Quantitative Imaging in Medicine and Surgery最新文献

Background: Diabetic retinopathy is a leading cause of vision impairment, often progressing to neovascular glaucoma. Early detection of neovascularisation of the iris (NVI) is crucial for timely intervention. Traditional diagnostic methods, such as slit-lamp examination, have limitations in identifying early-stage NVI. This study presents a deep learning-based automated approach for analysing iris fluorescein angiography (IFA) images to detect and quantify peripupillary leakage, a key indicator of NVI.

Methods: A dataset of 2,449 IFA images was used to train a YOLOv8n-based segmentation model for precise pupil localisation. A leakage circularity detection algorithm was developed to quantify peripupillary fluorescein leakage. The algorithm's performance was evaluated using an independent test set of 131 clinically standardized IFA images. Performance metrics included mean absolute error (MAE), mean absolute percentage error (MAPE), and intersection over union (IoU). Results were compared with manual annotations from two clinical experts.

Results: The proposed method demonstrated a significant reduction in MAE (20.81 degrees) and MAPE (21.64%) compared to Clinical Staff 1 (MAE: 34.23 degrees, MAPE: 58.38%) and Clinical Staff 2 (MAE: 43.17 degrees, MAPE: 75.71%). The algorithm achieved an IoU of 39.3%, slightly lower than Clinical Staff 1 (44.5%) and Clinical Staff 2 (41.7%), indicating high segmentation accuracy but minor spatial misalignment. The inter-clinician agreement yielded an IoU of 54.8%, highlighting subjectivity in human assessments.

Conclusions: The deep learning-based approach provides superior consistency and accuracy in quantifying peripupillary fluorescein leakage compared to manual expert annotations. While human experts demonstrated slightly higher spatial precision, the algorithm significantly reduces variability and subjectivity in leakage quantification. This automated method has the potential to enhance early detection of NVI, improve clinical workflow efficiency, and assist ophthalmologists in diagnosing DR. Further optimization will focus on refining spatial segmentation accuracy.

Background: Contrast-enhanced ultrasound (CEUS) and Ovarian-Adnexal Reporting and Data System ultrasound classification (O-RADS US) have been applied in the diagnosis and risk stratification of ovarian and adnexal masses. This study aimed to evaluate the diagnostic value and risk stratification efficacy of CEUS and O-RADS US for ovarian and adnexal masses.

Methods: A systematic review and meta-analysis of studies in the PubMed, Embase, Web of Science, and the Cochrane Library databased published until October 2024 was conducted. The Quality Assessment of Diagnostic Accuracy Studies-2 (QUADAS-2) was used for quality assessment. The Deeks funnel plot asymmetry test was used for the publication bias. The summary sensitivity, specificity, diagnostic odds ratio (DOR), and summary receiver operating characteristic (SROC) curve were used for the evaluation of diagnostic performance. The bivariate mixed-effects model in STATA 17.0 software (StataCorp) was used for the meta-analysis.

Results: This meta-analysis included a total of 21 studies, comprising 5,433 patients. CEUS was evaluated in seven studies, while O-RADS US was assessed in 15 studies, with one study using both methods for evaluation. The quality assessment revealed that bias risk and concerns regarding applicability were generally related to patient selection. The pooled sensitivity and specificity of CEUS were 93% [95% confidence interval (CI): 87-96%] and 91% (95% CI: 82-95%), respectively. For O-RADS US, the pooled sensitivity and specificity were 94% (95% CI: 87-98%) and 81% (95% CI: 72-88%), respectively.

Conclusions: CEUS and O-RADS US both exhibit high sensitivity in differentiating ovarian or adnexal masses, with CEUS also demonstrating very high specificity.

Background: Left atrial appendage (LAA) closure is a key stroke prevention strategy for non-valvular atrial fibrillation (AF). Cardiac computed tomography angiography (CTA) has emerged as an effective non-invasive alternative to transesophageal echocardiography for follow-up assessment. However, image quality and diagnostic confidence are often compromised by device-related artifacts and cardiac motion, highlighting the need for advanced motion correction techniques such as Snapshot Freeze 2 (SSF2), a second-generation whole-heart algorithm. The effects of SSF2 on patients with LAA closure have not yet been examined. Thus, this study sought to examine the ability of SSF2 to improve cardiac CTA image quality and diagnostic confidence for AF after LAA closure.

Methods: This retrospective study included 42 LAA closure patients who underwent single-heartbeat cardiac CTA on a 16-cm detector computed tomography scanner. The images were reconstructed using the standard reconstruction method (SRM), the first-generation Snapshot Freeze 1 (SSF1), and SSF2. Two experienced radiologists assessed the image artifacts caused by the LAA (at the proximal, central, and distal regions), boundaries, visualization clarity, and diagnostic confidence for LAA closures using a five-point scale. Regions of interest (ROIs) were placed in three areas of LAA closures with heavy artifacts to measure the standard deviation (SD). The Friedman test with a post-hoc analysis was used to compare the quantitative and qualitative results. All the continuous variables are presented as mean ± SD or median (interquartile range), while the subjective image scores are expressed as median (interquartile range) unless otherwise specified.

Results: In relation to the quantitative assessment, SSF2 had the smallest SD values, while the SRM had the largest SD values. SSF2 had significantly lower (better) artifact scores for the proximal, central, and distal regions of the LAA [1 (0-1), 1 (0-1), 1 (0-1), respectively] than the SRM [2 (1-2), 2 (1-3), 1 (0-3)] and SSF1 [1 (1-2), 2 (1-3), 1 (0-2)], as well as better scores for the LAA closure boundaries [4 (4-5)] and LAA visualization [4 (4-5)] than the SRM [3 (2-4) and 3 (3-4)] and SSF1 [4 (3-4) and 4 (3-4)], respectively. Excellent diagnostic confidence (i.e., scores of four or greater) was achieved in 30.9%, 52.4%, and 90.5% of the cases using the SRM, SSF1, and SSF2, respectively.

Conclusions: SSF2 significantly reduces the artifacts caused by LAA closure, and improves image quality, visualization, and diagnostic confidence in cardiac CTA of patients with LAA closure, compared with the SRM and SSF1. This could improve both the detection of device-related complications and patient management.

Background: The detection of colorectal liver metastases (CRLMs) in complex hepatic backgrounds due to chemotherapy-associated liver injury (CALI) and prior local CRLM treatment is challenging and requires advanced imaging modalities capable of providing both precise lesion detection and CALI assessments. This study aimed to evaluate the diagnostic performance of 5 Tesla (T) multimodal magnetic resonance imaging (MRI) for detecting CRLMs while accurately assessing the hepatic background.

Methods: A total of 35 post-chemotherapy patients with suspected CRLMs and a combined total of 118 MRI-identified lesions were prospectively enrolled. Participants underwent 5T liver multimodal MRI with acquisition of: T1-weighted (T1W) in/out-of-phase imaging and proton density fat fraction (PDFF), susceptibility-weighted imaging with fast technique (uSWIFT), and standard hepatic gadoxetic acid-enhanced MRI (EOB-MRI) with three-dimensional (3D) isotropic T1W fast spoiled gradient-recalled echo (FSPGR) hepatobiliary phase imaging and artificial intelligence (AI)-assisted compressed sensing (ACS-HBP) (acquisition voxel size 1.2 mm³, reconstructed voxel size 0.6 mm³, 300 slices). Image analysis was performed by two readers, blinded and randomized, who assessed CALI and CRLM diagnostic performance. Statistical analyses included inter-rater agreement (Cohen's kappa), diagnostic metrics-sensitivity, positive predictive value (PPV), and area under the receiver operating characteristic (ROC) curve (AUC)-and generalized estimating equations (GEEs) for identifying factors associated with CRLM detection and for comparing diagnostic performance.

Results: The combination of ACS-HBP and diffusion-weighted imaging (DWI) demonstrated excellent diagnostic performance in detecting CRLMs, yielding the highest sensitivity (97.2%) and PPV (94.6-95.5%), whereas for small lesions (≤10 mm), the combination yielded a sensitivity of 95.7%, outperforming DWI alone (66.0%). Neither CALI nor a prior history of local CRLM treatment had a significant impact on diagnostic performance (all P>0.05). On ACS-HBP imaging, 86.1% (93/108) of all lesions and 71.7% (33/46) of lesions ≤10 mm presented with a target sign (central hyperintensity with a hypointense rim), or reverse target sign (central hypointensity with a hyperintense rim).

Conclusions: The combination of ACS and 5T EOB-MRI demonstrates excellent diagnostic performance for CRLMs, including for ≤10 mm lesions, with high sensitivity and PPV across diverse hepatic backgrounds. On HBP imaging, CRLMs characteristically display target and reverse-target signs, especially in lesions ≤10 mm, facilitating differentiation from hepatic cysts.

Background: Although posterior lumbar interbody fusion (PLIF) is effective in treating degenerative lumbar spine diseases, postoperative adjacent segment disease (ASD) significantly affects long-term prognosis and quality of life. The complex pathogenesis may involve systemic inflammation, lumbar muscle degeneration, biomechanical changes, and surgical factors, but comprehensive studies are lacking. The aim of this study was to analyze the risk factors for ASD following PLIF and to provide guidance for clinical prevention and management.

Methods: A retrospective analysis was conducted on 500 patients who underwent PLIF from January 2017 to August 2022. Patients were divided into a non-ASD group (n=404) and an ASD group (n=96) based on postoperative imaging. Functional outcomes were assessed using the visual analog scale and the Oswestry Disability Index (ODI). Univariate and multivariate logistic regression analyses examined factors such as neutrophil-to-lymphocyte ratio (NLR), psoas major muscle index (PMI), multifidus muscle index (MMI), multifidus muscle fat infiltration rate (MFI), cage subsidence, pelvic incidence-lumbar lordosis mismatch, and nonunion to identify independent risk factors for ASD.

Results: The incidence of ASD was 19.2%. Postoperative visual analog scale and ODI scores improved significantly in both groups, but the non-ASD group experienced a greater degree of pain relief and functional recovery. Elevated NLR, decreased PMI and MMI, increased MFI, cage subsidence, lower Hounsfield units (HU) values, and nonunion were identified as independent risk factors for ASD.

Conclusions: ASD after PLIF is closely associated with systemic inflammation, lumbar muscle degeneration, cage subsidence, and nonunion. Interventions targeting these risk factors-such as optimizing inflammatory status preoperatively, enhancing lumbar muscle rehabilitation, improving surgical techniques to prevent cage subsidence, and promoting bone fusion-may reduce the incidence of ASD and improve long-term outcomes.

Background: Quantitative dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) can potentially be used to assess synovial inflammation in patients with knee osteoarthritis (OA). Precise methods are needed for adequate identification of changes in clinical trials. In this study, we aimed to evaluate test-retest repeatability of DCE-MRI within the synovium of patients with knee OA, and develop a semi-automatic method for synovial subregion assessment.

Methods: In this secondary explorative analysis, we included 31 adults with radiographically confirmed mild to moderate knee OA receiving sham treatment as participants of a randomized controlled trial evaluating genicular artery embolization. DCE-MRI of the knee was performed at baseline and at 1-month follow-up. The synovium was semi-automatically segmented and subdivided in 8 subregions based on blood vessel mapping. Quantitative DCE-MRI parameters were extracted from the synovium and its subregions by voxel-wise pharmacokinetic modeling. A two-way random effects model was used to estimate within-subject variance and between-subject variance, and intraclass correlation (ICC), within-subject standard deviation or coefficient of variation, and the repeatability coefficient (RC) were calculated.

Results: DCE-MRI parameter K^trans showed good repeatability with an ICC of 0.84 and a RC of 0.039. Semi-quantitative parameter IAUC₆₀ showed a similar ICC of 0.85 and a slightly higher RC of 0.090. For subregional assessment, ICCs for K^trans ranged between 0.70 and 0.89 while RCs ranged between 0.028 and 0.099.

Conclusions: Quantitative DCE-MRI biomarkers have good test-retest repeatability on both the whole synovium and synovial subregions, with K^trans showing the best performance.

Background: Risk stratification is particularly important for the prognosis of coronavirus disease 2019 (COVID-19), particularly regarding myocardial injury (MI). Non-contrast chest computed tomography (CT) is useful for diagnosing lung injury, but its potential for one-stop evaluation of MI remains unclear. This study aimed to develop a deep learning (DL) framework for the one-step prediction of MI in COVID-19 patients using non-contrast chest CT. This approach seeks to enable early MI screening, direct high-risk patients toward further diagnostic tests, and optimize the allocation of medical resources.

Methods: A group of 453 patients with COVID-19, including 230 patients with MI and 223 patients without MI, were retrospectively recruited. A two-stage DL framework was developed to first segment the left ventricle (LV) in the non-contrast chest CT images using fully convolutional networks with a ResNet-101 backbone (FCN-ResNet-101) module and then classify the status of MI using densely connected convolutional network with the structure type of 121 (DenseNet-121) module. The framework was trained in a training-validation dataset of 413 patients (MI or non-MI) with a cross-validation approach, and evaluated in a testing dataset of 40 patients.

Results: The proposed DL framework accurately obtained the segmentations of LV in non-contrast chest CT images with an intersection over union (IoU) of 0.8041, an accuracy (ACC) of 0.9949, and a Dice coefficient of 0.8672. Based on the segmentation, the DL framework further accurately determined MI status and obtained an area under the curve (AUC) of 0.8618 [95% confidence interval (CI): 0.8049-0.9187], an ACC of 0.7763, a sensitivity (SEN) of 0.8750, a specificity (SPE) of 0.6071, and an F1 score of 0.8317.

Conclusions: DL could determine the status of MI in non-contrast chest CT images of patients with COVID-19, providing one-stop convenience for early screening of MI.

Background: Amide proton transfer-weighted (APTw) imaging provides molecular-level information for comprehensive tumor characterization, aiding in diagnosis, grading, and treatment response monitoring. The study aimed to investigate the potential of APT imaging in simultaneously predicting three key markers of hepatocellular carcinoma (HCC) aggressiveness-microvascular invasion (MVI), Ki-67 labeling index (LI) and histologic grade-and compare its performance to that of diffusion-weighted imaging (DWI).

Methods: Fifty-six patients with histologically confirmed HCC underwent abdominal magnetic resonance imaging (MRI) including APTw and DWI. Tumor mean APTw and apparent diffusion coefficient (ADC) values were measured and correlated with Ki-67 LI. The differences of APTw and ADC values between MVI positive (MVI+) and negative (MVI-), high and low Ki-67 LI, and high- and low-grade HCC were analyzed. The diagnostic efficacy of APTw and ADC for predicting HCC aggressiveness was assessed using receiver operating characteristic (ROC) curve analysis.

Results: APTw positively correlated with Ki-67 LI (ρ=0.49, P<0.001) and ADC negatively correlated with Ki-67 LI (ρ=-0.27, P=0.04). HCC with MVI+ had higher APTw values than MVI- (1.68%±1.02% vs. 0.21%±1.29%; P<0.001). High Ki-67 LI had higher APTw median values than low Ki-67 LI (1.57% vs. 0.16%; P<0.001). High-grade HCC had higher APTw values than low-grade (1.54%±1.10% vs. 0.02%±1.25%; P<0.001). Diagnostic performance of APTw for predicting MVI+, high Ki-67 LI, and high-grade HCC, was promising with area under the curve values of 0.82 (ADC: 0.65), 0.85 (ADC: 0.76), and 0.82 (ADC: not applicable), respectively.

Conclusions: APTw imaging could potentially provide additional value in assessing HCC aggressiveness, complementing the information obtained from DWI.

Background: Fetal cardiac tumors are relatively rare among congenital heart diseases (CHDs). The presence of cardiac rhabdomyoma (CR) during fetal development may serve as the earliest clinical sign of tuberous sclerosis complex (TSC), with multiple CRs being a strong predictor of TSC. This study aimed to summarize the ultrasound image characteristics of fetal CR and follow up the comprehensive information of prenatal magnetic resonance imaging (MRI) examination, genetics, and postnatal changes in clinical outcomes to improve the detection rate and optimize the outcome of the diagnosis and treatment.

Methods: A retrospective analysis was performed on prenatal ultrasound diagnostic data from 36 fetal CRs, including lesion characteristics (number, location, size, and associated complications), as well as follow-up information during pregnancy and within 1 year of birth.

Results: Among the 36 cases of CR, 11 patients (31%) underwent MRI examinations. Of these, five cases presented with subependymal nodules on cranial MRI, suggestive of TSC; one case showed asymmetric bilateral lateral ventricles and dilation of the vein of Galen; and one case revealed multiple hemorrhages in the cerebral parenchyma and subependymal regions, along with left lateral ventricular enlargement. The remaining four MRI results were negative. Genetic testing identified abnormalities in 6 cases (17%), including five with TSC-related gene mutations and one with a whole exome sequencing (WES) anomaly. Pregnancy was terminated in 12 cases (33%). Postnatal follow-up demonstrated no significant change in CR size in one case, a reduction in CR size in one case, and complete regression of CR in three cases.

Conclusions: Echocardiography plays a critical role in diagnosing CR. For fetuses prenatally diagnosed with CR, routine cranial MRI and whole exome genetic testing should be performed to confirm the presence of TSC or other pathogenic variants, which is essential for early clinical intervention and decision-making.

Background: In gadoxetic acid-enhanced liver magnetic resonance imaging (MRI), an inadequate breath-holding strategy is an independent risk factor for image quality degradation in single-phase arterial acquisition. Therefore, optimizing breath-holding training protocols is critical. This study aimed to assess the effectiveness of a nongradual breath training protocol combined with contrast dilution for reducing arterial phase (AP) respiratory artifacts in upper abdomen gadoxetic acid-enhanced MRI.

Methods: In this retrospective study, we enrolled 126 patients (91 males; mean age 52.65±11.07 years) who underwent gadoxetic acid-enhanced upper abdomen MRI for the first and only time from May 2019 to February 2020. All patients were divided into four groups according to different breath-holding training strategies and contrast injection patterns (group 1: gradual strategy + diluted injection; group 2: nongradual strategy + diluted injection; group 3: gradual strategy + undiluted injection; group 4: nongradual strategy + undiluted injection). Two radiologists evaluated AP images in a blinded manner using a 5-point scale. Semiquantitative scores, signal-to-noise ratio (SNR), contrast-to-noise ratio (CNR) of the hepatic artery and portal vein to the liver parenchyma, and CNR of the hepatic artery to the portal vein were compared between the four groups via Mann-Whitney and Kruskal-Wallis tests. The differences in transient severe motion (TSM) incidence rates and respiratory artifact scores between the gradual and nongradual groups were calculated via the adjusted the Chi-squared test and the Mann-Whitney, respectively.

Results: The results of the 5-point scale evaluation showed that respiratory artifacts significantly differed between groups (group 1: 2.56±1.16; group 2: 1.53±0.62; group 3: 2.12±1.08; group 4: 1.79±0.89; P=0.002): groups 2 and 4 had smaller artifacts, and group 2 had the fewest number of artifacts. However, no statistically significant differences were observed in the scores for the hepatic artery (P=0.177), portal vein (P=0.214), abdominal aorta (P=0.599), and liver parenchyma (P=0.243) or for the total score (P=0.235). The groups significantly differed terms of portal vein SNR (P<0.001), liver parenchyma SNR (P=0.035), and the CNR of the portal vein to the liver parenchyma (P=0.008). These parameters were optimal in group 2. The incidence of TSM and the respiratory artifact score were significantly higher in the gradual group (groups 1 and 3) compared to the nongradual group (groups 2 and 4).

Conclusions: The nongradual breath-holding training strategy combined with an injection rate of 2 mL/s with 50% diluted gadoxetic acid could significantly reduce respiratory artifacts and improve AP image quality.