Investigative Radiology最新文献

Purpose: To investigate how artificial intelligence-assisted chest radiograph interpretation influences eye gaze patterns in novice and expert radiologists.

Materials and methods: This prospective eye tracking study included 6 novice radiology residents and 6 expert cardiothoracic radiologists. Fifty anonymized posteroanterior chest radiographs were interpreted under 3 conditions: Before AI, With AI, and After AI. An AI-Rad Companion Chest x-ray algorithm generated visual annotations and confidence scores. Eye movements were recorded at 250 Hz. Fixation sequences were analyzed using a 7 × 7 grid and quantified with normalized Levenshtein distance as a measure of systematic viewing behavior. Linear mixed-effects models assessed the effects of condition, expertise, and their interaction.

Results: Viewing patterns were significantly less systematic in the With AI and After AI conditions compared with Before AI. A significant interaction between condition and expertise was observed. Novice radiologists demonstrated markedly increased gaze dissimilarity during AI-assisted interpretation and remained less systematic after AI exposure. Expert radiologists showed no significant differences in gaze behavior across conditions, maintaining stable and systematic viewing patterns.

Conclusions: AI assistance disrupted systematic visual search behavior in novice radiologists but not in experts. These findings indicate that AI exposure may interfere with the stability of foundational perceptual search strategies in trainees, whereas experts integrate AI without altering established viewing patterns. Implications for radiology training warrant careful consideration.

Background: Gadolinium-based contrast agents (GBCAs) are crucial for enhancing MRI contrast, especially in neurological and oncological imaging. Achieving high relaxivity is a primary focus in contrast media research. Recently, digadoglucitol, the first dimeric macrocyclic gadolinium complex, was introduced, offering significantly higher relaxivity compared with first-generation commercial macrocyclic GBCAs.

Purpose: This study assesses the performance of digadoglucitol in preclinical models of neurological and oncological diseases. It explores whether digadoglucitol provides comparable diagnostic efficacy at half the dose and superior contrast enhancement at the full standard dose compared with gadoterate meglumine and gadobutrol.

Material and methods: All animal experiments and preclinical procedures adhered to national and international regulations. The study involved models of glioma, meningioma, ischemia, and breast adenocarcinoma, for a total of 112 animals. Digadoglucitol was administered at doses of 0.1 mmol Gd/kg and 0.05 mmol Gd/kg, whereas gadoterate meglumine and gadobutrol were given at 0.1 mmol Gd/kg. MRI sessions assessed signal enhancement and contrast-to-noise ratio (CNR) before and after contrast-agent administration.

Results: All pathologic models were accurately localized with consistent volumes across treatment groups. Tumour malignancy and ischemic lesions were confirmed by histologic analysis using standard haematoxylin-eosin staining and established pathologic criteria. Quantitative analysis revealed that both half and full doses of digadoglucitol produced significant signal enhancement and CNR in glioma, meningioma, ischemia, and breast models. The enhancement and CNR values were comparable (at half-dose) or exceeded (at full-dose) those achieved with gadoterate meglumine and gadobutrol. Statistical analysis confirmed the significance of these results.

Conclusion: Digadoglucitol shows high diagnostic efficacy in preclinical models of neurological and oncological diseases. It delivers comparable effectiveness at half the dose and superior contrast enhancement at the full standard dose compared with established GBCAs. These preclinical findings suggest that digadoglucitol is a promising contrast agent to be considered for clinical translation, potentially allowing for reduced dosage while maintaining or improving diagnostic accuracy.

Background: Metallic implants can cause relevant artifacts in computed tomography (CT) imaging, affecting the quality and diagnostic utility of scans. Previous advancements in metal artifact reduction techniques have shown promise but still exhibit limitations in artifact reduction, particularly close to metal implants.

Purpose: To evaluate a novel, advanced iterative metal artifact reduction (iMAR) algorithm for photon-counting detector CT in an experimental study focused on visualizing the vicinity of a fixation nail implant.

Methods: Three bovine femur bones with titanium-based trochanteric fixation nail implants were scanned on a clinical photon-counting detector CT scanner. Images were reconstructed (1) without iMAR, (2) with the current iMAR algorithm, and (3) with a new prototype iMAR algorithm. The new iMAR prototype algorithm advances state-of-the-art iMAR for photon-counting detector CT by utilizing intrinsically available spectral information. Attenuation and artifact severity (SD of attenuation) were quantified by placing regions-of-interest on each reconstruction across 3 different axial slices: One in the bone marrow immediately adjacent to the metal implant and one in the water adjacent to the femur with the implant. Qualitative image quality, newly introduced artifacts, and diagnostic confidence were rated by 3 radiologists using 5-point Likert scales. Differences between reconstructions were tested using the Friedman test with Wilcoxon post hoc tests; interreader agreement was assessed using Krippendorff alpha.

Results: Artifact severity in the bone adjacent to the implant significantly decreased from 226 HU (no iMAR) to 174 HU (current iMAR) and to 159 HU with the new iMAR (P < 0.001). Adjacent to the femur, artifact severity decreased from 63 HU to 48 HU and to 29 HU, respectively (P < 0.05). Qualitative scores differed significantly between reconstructions (P < 0.05), with highest ratings for new iMAR across all categories. Current iMAR introduced new artifacts near the implant, which did not occur with new iMAR (P < 0.05).

Conclusion: Experimental evidence from a bovine femur implant model suggests that a new, advanced iterative metal artifact reduction algorithm leveraging intrinsic spectral information from photon-counting detector CT effectively reduces metal artifacts and further improves the visualization of the metal-bone interface. Thus, this technique has the potential to enhance the assessment of implant-related complications such as aseptic loosening.

Objectives: To propose and validate a simplified method for 3D simultaneous post-contrast parametric mapping and synthetic late gadolinium enhancement (LGE) imaging at 0.55T for comprehensive whole-heart myocardial tissue characterization.

Materials and methods: A 3D joint T1/T2 mapping research sequence is adopted from a previous study. Three interleaved volumes with inversion recovery (IR) preparation, no magnetization preparation, and T2 preparation were acquired with image navigators to enable 100% respiratory scan efficiency. Intrinsically co-registered 3D T1, T2, and proton density maps were calculated using a dictionary-matching method, and Bloch equation-based IR and T2 preparation-IR (T2IR) signal models were proposed to generate multi-contrast 3D synthetic LGE images. In vivo evaluation included 10 data sets from a porcine myocardial infarction model to validate the performance of the proposed 3D method in comparison with that of separately scanned 2D reference sequences including post-contrast T1 mapping, pre-contrast T2 mapping, and LGE.

Results: For the 10 swine data sets, 2D and 3D T1/T2 maps had consistent findings regarding the changes in T1/T2 values of myocardial infarction, presenting significantly decreased post-contrast T1 (2D: 279±48 vs. 472±44 ms, P<0.01; 3D: 355±32 vs. 597±48 ms, P<0.01) and increased T2 (2D: 102.4±11.5 vs. 66.4±3.1 ms, P<0.01; 3D: 71.0±5.3 vs. 39.4±4.5 ms, P<0.01) in scar compared with remote myocardium. 3D multi-contrast LGE images were successfully generated without additional scan and provided excellent image contrasts. Compared with 2D LGE, 3D synthetic bright-blood IR-LGE had improved scar-to-myocardium contrast (P<0.01) with comparable image contrasts of scar-to-blood (P=0.08) and blood-to-myocardium (P=0.71), synthetic gray-blood IR-LGE had improved scar-to-blood and scar-to-myocardium contrast (P<0.01) with comparable blood-to-myocardium contrast (P=0.06), whereas synthetic dark-blood T2IR-LGE demonstrated significant differences regarding all tissue contrasts (P<0.01).

Conclusions: The proposed method provided imaging findings consistent with 2D references and shows promise for comprehensive myocardial tissue characterization in a single simple scan.

Objectives: Intramyocardial hemorrhage (IMH) is the most severe form of injury associated with reperfusion therapy during acute myocardial infarction (AMI). Although T2*-weighted cardiac MRI is regarded as the reference standard for IMH detection, its application is limited in the hyperacute phase. Ferumoxytol, an iron-based contrast agent characterized by potent T1-shortening and a long intravascular half-life, may enable earlier detection. This study assessed a T1-weighted approach using ferumoxytol-enhanced MRI (FE-MRI) for early detection of IMH after ischemia-reperfusion injury in a swine model.

Materials and methods: IMH was induced in 22 Yorkshire swine using a closed-chest ischemia-reperfusion model with intracoronary collagenase administration. FE-MRI was performed immediately after reperfusion using T1-weighted cine imaging and serial T1 mapping with a modified Look-Locker inversion recovery (MOLLI) pulse sequence. Imaging findings were compared among IMH-positive (IMH+), IMH-negative (IMH-), and control animals. Results were validated through gross pathology, histology, and electron microscopy.

Results: Nine animals completed imaging, of which 5 were IMH+ and 4 IMH- based on histopathology and FE-MRI. In IMH+ animals, the T1 within hemorrhagic reperfused myocardium was significantly lower than remote myocardium [420.8 ms (380.5, 656.3) vs. 806.0 ms (781.0, 818.8); P <0.001], and lower than healthy myocardium in the control animal [808.2 ms (796.3, 811.0)]. IMH- animals exhibited significantly higher myocardial T1 than IMH+ animals in both non-hemorrhagic reperfused [807.8 ms (786.1, 830.3); P <0.001] and remote myocardium [805.6 ms (786.1, 825.5); P <0.001]. A mixed-effects model confirmed significant cohort-specific T1 decrease within hemorrhagic reperfused tissue of IMH+ animals (estimate= -0.49±0.15; P =0.001), with a significant time-dependent effect (β= -0.125±0.03; P <0.001). Remote myocardial and left ventricular blood pool T1 did not differ between groups (both P >0.05). Gross and microscopic findings confirmed extravasated erythrocytes and disrupted myocardial architecture, consistent with hemorrhagic injury.

Conclusions: This proof-of-concept study demonstrates the feasibility and potential of T1-weighted FE-MRI as a unique approach for detecting IMH immediately after ischemia-reperfusion injury. Early identification of IMH by FE-MRI may facilitate translational efforts to develop and evaluate targeted therapies that reduce IMH burden and improve post-AMI outcomes.

Background: Despite the growing number of artificial intelligence (AI)-based applications used in radiology, no structured framework exists to assess their case-level reliability or to document overridden outputs in reports.

Purpose: To develop and evaluate the Artificial Intelligence Reporting and Data System (AI-RADS), a structured framework for an objective, case-level assessment of AI output reliability, clinical utility, and recommended actions in radiology.

Materials and methods: The AI-RADS framework was tested in a retrospective, multireader study. Here, 5 board-certified radiologists independently evaluated 350 cases processed by 7 representative AI applications for image-based and generative tasks. Each case was assigned one of 5 AI-RADS categories, applicable modifiers, and an independent correctness rating as a reference. Interreader agreement was quantified using Krippendorff's α with 95% CIs.

Results: Substantial interreader agreement was observed for the core AI-RADS categories in both image-based (Krippendorff's α=0.87; 95% CI: 0.83-0.91) and generative AI tasks (Krippendorff's α=0.93; 95% CI: 0.91-0.95). Reader-assigned correctness aligned well with AI-RADS categories 1 to 2, which indicate outputs suitable for integration into clinical workflows. Outputs rated as "incorrect" were predominantly assigned to categories 4 to 5, warranting override or removal from display.

Conclusion: AI-RADS provides a structured framework for the case-level evaluation of AI output reliability, clinical utility, and consequences for report communication. This multireader study demonstrated substantial interreader agreement and applicability across various AI applications.

Objectives: To evaluate the contrast enhancement of the breast parenchyma in MRI using gadopiclenol (GP) at half the standard dose of gadolinium compared with gadobutrol (GB) at the standard dose of gadolinium.

Materials and methods: This retrospective, single-center study considered 319 consecutive female patients who underwent breast MRI with a half-standard dose of gadopiclenol (0.05 mmol/kg) between January and March of 2025. Complete data sets of patients who had received a prior cycle-appropriate MRI using a standard dose of gadobutrol (0.1 mmol/kg) within 2 years, without interim therapy, were evaluated. Absolute signal enhancement of the aorta, axillary lymph nodes, and breast parenchyma was measured to provide an objective assessment. Two independent radiologists evaluated subjective image quality and diagnostic confidence using a 5-point Likert Scale in a side-by-side comparison. Statistical analyses were performed using Wilcoxon signed-rank tests and weighted Cohen kappa.

Results: A total of 145 patients (mostly with hereditary breast and ovarian cancer syndrome) met the inclusion criteria. GP revealed significantly higher absolute enhancement values for all regions (eg, breast parenchyma: GP: 41.8, IQR: 27.1 to 65.4 vs GB: 33.4, IQR: 21.6 to 56.3; P < 0.001). Both contrast agents achieved excellent overall diagnostic confidence ratings (5; IQR: 4 to 5) and demonstrated moderate agreement (GP: κ = 0.59/ GB: κ = 0.48). GP received more excellent ratings (71.9% vs 56.8%; P = 0.002). Most pairs were considered equal on terms of image quality (R1: 90/145; R2: 92/145; P = 0.845) with fair agreement (κ = 0.366). Agreement was achieved in 97 of the 145 cases (67 "equal," 21 "pro GP", and 9 "pro GB"). The combination "equal & pro GP" accounted for most of the remaining comparisons (37/48).

Conclusion: Breast MRI using gadopiclenol with half standard gadolinium dose offers equivalent to superior objective contrast enhancement and subjective diagnostic confidence compared with a gadobutrol-enhanced MRI with full standard gadolinium dose.

Since the discovery of gadolinium (Gd) in the brain following the administration of gadolinium-based contrast agents (GBCAs), considerable progress has been made in understanding their pharmacokinetics and neurotoxicology. This review summarizes animal studies assessing the presence of Gd after GBCA administration, with a specific focus on functional and behavioral outcomes, rather than providing a comprehensive overview of all aspects of Gd presence in the brain. These findings indicate that Gd accumulation in the brain depends on the chemical structure of GBCAs, with linear agents exhibiting greater retention and slower clearance than macrocyclic agents. Gd distribution is nonhomogeneous, primarily localized in deep gray matter structures, and is influenced by cerebrospinal fluid-mediated transport and perivascular deposition. Although motor and cognitive functions are generally unaffected under normal conditions, prolonged exposure to linear GBCAs or preexisting conditions such as inflammation or metabolic disorders may increase neurotoxic risks, resulting in motor and cognitive deficits. Pain and sensory hypersensitivity are frequently and reproducibly observed, particularly when linear agents are used. We will also discuss the potential mechanisms of neurotoxicity caused by free Gd3+ ion. However, these mechanistic findings are limited because the studies cannot be extrapolated to clinical practice. Future studies should investigate the potential associations between GBCA exposure and neurodegenerative diseases. These insights are essential for enhancing GBCA safety and informing clinical guidelines.

Objectives: Photon-counting CT (PCCT) represents a newer CT technology with reduced electronic noise and potentially better dose efficiency than conventional CT. However, it remains unclear how vertical off-center positioning affects dose and image quality across a spectrum of patient sizes. The aim was to quantify the effects of vertical off-centering on radiation dose and image noise in PCCT using anthropomorphic phantoms representing both adult and pediatric body sizes.

Materials and methods: Three anthropomorphic phantoms (adult male, 10-year-old, and 5-year-old) were scanned on a commercially available PCCT system at multiple vertical offsets using a posteroanterior localizer with the x-ray tube positioned below the phantom. Chest and abdomen protocols were used, with radiation doses, Monte Carlo-simulated organ doses, and image noise recorded at each offset.

Results: Off-centering markedly affected radiation dose, whereas image noise differed primarily between the predefined image quality levels. A strong linear relationship was observed between vertical offset and CTDIvol [median R2 (IQL) = 0.85 (0.78-0.98)]. Downward off-centering (-4 cm) increased radiation dose by up to 16% in adults and 17% in pediatric phantoms for both chest and abdominal scans, with the largest effects in chest scans without tin filtration. Upward off-centering (+4 cm) reduced dose by up to 11% in adults and 8% in pediatric phantoms. Larger phantoms showed steeper regression slopes, indicating stronger dose dependence on positioning. In contrast, no consistent dependence of image noise on vertical off-centering was observed within a given image quality level [median R2 = 0.23 (0.03-0.52)]. Across all offsets, the overall variation reached +72%/-47% in chest PCCT and +66%/-13% in abdominal PCCT.

Conclusions: Vertical off-center positioning substantially affects radiation dose in PCCT, whereas image noise appears largely independent of vertical positioning within a given image quality level. Meticulous isocenter alignment remains crucial for both adult and pediatric imaging to avoid unnecessary radiation and sustain diagnostic image quality.