Constructing a Large Language Model to Generate Impressions from Findings in Radiology Reports.

Background The specialization and complexity of radiology makes the automatic generation of radiologic impressions (ie, a diagnosis with differential diagnosis and management recommendations) challenging. Purpose To develop a large language model (LLM) that generates impressions based on imaging findings and to evaluate its performance in professional and linguistic dimensions. Materials and Methods Six radiologists recorded imaging examination findings from August 2 to 31, 2023, at Shanghai General Hospital and used the developed LLM before routinely writing report impressions for multiple radiologic modalities (CT, MRI, radiography, mammography) and anatomic sites (cranium and face, neck, chest, upper abdomen, lower abdomen, vessels, bone and joint, spine, breast), making necessary corrections and completing the radiologic impression. A subset was defined to investigate cases where the LLM-generated impressions differed from the final radiologist impressions by excluding identical and highly similar cases. An expert panel scored the LLM-generated impressions on a five-point Likert scale (5 = strongly agree) based on scientific terminology, coherence, specific diagnosis, differential diagnosis, management recommendations, correctness, comprehensiveness, harmlessness, and lack of bias. Results In this retrospective study, an LLM was pretrained using 20 GB of medical and general-purpose text data. The fine-tuning data set comprised 1.5 GB of data, including 800 radiology reports with paired instructions (describing the output task in natural language) and outputs. Test set 2 included data from 3988 patients (median age, 56 years [IQR, 40-68 years]; 2159 male). The median recall, precision, and F1 score of LLM-generated impressions were 0.775 (IQR, 0.56-1), 0.84 (IQR, 0.611-1), and 0.772 (IQR, 0.578-0.957), respectively, using the final impressions as the reference standard. In a subset of 1014 patients (median age, 57 years [IQR, 42-69 years]; 528 male), the overall median expert panel score for LLM-generated impressions was 5 (IQR, 5-5), ranging from 4 (IQR, 3-5) to 5 (IQR, 5-5). Conclusion The developed LLM generated radiologic impressions that were professionally and linguistically appropriate for a full spectrum of radiology examinations. © RSNA, 2024 Supplemental material is available for this article.

Knowledge-Augmented Deep Learning for Segmenting and Detecting Cerebral Aneurysms With CT Angiography: A Multicenter Study.

Background Deep learning (DL) could improve the labor-intensive, challenging processes of diagnosing cerebral aneurysms but requires large multicenter data sets. Purpose To construct a DL model using a multicenter data set for accurate cerebral aneurysm segmentation and detection on CT angiography (CTA) images and to compare its performance with radiology reports. Materials and Methods Consecutive head or head and neck CTA images of suspected unruptured cerebral aneurysms were gathered retrospectively from eight hospitals between February 2018 and October 2021 for model development. An external test set with reference standard digital subtraction angiography (DSA) scans was obtained retrospectively from one of the eight hospitals between February 2022 and February 2023. Radiologists (reference standard) assessed aneurysm segmentation, while model performance was evaluated using the Dice similarity coefficient (DSC). The model's aneurysm detection performance was assessed by sensitivity and comparing areas under the receiver operating characteristic curves (AUCs) between the model and radiology reports in the DSA data set with use of the DeLong test. Results Images from 6060 patients (mean age, 56 years ± 12 [SD]; 3375 [55.7%] female) were included for model development (training: 4342; validation: 1086; and internal test set: 632). Another 118 patients (mean age, 59 years ± 14; 79 [66.9%] female) were included in an external test set to evaluate performance based on DSA. The model achieved a DSC of 0.87 for aneurysm segmentation performance in the internal test set. Using DSA, the model achieved 85.7% (108 of 126 aneurysms [95% CI: 78.1, 90.1]) sensitivity in detecting aneurysms on per-vessel analysis, with no evidence of a difference versus radiology reports (AUC, 0.93 [95% CI: 0.90, 0.95] vs 0.91 [95% CI: 0.87, 0.94]; P = .67). Model processing time from reconstruction to detection was 1.76 minutes ± 0.32 per scan. Conclusion The proposed DL model could accurately segment and detect cerebral aneurysms at CTA with no evidence of a significant difference in diagnostic performance compared with radiology reports. © RSNA, 2024 Supplemental material is available for this article. See also the editorial by Payabvash in this issue.

Fully automated artificial intelligence-based coronary CT angiography image processing: efficiency, diagnostic capability, and risk stratification.

OBJECTIVES: To prospectively investigate whether fully automated artificial intelligence (FAAI)-based coronary CT angiography (CCTA) image processing is non-inferior to semi-automated mode in efficiency, diagnostic ability, and risk stratification of coronary artery disease (CAD). MATERIALS AND METHODS: Adults with indications for CCTA were prospectively and consecutively enrolled at two hospitals and randomly assigned to either FAAI-based or semi-automated image processing using equipment workstations. Outcome measures were workflow efficiency, diagnostic accuracy for obstructive CAD (≥ 50% stenosis), and cardiovascular events at 2-year follow-up. The endpoints included major adverse cardiovascular events, hospitalization for unstable angina, and recurrence of cardiac symptoms. The non-inferiority margin was 3 percentage difference in diagnostic accuracy and C-index. RESULTS: In total, 1801 subjects (62.7 ± 11.1 years) were included, of whom 893 and 908 were assigned to the FAAI-based and semi-automated modes, respectively. Image processing times were 121.0 ± 18.6 and 433.5 ± 68.4 s, respectively (p <0.001). Scan-to-report release times were 6.4 ± 2.7 and 10.5 ± 3.8 h, respectively (p < 0.001). Of all subjects, 152 and 159 in the FAAI-based and semi-automated modes, respectively, subsequently underwent invasive coronary angiography. The diagnostic accuracies for obstructive CAD were 94.7% (89.9-97.7%) and 94.3% (89.5-97.4%), respectively (difference 0.4%). Of all subjects, 779 and 784 in the FAAI-based and semi-automated modes were followed for 589 ± 182 days, respectively, and the C-statistic for cardiovascular events were 0.75 (0.67 to 0.83) and 0.74 (0.66 to 0.82) (difference 1%). CONCLUSIONS: FAAI-based CCTA image processing significantly improves workflow efficiency than semi-automated mode, and is non-inferior in diagnosing obstructive CAD and risk stratification for cardiovascular events. CLINICAL RELEVANCE STATEMENT: Conventional coronary CT angiography image processing is semi-automated. This observation shows that fully automated artificial intelligence-based image processing greatly improves efficiency, and maintains high diagnostic accuracy and the effectiveness in stratifying patients for cardiovascular events. KEY POINTS: • Coronary CT angiography (CCTA) relies heavily on high-quality and fast image processing. • Full-automation CCTA image processing is clinically non-inferior to the semi-automated mode. • Full automation can facilitate the application of CCTA in early detection of coronary artery disease.

Quercetin ameliorates ulcerative colitis by activating aryl hydrocarbon receptor to improve intestinal barrier integrity.

Ulcerative colitis (UC) pathogenesis is largely associated with intestinal epithelial barrier dysfunction. A therapeutic approach to UC involves the repair of damaged intestinal barrier. Our study aimed to investigate whether aryl hydrocarbon receptor (AhR) mediated the intestinal barrier repair effects of quercetin to ameliorate UC. 3% dextran sulfate sodium was used to induce colitic mice, and quercetin (25, 50, and 100 mg/kg) was administered orally for 10 days to assess the therapeutic effects. In vitro, Caco-2 cells were used to explore the effect of quercetin on tight junction protein expression and AhR activation. The results showed that quercetin alleviated colitic mice by restoring tight junctions (TJs) integrity via an AhR-dependent manner (p < 0.05). In vitro, quercetin dose-dependently elevated the expressions of TJs protein ZO-1 and Claudin1, and activated AhR by enhancing the expression of CYP1A1 and facilitating AhR nuclear translocation in Caco-2 cells (p < 0.05). While AhR antagonist CH223191 reversed the therapeutic effects of quercetin (p < 0.05) and blocked quercetin-induced AhR activation and enhancement of TJs protein (p < 0.05). In conclusion, quercetin repaired intestinal barrier dysfunction by activating AhR-mediated enhancement of TJs to alleviate UC. Our research offered new perspectives on how quercetin enhanced intestinal barrier function.

Motion artifact removal in coronary CT angiography based on generative adversarial networks.

OBJECTIVES: Coronary motion artifacts affect the diagnostic accuracy of coronary CT angiography (CCTA), especially in the mid right coronary artery (mRCA). The purpose is to correct CCTA motion artifacts of the mRCA using a GAN (generative adversarial network). METHODS: We included 313 patients with CCTA scans, who had paired motion-affected and motion-free reference images at different R-R interval phases in the same cardiac cycle and included another 53 CCTA cases with invasive coronary angiography (ICA) comparison. Pix2pix, an image-to-image conversion GAN, was trained by the motion-affected and motion-free reference pairs to generate motion-free images from the motion-affected images. Peak signal-to-noise ratio (PSNR), structural similarity (SSIM), Dice similarity coefficient (DSC), and Hausdorff distance (HD) were calculated to evaluate the image quality of GAN-generated images. RESULTS: At the image level, the median of PSNR, SSIM, DSC, and HD of GAN-generated images were 26.1 (interquartile: 24.4-27.5), 0.860 (0.830-0.882), 0.783 (0.714-0.825), and 4.47 (3.00-4.47), respectively, significantly better than the motion-affected images (p < 0.001). At the patient level, the image quality results were similar. GAN-generated images improved the motion artifact alleviation score (4 vs. 1, p < 0.001) and overall image quality score (4 vs. 1, p < 0.001) than those of the motion-affected images. In patients with ICA comparison, GAN-generated images achieved accuracy of 81%, 85%, and 70% in identifying no, < 50%, and ≥ 50% stenosis, respectively, higher than 66%, 72%, and 68% for the motion-affected images. CONCLUSION: Generative adversarial network-generated CCTA images greatly improved the image quality and diagnostic accuracy compared to motion-affected images. KEY POINTS: • A generative adversarial network greatly reduced motion artifacts in coronary CT angiography and improved image quality. • GAN-generated images improved diagnosis accuracy of identifying no, < 50%, and ≥ 50% stenosis.

Wogonin regulates colonocyte metabolism via PPARγ to inhibit Enterobacteriaceae against dextran sulfate sodium-induced colitis in mice.

To investigate the potential effects and mechanism of wogonin on dextran sulfate sodium (DSS)-induced colitis, 70 male mice were administered wogonin (12.5, 25, 50 mg·kg-1 ·d-1 , i.g.) for 10 days, meanwhile, in order to induce colitis, the mice were free to drink 3% DSS for 6 days. We found that wogonin could obviously ameliorate DSS-induced colitis, including preventing colon shortening and inhibiting pathological damage. In addition, wogonin could increase the expression of PPARγ, which not only restores intestinal epithelial hypoxia but also inhibits iNOS protein to reduce intestinal nitrite levels. All these effects facilitated a reduction in the abundance of Enterobacteriaceae in DSS-induced colitis mice. Therefore, compared with the DSS group, the number of Enterobacteriaceae in the intestinal flora was significantly reduced after administration of wogonin or rosiglitazone by 16s rDNA technology. We also verified that wogonin could promote the expression of PPARγ mRNA and protein in Caco-2 cells, and this effect disappeared when PPARγ signal was inhibited. In conclusion, our study suggested that wogonin can activate the PPARγ signal of the Intestinal epithelium to ameliorate the Intestinal inflammation caused by Enterobacteriaceae bacteria expansion.

Deep Learning Reconstruction Shows Better Lung Nodule Detection for Ultra-Low-Dose Chest CT.

Background Ultra-low-dose (ULD) CT could facilitate the clinical implementation of large-scale lung cancer screening while minimizing the radiation dose. However, traditional image reconstruction methods are associated with image noise in low-dose acquisitions. Purpose To compare the image quality and lung nodule detectability of deep learning image reconstruction (DLIR) and adaptive statistical iterative reconstruction-V (ASIR-V) in ULD CT. Materials and Methods Patients who underwent noncontrast ULD CT (performed at 0.07 or 0.14 mSv, similar to a single chest radiograph) and contrast-enhanced chest CT (CECT) from April to June 2020 were included in this prospective study. ULD CT images were reconstructed with filtered back projection (FBP), ASIR-V, and DLIR. Three-dimensional segmentation of lung tissue was performed to evaluate image noise. Radiologists detected and measured nodules with use of a deep learning-based nodule assessment system and recognized malignancy-related imaging features. Bland-Altman analysis and repeated-measures analysis of variance were used to evaluate the differences between ULD CT images and CECT images. Results A total of 203 participants (mean age ± standard deviation, 61 years ± 12; 129 men) with 1066 nodules were included, with 100 scans at 0.07 mSv and 103 scans at 0.14 mSv. The mean lung tissue noise ± standard deviation was 46 HU ± 4 for CECT and 59 HU ± 4, 56 HU ± 4, 53 HU ± 4, 54 HU ± 4, and 51 HU ± 4 in FBP, ASIR-V level 40%, ASIR-V level 80% (ASIR-V-80%), medium-strength DLIR, and high-strength DLIR (DLIR-H), respectively, of ULD CT scans (P < .001). The nodule detection rates of FBP reconstruction, ASIR-V-80%, and DLIR-H were 62.5% (666 of 1066 nodules), 73.3% (781 of 1066 nodules), and 75.8% (808 of 1066 nodules), respectively (P < .001). Bland-Altman analysis showed the percentage difference in long diameter from that of CECT was 9.3% (95% CI of the mean: 8.0, 10.6), 9.2% (95% CI of the mean: 8.0, 10.4), and 6.2% (95% CI of the mean: 5.0, 7.4) in FBP reconstruction, ASIR-V-80%, and DLIR-H, respectively (P < .001). Conclusion Compared with adaptive statistical iterative reconstruction-V, deep learning image reconstruction reduced image noise, increased nodule detection rate, and improved measurement accuracy on ultra-low-dose chest CT images. © RSNA, 2022 Online supplemental material is available for this article. See also the editorial by Lee in this issue.

Is small airway dysfunction an abnormal phenomenon for patients with normal forced expiratory volume in 1 second and the ratio of forced expiratory volume in 1 second to forced vital capacity?

BACKGROUND: The clinical significance of small airway dysfunction (SAD) determined with spirometry in patients with normal forced expiratory volume in 1 second (FEV1) and the ratio of FEV1 to forced vital capacity (FVC) is controversial. OBJECTIVE: To determine whether SAD presents histologic abnormalities in the setting of normal computed tomography (CT) imaging and FEV1 and FEV1/FVC. METHODS: A cross-sectional study was performed in 64 patients undergoing thoracotomy for pulmonary nodules. Thoracic high-resolution CT (HRCT), bronchodilation test, and fractional exhaled nitric oxide (FENO) and its alveolar component (nitric oxide alveolar concentration [CANO]) were obtained before surgery. Lung pathology and levels of cytokines in lung tissue were measured. The patients were divided into SAD and small airway normal function groups according to forced expiratory flow at 75% and 50% of the FVC (maximal expiratory flow [MEF] 25, MEF50) and maximum midexpiratory flow. RESULTS: The MEF50, MEF25, and maximum midexpiratory flow were strongly negatively correlated with CANO (r, -0.42, -0.42, -0.40, respectively; P ≤ .001 for all). The MEFs were mildly negatively correlated with interleukin (IL)-6 and macrophages in lung tissue (r < -0.25, P < .001 for all). The CANO (P < .001), airspace size (mean linear intercept) (P = .02), macrophages (P = .003), IL-6 (P = .003), and IL-8 (P = .008) in lung tissue were higher in patients with SAD (n = 35) than those with small airway normal function (n = 29). A total of 8 patients (22.86%) with SAD and 2 (6.90%) without SAD had pneumatoceles (P = .10). CONCLUSION: Patients with pulmonary nodules and SAD were more likely to have abnormal inflammation and emphysematous destruction than patients without SAD. Thus, SAD indicates histologic abnormalities in patients with normal CT imaging and FEV1 and FEV1/FVC.

Baicalein ameliorates ulcerative colitis by improving intestinal epithelial barrier via AhR/IL-22 pathway in ILC3s.

Ulcerative colitis (UC) is a chronic inflammatory disease of the gastrointestinal tract, which is closely related to gut barrier dysfunction. Emerging evidence shows that interleukin-22 (IL-22) derived from group 3 innate lymphoid cells (ILC3s) confers benefits on intestinal barrier, and IL-22 expression is controlled by aryl hydrocarbon receptor (AhR). Previous studies show that baicalein protects the colon from inflammatory damage. In this study we elucidated the molecular mechanisms underlying the protective effect of baicalein on intestinal barrier function in colitis mice. Mice were administered baicalein (10, 20, 40 mg·kg-1·d-1, i.g.) for 10 days; the mice freely drank 3% dextran sulfate sodium (DSS) on D1-D7 to induce colitis. We showed that baicalein administration simultaneously ameliorated gut inflammation, decreased intestinal permeability, restored tight junctions of colons possibly via promoting AhR/IL-22 pathway. Co-administration of AhR antagonist CH223191 (10 mg/kg, i.p.) partially blocked the therapeutic effects of baicalein in colitis mice, whereas AhR agonist FICZ (1 µg, i.p.) ameliorated symptoms and gut barrier function in colitis mice. In a murine lymphocyte line MNK-3, baicalein (5-20 µM) dose-dependently increased the expression of AhR downstream target protein CYP1A1, and enhanced IL-22 production through facilitating AhR nuclear translocation, these effects were greatly diminished in shAhR-MNK3 cells, suggesting that baicalein induced IL-22 production in AhR-dependent manner. To further clarify that, we constructed an in vitro system consisting of MNK-3 and Caco-2 cells, in which MNK-3 cell supernatant treated with baicalein could decrease FITC-dextran permeability and promoted the expression of tight junction proteins ZO-1 and occluding in Caco-2 cells. In conclusion, this study demonstrates that baicalein ameliorates colitis by improving intestinal epithelial barrier via AhR/IL-22 pathway in ILC3s, thus providing a potential therapy for UC.

Combined analysis of imaging tumor capsule with imaging tumor size guides the width of resection margin for solitary hepatocellular carcinoma.

BACKGROUND: The optimal width of resection margin (RM) for hepatocellular carcinoma (HCC) remains controversial. This study aimed to investigate the value of imaging tumor capsule (ITC) and imaging tumor size (ITS) in guiding RM width for patients with HCC. METHODS: Patients who underwent hepatectomy for HCC in our center were retrospectively reviewed. ITC (complete/incomplete) and ITS (≤ 3 cm/> 3 cm) were assessed by preoperative magnetic resonance imaging (MRI). Using subgroup analyses based on ITC and ITS, the impact of RM width [narrow RM (< 5 mm)/wide RM (≥ 5 mm)] on recurrence-free survival (RFS), overall survival (OS), and RM recurrence was analyzed. RESULTS: A total of 247 patients with solitary HCC were included. ITC and ITS were independent predictors for RFS and OS in the entire cohort. In patients with ITS ≤ 3 cm, neither ITC nor RM width showed a significant impact on prognosis, and the incidence of RM recurrence was comparable between the narrow RM and wide RM groups (15.6% vs. 4.3%, P = 0.337). In patients with ITS > 3 cm and complete ITC, the narrow RM group exhibited comparable RFS, OS, and incidence of RM recurrence with the wide RM group (P = 0.606, 0.916, and 0.649, respectively). However, in patients with ITS > 3 cm and incomplete ITC, the wide RM group showed better RFS and OS and a lower incidence of RM recurrence compared with the narrow RM group (P = 0.037, 0.018, and 0.046, respectively). CONCLUSIONS: As MRI-based preoperative markers, conjoint analysis of ITC with ITS aids in determining RM width for solitary HCC patients. Narrow RM is applicable in patients with ITS ≤ 3 cm regardless of ITC status and in those with ITS > 3 cm and complete ITC. Wide RM is preferred in those with ITS > 3 cm and incomplete ITC.

Human-recognizable CT image features of subsolid lung nodules associated with diagnosis and classification by convolutional neural networks.

OBJECTIVES: The interpretability of convolutional neural networks (CNNs) for classifying subsolid nodules (SSNs) is insufficient for clinicians. Our purpose was to develop CNN models to classify SSNs on CT images and to investigate image features associated with the CNN classification. METHODS: CT images containing SSNs with a diameter of ≤ 3 cm were retrospectively collected. We trained and validated CNNs by a 5-fold cross-validation method for classifying SSNs into three categories (benign and preinvasive lesions [PL], minimally invasive adenocarcinoma [MIA], and invasive adenocarcinoma [IA]) that were histologically confirmed or followed up for 6.4 years. The mechanism of CNNs on human-recognizable CT image features was investigated and visualized by gradient-weighted class activation map (Grad-CAM), separated activation channels and areas, and DeepDream algorithm. RESULTS: The accuracy was 93% for classifying 586 SSNs from 569 patients into three categories (346 benign and PL, 144 MIA, and 96 IA in 5-fold cross-validation). The Grad-CAM successfully located the entire region of image features that determined the final classification. Activated areas in the benign and PL group were primarily smooth margins (p < 0.001) and ground-glass components (p = 0.033), whereas in the IA group, the activated areas were mainly part-solid (p < 0.001) and solid components (p < 0.001), lobulated shapes (p < 0.001), and air bronchograms (p < 0.001). However, the activated areas for MIA were variable. The DeepDream algorithm showed the image features in a human-recognizable pattern that the CNN learned from a training dataset. CONCLUSION: This study provides medical evidence to interpret the mechanism of CNNs that helps support the clinical application of artificial intelligence. KEY POINTS: • CNN achieved high accuracy (93%) in classifying subsolid nodules on CT images into three categories: benign and preinvasive lesions, MIA, and IA. • The gradient-weighted class activation map (Grad-CAM) located the entire region of image features that determined the final classification, and the visualization of the separated activated areas was consistent with radiologists' expertise for diagnosing subsolid nodules. • DeepDream showed the image features that CNN learned from a training dataset in a human-recognizable pattern.

Classification of moving coronary calcified plaques based on motion artifacts using convolutional neural networks: a robotic simulating study on influential factors.

BACKGROUND: Motion artifacts affect the images of coronary calcified plaques. This study utilized convolutional neural networks (CNNs) to classify the motion-contaminated images of moving coronary calcified plaques and to determine the influential factors for the classification performance. METHODS: Two artificial coronary arteries containing four artificial plaques of different densities were placed on a robotic arm in an anthropomorphic thorax phantom. Each artery moved linearly at velocities ranging from 0 to 60 mm/s. CT examinations were performed with four state-of-the-art CT systems. All images were reconstructed with filtered back projection and at least three levels of iterative reconstruction. Each examination was performed at 100%, 80% and 40% radiation dose. Three deep CNN architectures were used for training the classification models. A five-fold cross-validation procedure was applied to validate the models. RESULTS: The accuracy of the CNN classification was 90.2 ± 3.1%, 90.6 ± 3.5%, and 90.1 ± 3.2% for the artificial plaques using Inception v3, ResNet101 and DenseNet201 CNN architectures, respectively. In the multivariate analysis, higher density and increasing velocity were significantly associated with higher classification accuracy (all P < 0.001). The classification accuracy in all three CNN architectures was not affected by CT system, radiation dose or image reconstruction method (all P > 0.05). CONCLUSIONS: The CNN achieved a high accuracy of 90% when classifying the motion-contaminated images into the actual category, regardless of different vendors, velocities, radiation doses, and reconstruction algorithms, which indicates the potential value of using a CNN to correct calcium scores.

Motion-corrected coronary calcium scores by a convolutional neural network: a robotic simulating study.

OBJECTIVE: To classify motion-induced blurred images of calcified coronary plaques so as to correct coronary calcium scores on nontriggered chest CT, using a deep convolutional neural network (CNN) trained by images of motion artifacts. METHODS: Three artificial coronary arteries containing nine calcified plaques of different densities (high, medium, and low) and sizes (large, medium, and small) were attached to a moving robotic arm. The artificial arteries moving at 0-90 mm/s were scanned to generate nine categories (each from one calcified plaque) of images with motion artifacts. An inception v3 CNN was fine-tuned and validated. Agatston scores of the predicted classification by CNN were considered as corrected scores. Variation of Agatston scores on moving plaque and by CNN correction was calculated using the scores at rest as reference. RESULTS: The overall accuracy of CNN classification was 79.2 ± 6.1% for nine categories. The accuracy was 88.3 ± 4.9%, 75.9 ± 6.4%, and 73.5 ± 5.0% for the high-, medium-, and low-density plaques, respectively. Compared with the Agatston score at rest, the overall median score variation was 37.8% (1st and 3rd quartile, 10.5% and 68.8%) in moving plaques. CNN correction largely decreased the variation to 3.7% (1.9%, 9.1%) (p < 0.001, Mann-Whitney U test) and improved the sensitivity (percentage of non-zero scores among all the scores) from 65 to 85% for detection of coronary calcifications. CONCLUSIONS: In this experimental study, CNN showed the ability to classify motion-induced blurred images and correct calcium scores derived from nontriggered chest CT. CNN correction largely reduces the overall Agatston score variation and increases the sensitivity to detect calcifications. KEY POINTS: • A deep CNN architecture trained by CT images of motion artifacts showed the ability to correct coronary calcium scores from blurred images. • A correction algorithm based on deep CNN can be used for a tenfold reduction in Agatston score variations from 38 to 3.7% of moving coronary calcified plaques and to improve the sensitivity from 65 to 85% for the detection of calcifications. • This experimental study provides a method to improve its accuracy for coronary calcium scores that is a fundamental step towards a real clinical scenario.

Feasibility of bronchial wall quantification in low- and ultralow-dose third-generation dual-source CT: An ex vivo lung study.

PURPOSE: To investigate image quality and bronchial wall quantification in low- and ultralow-dose third-generation dual-source computed tomography (CT). METHODS: A lung specimen from a formerly healthy male was scanned using third-generation dual-source CT at standard-dose (51 mAs/120 kV, CTDIvol 3.41 mGy), low-dose (1/4th and 1/10th of standard dose), and ultralow-dose setting (1/20th). Low kV (70, 80, 90, and Sn100 kV) scanning was applied in each low/ultralow-dose setting, combined with adaptive mAs to keep a constant dose. Images were reconstructed at advanced modeled iterative reconstruction (ADMIRE) levels 1, 3, and 5 for each scan. Bronchial wall were semi-automatically measured from the lobar level to subsegmental level. Spearman correlation analysis was performed between bronchial wall quantification (wall thickness and wall area percentage) and protocol settings (dose, kV, and ADMIRE). ANOVA with a post hoc pairwise test was used to compare signal-to-noise ratio (SNR), noise and bronchial wall quantification values among standard- and low/ultralow-dose settings, and among ADMIRE levels. RESULTS: Bronchial wall quantification had no correlation with dose level, kV, or ADMIRE level (|correlation coefficients| < 0.3). SNR and noise showed no statistically significant differences at different kV in the same ADMIRE level (1, 3, or 5) and in the same dose group (P > 0.05). Generally, there were no significant differences in bronchial wall quantification among the standard- and low/ultralow-dose settings, and among different ADMIRE levels (P > 0.05). CONCLUSION: The combined use of low/ultralow-dose scanning and ADMIRE does not influence bronchial wall quantification compared to standard-dose CT. This specimen study suggests the potential that an ultralow-dose scan can be used for bronchial wall quantification.

Convolutional neural network to predict the local recurrence of giant cell tumor of bone after curettage based on pre-surgery magnetic resonance images.

OBJECTIVE: To predict the local recurrence of giant cell bone tumors (GCTB) on MR features and the clinical characteristics after curettage using a deep convolutional neural network (CNN). METHODS: MR images were collected from 56 patients with histopathologically confirmed GCTB after curettage who were followed up for 5.8 years (range, 2.0 to 9.5 years). The inception v3 CNN architecture was fine-tuned by two categories of the MR datasets (recurrent and non-recurrent GCTB) obtained through data augmentation and was validated using fourfold cross-validation to evaluate its generalization ability. Twenty-eight cases (50%) were chosen as the training dataset for the CNN and four radiologists, while the remaining 28 cases (50%) were used as the test dataset. A binary logistic regression model was established to predict recurrent GCTB by combining the CNN prediction and patient features (age and tumor location). Accuracy and sensitivity were used to evaluate the prediction performance. RESULTS: When comparing the CNN, CNN regression, and radiologists, the accuracies of the CNN and CNN regression models were 75.5% (95% CI 55.1 to 89.3%) and 78.6% (59.0 to 91.7%), respectively, which were higher than the 64.3% (44.1 to 81.4%) accuracy of the radiologists. The sensitivities were 85.7% (42.1 to 99.6%) and 87.5% (47.3 to 99.7%), respectively, which were higher than the 58.3% (27.7 to 84.8%) sensitivity of the radiologists (p < 0.05). CONCLUSION: The CNN has the potential to predict recurrent GCTB after curettage. A binary regression model combined with patient characteristics improves its prediction accuracy. KEY POINTS: • Convolutional neural network (CNN) can be trained successfully on a limited number of pre-surgery MR images, by fine-tuning a pre-trained CNN architecture. • CNN has an accuracy of 75.5% to predict post-surgery recurrence of giant cell tumors of bone, which surpasses the 64.3% accuracy of human observation. • A binary logistic regression model combining CNN prediction rate, patient age, and tumor location improves the accuracy to predict post-surgery recurrence of giant cell bone tumors to 78.6%.

Contrast-optimized composite image derived from multigradient echo cardiac magnetic resonance imaging improves reproducibility of myocardial contours and T2* measurement.

OBJECTIVES: Reproducibility of myocardial contour determination in cardiac magnetic resonance imaging is important, especially when determining T2* values per myocardial segment as a prognostic factor of heart failure or thalassemia. A method creating a composite image with contrasts optimized for drawing myocardial contours is introduced and compared with the standard method on a single image. MATERIALS AND METHODS: A total of 36 short-axis slices from bright-blood multigradient echo (MGE) T2* scans of 21 patients were acquired at eight echo times. Four observers drew free-hand myocardial contours on one manually selected T2* image (method 1) and on one image composed by blending three images acquired at TEs providing optimum contrast-to-noise ratio between the myocardium and its surrounding regions (method 2). RESULTS: Myocardial contouring by method 2 met higher interobserver reproducibility than method 1 (P < 0.001) with smaller Coefficient of variance (CoV) of T2* values in the presence of myocardial iron accumulation (9.79 vs. 15.91%) and in both global myocardial and mid-ventricular septum regions (12.29 vs. 16.88 and 5.76 vs. 8.16%, respectively). CONCLUSION: The use of contrast-optimized composite images in MGE data analysis improves reproducibility of myocardial contour determination, leading to increased consistency in the calculated T2* values enhancing the diagnostic impact of this measure of iron overload.

Chronic respiratory symptoms associated with airway wall thickening measured by thin-slice low-dose CT.

OBJECTIVE: In lung cancer screening, the prevalence of chronic respiratory symptoms is high among heavy smokers. The purpose of this study was to compare CT-derived airway wall measurements between male smokers with and those without chronic respiratory symptoms. MATERIALS AND METHODS: Fifty male heavy smokers with chronic respiratory symptoms (cough, excessive mucus secretion, dyspnea, and wheezing) and 50 without any respiratory symptom were randomly selected from the Dutch-Belgian Randomized Lung Cancer Screening Trial. Thin-slice low-dose CT images were evaluated with dedicated software for airway measurements. Wall area percentage and airway wall thickness were measured from trachea to bronchi in five different pulmonary lobes of airways with a luminal diameter of 5 mm or greater. Association between airway wall measurements and respiratory symptoms was analyzed by multiple linear regression adjusted for age, body mass index, smoking status, emphysema, and pulmonary function. RESULTS: After adjustment for relevant factors, a significant positive association between airway wall measurements and respiratory symptoms was found in airways with a luminal diameter between 5 to 10 mm (p < 0.01), but not in airways measuring 10 mm or greater (p > 0.05). At the airway level between 5 to 10 mm, the mean wall area percentages were 51.5% ± 7.9%. Airway wall thicknesses were 1.54 ± 0.39 mm and 1.37 ± 0.35 mm (p < 0.001). CONCLUSION: Male heavy smokers with chronic respiratory symptoms in lung cancer screening, who are at high-risk of chronic bronchitis, have bronchial wall thickening in airways with a luminal diameter of 5-10 mm but not in larger airways.

Small irregular pulmonary nodules in low-dose CT: observer detection sensitivity and volumetry accuracy.

OBJECTIVE: The purpose of this study is to evaluate observer detection and volume measurement of small irregular solid artificial pulmonary nodules on 64-MDCT in an anthropomorphic thoracic phantom. MATERIALS AND METHODS: Forty in-house-made solid pulmonary nodules (lobulated and spiculated; actual volume, 5.1-88.4 mm3; actual CT densities, -51 to 157 HU) were randomly placed inside an anthropomorphic thoracic phantom with pulmonary vasculature. The phantom was examined on two 64-MDCT scanners, using a scan protocol as applied in lung cancer screening. Two independent blinded observers screened for pulmonary nodules. Nodule volume was evaluated semiautomatically using dedicated software and was compared with the actual volume using an independent-samples t test. The interscanner and interobserver agreement of volumetry was assessed using Bland-Altman analysis. RESULTS: Observer detection sensitivity increased along with increasing size of irregular nodules. Sensitivity was 100% when the actual volume was at least 69 mm3, regardless of specific observer, scanner, nodule shape, and density. Overall, nodule volume was underestimated by (mean±SD) 18.9±11.8 mm3 (39%±21%; p<0.001). The relative interscanner difference of volumetry was 3.3% (95% CI, -33.9% to 40.4%). The relative interobserver difference was 0.6% (-33.3% to 34.5%). CONCLUSION: Small irregular solid pulmonary nodules with an actual volume of at least 69 mm3 are reliably detected on 64-MDCT. However, CT-derived volume of those small nodules is largely underestimated, with considerable variation.

Risk stratification of thymic epithelial tumors based on peritumor CT radiomics and semantic features.

OBJECTIVES: To develop and validate nomograms combining radiomics and semantic features to identify the invasiveness and histopathological risk stratification of thymic epithelial tumors (TET) using contrast-enhanced CT. METHODS: This retrospective multi-center study included 224 consecutive cases. For each case, 6764 intratumor and peritumor radiomics features and 31 semantic features were collected. Multi-feature selections and decision tree models were performed on radiomics features and semantic features separately to select the most important features for Masaoka-Koga staging and WHO classification. The selected features were then combined to create nomograms for the two systems. The performance of the radiomics model, semantic model, and combined model was evaluated using the area under the receiver operating characteristic curves (AUCs). RESULTS: One hundred eighty-seven cases (56.5 years ± 12.3, 101 men) were included, with 62 cases as the external test set. For Masaoka-Koga staging, the combined model, which incorporated five peritumor radiomics features and four semantic features, showed an AUC of 0.958 (95% CI: 0.912-1.000) in distinguishing between early-stage (stage I/II) and advanced-stage (III/IV) TET in the external test set. For WHO classification, the combined model incorporating five peritumor radiomics features and two semantic features showed an AUC of 0.857 (0.760-0.955) in differentiating low-risk (type A/AB/B1) and high-risk (B2/B3/C) TET. The combined models showed the most effective predictive performance, while the semantic models exhibited comparable performance to the radiomics models in both systems (p > 0.05). CONCLUSION: The nomograms combining peritumor radiomics features and semantic features could help in increasing the accuracy of grading invasiveness and risk stratification of TET. CRITICAL RELEVANCE STATEMENT: Peripheral invasion and histopathological type are major determinants of treatment and prognosis of TET. The integration of peritumoral radiomics features and semantic features into nomograms may enhance the accuracy of grading invasiveness and risk stratification of TET. KEY POINTS: Peritumor region of TET may suggest histopathological and invasive risk. Peritumor radiomic and semantic features allow classification by Masaoka-Koga staging (AUC: 0.958). Peritumor radiomic and semantic features enable the classification of histopathological risk (AUC: 0.857).

Paeoniflorin ameliorates chronic colitis via the DR3 signaling pathway in group 3 innate lymphoid cells.

Inhibiting the death receptor 3 (DR3) signaling pathway in group 3 innate lymphoid cells (ILC3s) presents a promising approach for promoting mucosal repair in individuals with ulcerative colitis (UC). Paeoniflorin, a prominent component of Paeonia lactiflora Pall., has demonstrated the ability to restore barrier function in UC mice, but the precise mechanism remains unclear. In this study, we aimed to delve into whether paeoniflorin may promote intestinal mucosal repair in chronic colitis by inhibiting DR3 signaling in ILC3s. C57BL/6 mice were subjected to random allocation into 7 distinct groups, namely the control group, the 2 % dextran sodium sulfate (DSS) group, the paeoniflorin groups (25, 50, and 100 mg/kg), the anti-tumor necrosis factor-like ligand 1A (anti-TL1A) antibody group, and the IgG group. We detected the expression of DR3 signaling pathway proteins and the proportion of ILC3s in the mouse colon using Western blot and flow cytometry, respectively. Meanwhile, DR3-overexpressing MNK-3 cells and 2 % DSS-induced Rag1-/- mice were used for verification. The results showed that paeoniflorin alleviated DSS-induced chronic colitis and repaired the intestinal mucosal barrier. Simultaneously, paeoniflorin inhibited the DR3 signaling pathway in ILC3s and regulated the content of cytokines (Interleukin-17A, Granulocyte-macrophage colony stimulating factor, and Interleukin-22). Alternatively, paeoniflorin directly inhibited the DR3 signaling pathway in ILC3s to repair mucosal damage independently of the adaptive immune system. We additionally confirmed that paeoniflorin-conditioned medium (CM) restored the expression of tight junctions in Caco-2 cells via coculture. In conclusion, paeoniflorin ameliorates chronic colitis by enhancing the intestinal barrier in an ILC3-dependent manner, and its mechanism is associated with the inhibition of the DR3 signaling pathway.