Ultrasonography最新文献

Purpose: This study aimed to evaluate the diagnostic accuracy of three multimodal large language models (LLMs) in radiological image interpretation and to assess the impact of prompt engineering strategies and input conditions.

Methods: This study analyzed 67 radiological quiz cases from the Korean Society of Ultrasound in Medicine. Three multimodal LLMs (Claude 3.5 Sonnet, GPT-4o, and Gemini-1.5-Pro-002) were evaluated using six types of prompts (basic [without system prompt], original [specific instructions], chain-of-thought, reflection, multiagent, and artificial intelligence [AI]-generated). Performance was assessed across various factors, including tumor versus non-tumor status, case rarity, difficulty, and knowledge cutoff dates. A subgroup analysis compared diagnostic accuracy between imaging-only inputs and combined imaging-descriptive text inputs.

Results: With imaging-only inputs, Claude 3.5 Sonnet achieved the highest overall accuracy (46.3%, 186/402), followed by GPT-4o (43.5%, 175/402) and Gemini-1.5-Pro-002 (39.8%, 160/402). AI-generated prompts yielded superior combined accuracy across all three models, with improvements over the basic (5.5%, P=0.035), chain-of-thought (4.0%, P=0.169), and multiagent prompts (3.5%, P=0.248). The integration of descriptive text significantly enhanced diagnostic accuracy for Claude 3.5 Sonnet (46.3% to 66.2%, P<0.001), GPT-4o (43.5% to 57.5%, P<0.001), and Gemini-1.5-Pro-002 (39.8% to 60.4%, P<0.001). Model performance was significantly influenced by case rarity for GPT-4o (rare: 6.7% vs. non-rare: 53.9%, P=0.001) and by knowledge cutoff dates for Claude 3.5 Sonnet (post-cutoff: 23.5% vs. pre-cutoff: 64.0%, P=0.005).

Conclusion: Claude 3.5 Sonnet achieved the highest diagnostic accuracy in radiological quiz cases, followed by GPT-4o and Gemini-1.5-Pro-002. The use of AI-generated prompts and the integration of descriptive text inputs enhanced model performance.

Purpose: This study aimed to evaluate the prognostic value of thrombus stiffness, measured by strain elastography, in independently predicting the success of endovascular treatment for thrombosed hemodialysis fistulas.

Methods: The study employed a retrospective observational design. Forty-nine patients with chronic kidney disease undergoing hemodialysis and experiencing fistula dysfunction were included. Various factors were evaluated, including patient age, sex, comorbidities, arteriovenous fistula (AVF) type, use of antiplatelet agents, duration of thrombosed AVF dysfunction, thrombus length, thrombus stiffness, and thrombus strain ratio (SR). Thrombus stiffness was measured using strain ultrasound elastography.

Results: The presence of comorbidities (P=0.018), duration of AVF dysfunction due to thrombosis (P=0.005), thrombus stiffness (P<0.001), and thrombus SR (P<0.001) were statistically significant predictors of treatment success, while other factors were not. Thrombus stiffness and SR, assessed by two independent radiologists with excellent inter-reader agreement, demonstrated high reliability. The optimal SR cut-off for treatment success was 1.4 (sensitivity, 0.947; specificity, 0.901; area under the curve, 0.935). Multivariable logistic regression analysis revealed that both thrombus SR and thrombus stiffness significantly influenced treatment outcomes (P=0.003 and P=0.007, respectively). A 0.1-unit increase in thrombus SR was associated with 9.37% reduction in the likelihood of treatment success, while a smilar increase in thrombus stiffness exhibited an 8.06% reduction, underscoring their importance as prognostic factors in clinical setting.

Conclusion: This study demonstrates the utility of strain ultrasound elastography in assessing thrombus stiffness in thrombosed hemodialysis AVFs and its correlation with treatment success.

Purpose: The aim of this study was to investigate the feasibility of an intelligent handheld ultrasound (US) device for assisting non-expert general practitioners (GPs) in detecting carotid plaques (CPs) in community populations.

Methods: This prospective parallel controlled trial recruited 111 consecutive community residents. All of them underwent examinations by non-expert GPs and specialist doctors using handheld US devices (setting A, setting B, and setting C). The results of setting C with specialist doctors were considered the gold standard. Carotid intima-media thickness (CIMT) and the features of CPs were measured and recorded. The diagnostic performance of GPs in distinguishing CPs was evaluated using a receiver operating characteristic curve. Inter-observer agreement was compared using the intragroup correlation coefficient (ICC). Questionnaires were completed to evaluate clinical benefits.

Results: Among the 111 community residents, 80, 96, and 112 CPs were detected in settings A, B, and C, respectively. Setting B exhibited better diagnostic performance than setting A for detecting CPs (area under the curve, 0.856 vs. 0.749; P<0.01). Setting B had better consistency with setting C than setting A in CIMT measurement and the assessment of CPs (ICC, 0.731 to 0.923). Moreover, measurements in setting B required less time than the other two settings (44.59 seconds vs. 108.87 seconds vs. 126.13 seconds, both P<0.01).

Conclusion: Using an intelligent handheld US device, GPs can perform CP screening and achieve a diagnostic capability comparable to that of specialist doctors.

Purpose: This study aimed to evaluate our institution's experience in using artificial intelligence (AI) decision support (DS) as part of the clinical workflow to triage patients with Breast Imaging Reporting and Data System (BI-RADS) 3 sonographic lesions whose follow-up was delayed during the coronavirus disease 2019 (COVID-19) pandemic, against subsequent imaging and/or pathologic follow-up results.

Methods: This retrospective study included patients with a BI-RADS category 3 (i.e., probably benign) breast ultrasound assessment from August 2019-December 2019 whose follow-up was delayed during the COVID-19 pandemic and whose breast ultrasounds were re-reviewed using Koios DS Breast AI as part of the clinical workflow for triaging these patients. The output of Koios DS was compared with the true outcome of a presence or absence of breast cancer defined by resolution/stability on imaging follow-up for at least 2 years or pathology results.

Results: The study included 161 women (mean age, 52 years) with 221 BI-RADS category 3 sonographic lesions. Of the 221 lesions, there were two confirmed cancers (0.9% malignancy rate). Koios DS assessed 112/221 lesions (50.7%) as benign, 42/221 lesions (19.0%) as probably benign, 64/221 lesions (29.0%) as suspicious, and 3/221 lesions (1.4%) as probably malignant. Koios DS had a sensitivity of 100% (2/2; 95% confidence interval [CI], 16% to 100%), specificity of 70% (154/219; 95% CI, 64% to 76%), negative predictive value of 100% (154/154; 95% CI, 98% to 100%), and false-positive rate of 30% (65/219; 95% CI, 24% to 36%).

Conclusion: When many follow-up appointments are delayed, e.g., natural disaster, or scenarios where resources are limited, breast ultrasound AI DS can help triage patients with probably benign breast ultrasounds.

Radiogenomics, an extension of radiomics, explores the relationship between imaging features and underlying gene expression patterns. This field is instrumental in providing reliable imaging surrogates, thus potentially representing an alternative to genetic testing. The rapidly growing area of radiogenomics that utilizes ultrasound (US) imaging seeks to elucidate the connections between US image characteristics and genomic data. In this review, the authors outline the radiogenomics workflow and summarize the applications of US-based radiogenomics. These include the prediction of gene variations, molecular subtypes, and other biological characteristics, as well as the exploration of the relationships between US phenotypes and cancer gene profiles. Although the field faces various challenges, US-based radiogenomics offers promising prospects and avenues for future research.

Purpose: This study evaluated the educational impact of an artificial intelligence (AI)-based decision support system for breast ultrasonography (US) on medical professionals not specialized in breast imaging.

Methods: In this multi-case, multi-reader study, educational materials, including American College of Radiology Breast Imaging Reporting and Data System (BI-RADS) descriptors, were provided alongside corresponding AI results during training. The AI system presented results in the form of AIheatmaps, AI scores, and AI-provided BI-RADS assessment categories. Forty-two readers evaluated the test set in three sessions: the first session (S1) occurred before the educational intervention, the second session (S2) followed education without AI assistance, and the third session (S3) took place after education with AI assistance. The area under the receiver operating characteristic curve (AUC), sensitivity, specificity, and overall performance, were compared between the sessions.

Results: The mean sensitivity increased from 66.5% (95% confidence interval [CI], 59.2% to 73.7%) to 88.7% (95% CI, 84.1% to 93.3%), with a statistically significant difference (P<0.001), and the AUC non-significantly increased from 0.664 (95% CI, 0.606 to 0.723) to 0.684 (95% CI, 0.620 to 0.748) (P=0.300). Both measures were higher in S2 than in S1. The AI-achieved AUC was comparable to that of the expert reader (0.747 [95% CI, 0.640 to 0.855] vs. 0.803 [95% CI, 0.706 to 0.900], P=0.217). Additionally, with AI assistance, the mean AUC for inexperienced readers was not significantly different from that of the expert reader (0.745 [95% CI, 0.660 to 0.830] vs. 0.803 [95% CI, 0.706 to 0.900], P=0.120).

Conclusion: The mean AUC and sensitivity improved after incorporating AI into breast US education and interpretation. AI systems with high-level performance for breast US can potentially be used as educational tools in the interpretation of breast US images.

Purpose: This study assessed the reproducibility and clinical value of the vascular index (VI), derived from superb microvascular imaging (SMI) using Doppler ultrasonography, for evaluating renal function in transplanted kidneys.

Methods: This retrospective study included 63 renal transplant patients who underwent grayscale and Doppler ultrasonography with SMI from January 2022 to February 2023. The VI of the transplanted kidneys was measured using three methods (VIbox, VIF1, VIF2). The VI was compared across chronic kidney disease (CKD) groups categorized by estimated glomerular filtration rate (eGFR) and Kidney Disease: Improving Global Outcomes (KDIGO) CKD risk groups based on eGFR and albuminuria. The correlation between VI and renal function was evaluated. Univariate and multivariate linear regression analyses were used to identify predictors of eGFR.

Results: Significant differences in VI were observed among CKD groups based on eGFR (VIbox, P=0.001; VIF1, P<0.001; VIF2, P<0.001) and KDIGO CKD groups based on eGFR and albuminuria (VIbox, P=0.039; VIF1, P=0.001; VIF2, P<0.001). VIF1 and VIF2 demonstrated moderate/high correlations with eGFR (r=0.627, P<0.001 and r=0.657, P<0.001, respectively) and serum creatinine (r=-0.626, P<0.001 and r=-0.649, P<0.001, respectively). VIbox moderately correlated with eGFR (r=0.445, P<0.001). Multivariate regression identified the urine albumincreatinine ratio (ACR) (adjusted odds ratio [aOR], 1.122; 95% confidence interval [CI], -0.007 to, 0.000; P=0.030) and VIF2 (aOR, 1.114; 95% CI, 0.466 to 1.235; P<0.001) were independently associated with eGFR.

Conclusion: The VI measured by drawing a region of interest along the border of the transplanted kidney in SMI (VIF2) is highly reproducible and correlates well with eGFR. Both VIF2 and ACR independently predict eGFR.

Purpose: This study assessed the performance of the ultrasound-guided attenuation parameter (UGAP) in diagnosing and grading hepatic steatosis in patients with metabolic dysfunctionassociated steatotic liver disease (MASLD). Magnetic resonance imaging proton density fat fraction (MRI-PDFF) served as the reference standard.

Methods: Patients with hepatic steatosis were enrolled in this prospective study and underwent UGAP measurements. MRI-PDFF values of ≥5%, ≥15%, and ≥25% were used as references for the diagnosis of steatosis grades ≥S1, ≥S2, and S3, respectively. Spearman correlation coefficients and area under the receiver operating characteristic curves (AUCs) were calculated.

Results: Between July 2023 and June 2024, the study included 88 patients (median age, 40 years; interquartile range [IQR], 36 to 46 years), of whom 54.5% (48/88) were men and 45.5% (40/88) were women. Steatosis grades exhibited the following distribution: 22.7% (20/88) had S0, 50.0% (44/88) had S1, 21.6% (19/88) had S2, and 5.7% (5/88) had S3. The success rate for UGAP measurements was 100%. The median UGAP value was 0.74 dB/cm/MHz (IQR, 0.65 to 0.82 dB/ cm/MHz), and UGAP values were positively correlated with MRI-PDFF (r=0.77, P<0.001). The AUCs of UGAP for the diagnoses of ≥S1, ≥S2, and S3 steatosis were 0.91, 0.90, and 0.88, respectively. In the subgroup analysis, 98.4% (60/61) of patients had valid controlled attenuation parameter (CAP) values. UGAP measurements were positively correlated with CAP values (r=0.65, P<0.001).

Conclusion: Using MRI-PDFF as the reference standard, UGAP demonstrates good diagnostic performance in the detection and grading of hepatic steatosis in patients with MASLD.