performing tasks ranging from image reconstruction and workflow optimization to image analysis, reporting support, and multimodal clinical decision support. This narrative review provides a structured overview of the methodological foundations and clinical evolution of AI in radiology. We first outline the historical development from early rule-based expert systems and classical feature engineering to data-driven machine learning. We then explain the core principles of neural networks, including network depth, nonlinear activation functions, loss-based optimization, and backpropagation, before discussing convolutional neural networks (CNNs) as the dominant architecture for image-based learning in radiology. Particular emphasis is placed on inductive biases such as locality, parameter sharing, and translation equivariance, as well as on encoder-decoder architectures and U-Net-based approaches for segmentation. Subsequently, we discuss attention mechanisms, Transformer architectures, and their role in modeling long-range dependencies in medical imaging. We distinguish Vision Transformers, which process images as token sequences, from vision-language models, which align imaging data with textual information and enable multimodal tasks such as report generation, image-text retrieval, and visual question answering. Recent radiology-specific foundation models and large language models are considered in the context of workflow support, structured reporting, knowledge integration, and emerging multimodal image interpretation. Finally, we address key challenges for clinical translation, including data requirements, bias, generalizability, hallucinations, interpretability, validation, federated learning, and regulatory aspects such as the EU AI Act and the Medical Device Regulation. A basic understanding of these methodological principles is essential for radiologists to critically evaluate current AI systems and to contribute to their safe, transparent, and clinically meaningful implementation.
Purpose Photon-counting CT (PCCT) and spectral shaping enable new approaches in pediatric imaging, but the effects of clinically implemented tin-filtered pediatric chest protocols on radiation dose and image noise across image quality levels (IQLs) remain insufficiently characterized. This study compared a clinical Sn100 kVp ultra-high resolution (UHR) protocol with a non-tin-filtered 90 kVp standard-resolution protocol using anthropomorphic pediatric phantoms. Materials and Methods Three pediatric phantoms (1, 5, and 10 years) were scanned on a commercial PCCT system across IQLs 1-300 using the clinical Sn100 kVp UHR and non-tin-filtered 90 kVp standard-resolution protocols. Radiation dose and ROI-based image noise in lung parenchyma and bone were recorded. Results The Sn100 kVp UHR protocol showed lower lung-parenchymal noise across all phantom sizes and IQLs and generally lower bone noise. In the ultra-low-dose range (IQL 1-10), lung-parenchymal noise reductions ranged from 31% to 52% and bone-noise reductions reached 58%. Across most IQLs, the Sn100 kVp UHR protocol was associated with higher radiation dose. At high IQLs, dose plateaus occurred at approximately IQL 150 and 100 in the 5- and 10-year-old phantoms, followed by dose crossovers at approximately IQL 220 and 150, respectively. At very low IQLs, the Sn100 kVp UHR protocol showed a dose advantage up to IQL 5, 4, and 3 in the 1-, 5-, and 10-year-old phantoms, respectively. Conclusions The Sn100 kVp UHR protocol was associated with lower image noise but was not universally dose-saving compared with the non-tin-filtered 90 kVp standard-resolution protocol. In addition, the observed dose behavior was influenced not only by spectral shaping but also by detector readout mode and protocol-specific technical constraints. These findings may support future pediatric chest PCCT protocol optimization.
Purpose To evaluate radiation exposure in peripheral, abdominal, and cerebral digital subtraction angiography (DSA) across three consecutive equipment generations and to identify technological potential for dose reduction. Material and Methods In this retrospective single-center study, 450 patients were examined using three different platforms (150 per system): a conventional system (Allura Xper) and two subsequent generations featuring advanced noise reduction (Allura Clarity and Azurion). The patient cohort was equally distributed among head, abdomen, and lower limb regions. Primary endpoints included total dose-area product (DAP), fluoroscopic dose rate, dose per DSA frame, and the DAP per 3D rotational angiography (3D-RA) for the head. Subjective image quality was evaluated using a 5-point Likert scale. Results Transitioning from the conventional system to the first generation with advanced image processing (Allura Clarity) resulted in a significant reduction in total DAP of 70% (abdomen), 65% (head), and 69% (lower limbs; each p<0.001). This effect was driven by a significant decrease in fluoroscopic dose rate (34-50%) and dose per DSA frame (68-74%). Transitioning to the latest system generation (Azurion) yielded a non-significant yet further reduction of 6-17%. Cerebral 3D-RA demonstrated stable dose requirements across all three generations (p=0.123). Subjective image quality improved significantly with the introduction of noise reduction algorithms (p<0.001), with no significant difference between the two most recent systems (p>0.05). Conclusions Advanced image processing algorithms are an effective tool for radiation dose reduction in angiography. However, a technological dose plateau appears to have been reached, as the current equipment generation has not shown further significant improvements in radiation dose requirements compared to its immediate predecessor. Dose requirements for 3D-RA do not, at present, benefit significantly from the advantages of modern image processing algorithms.
Purpose To comprehensively evaluate anatomical, degenerative, and synovial factors associated with Baker cyst presence and volume on knee MRI.Materials and Methods This retrospective study included 219 patients who underwent knee MRI between January 2025 and December 2025 (119 with Baker cysts, 100 controls). Recorded variables included medial femoral condyle dimensions (distal and posterior articular diameters, width, length), patellofemoral/trochlear morphology, degenerative findings, effusion grade, the arrangement pattern between the medial head of the gastrocnemius and the semimembranosus tendon and Baker cyst volume. Group comparisons and multivariable logistic and log-linear models were performed. Results Patients with cysts had a significantly smaller posterior articular diameter of the medial condyle than the controls (median 29.8 vs. 32.5 mm, p<0.001), the strongest anatomical discriminator among condylar metrics. Meniscal degeneration, meniscal tear, and anterior cruciate ligament mucoid degeneration showed significant association with cyst presence. Cyst volume correlated positively with effusion (rho=0.38, p<0.001), medial condyle size parameters, and degenerative findings. Tendon arrangement morphology showed an ordinal trend with volume (concave -> flat -> convex). In multivariable analysis for cyst presence, the posterior articular diameter of the medial condyle remained an independent anatomical predictor. However, in the log-linear model for cyst volume, effusion emerged as the strongest determinant, with each grade increase associated with a 54% enlargement. The final multivariable model demonstrated good to excellent discrimination for predicting cyst presence (AUC=0.80-0.81). Conclusion Baker cyst presence is linked to meniscal and ligament degeneration with additional anatomical features, particularly reduced posterior articular diameter of the medial condyle, while cyst enlargement is primarily associated with effusion and tendon arrangement.
Purpose Diffusion MRI is sensitive to tissue abnormalities in multiple sclerosis (MS), but conventional region-of-interest measures reduce spatially structured signal patterns to a single summary value. We evaluated boundary-anchored thalamic apparent diffusion coefficient (ADC) profiling in MS using line-based edge-anchored profiling (LEAP). Materials and Methods In this retrospective matched-control study, age- and sex-matched controls (CTRL, n = 28) and participants with relapsing-remitting MS (RRMS, n = 14) or secondary progressive MS (SPMS, n = 14) were analyzed on the same 3T MRI scanner using the same diffusion protocol. LEAP was applied bilaterally at the medial thalamic surface adjacent to the third ventricle. Participant-averaged bilateral summary metrics and exploratory side-specific profile-shape metrics were derived from ADC curves. Results The clearest signal was observed in comparisons between MS and matched controls, with the largest number of differences seen for SPMS versus controls. Compared with controls, SPMS showed a higher ventricle-near mean ADC (p = 0.004), higher profile maximum ADC (p = 0.003), and a steeper early profile slope (p = 0.005). RRMS generally showed intermediate values. Direct RRMS-versus-SPMS comparisons of bilateral summary metrics were limited. Exploratory right-sided profile-shape metrics and left-right asymmetry measures suggested possible phenotype-related differences. Conclusion Boundary-anchored thalamic ADC profiling detected spatially structured abnormalities in MS relative to matched controls, with the strongest deviation in SPMS. Exploratory shape-based findings should be considered hypothesis-generating and suggest additional profile-level information beyond bilateral summary values.
