Accelerated 3D MR neurography of the brachial plexus using deep learning-constrained compressed sensing.

OBJECTIVES: To explore the use of deep learning-constrained compressed sensing (DLCS) in improving image quality and acquisition time for 3D MRI of the brachial plexus. METHODS: Fifty-four participants who underwent contrast-enhanced imaging and forty-one participants who underwent unenhanced imaging were included. Sensitivity encoding with an acceleration of 2 × 2 (SENSE4x), CS with an acceleration of 4 (CS4x), and DLCS with acceleration of 4 (DLCS4x) and 8 (DLCS8x) were used for MRI of the brachial plexus. Apparent signal-to-noise ratios (aSNRs), apparent contrast-to-noise ratios (aCNRs), and qualitative scores on a 4-point scale were evaluated and compared by ANOVA and the Friedman test. Interobserver agreement was evaluated by calculating the intraclass correlation coefficients. RESULTS: DLCS4x achieved higher aSNR and aCNR than SENSE4x, CS4x, and DLCS8x (all p < 0.05). For the root segment of the brachial plexus, no statistically significant differences in the qualitative scores were found among the four sequences. For the trunk segment, DLCS4x had higher scores than SENSE4x (p = 0.04) in the contrast-enhanced group and had higher scores than SENSE4x and DLCS8x in the unenhanced group (all p < 0.05). For the divisions, cords, and branches, DLCS4x had higher scores than SENSE4x, CS4x, and DLCS8x (all p ≤ 0.01). No overt difference was found among SENSE4x, CS4x, and DLCS8x in any segment of the brachial plexus (all p > 0.05). CONCLUSIONS: In three-dimensional MRI for the brachial plexus, DLCS4x can improve image quality compared with SENSE4x and CS4x, and DLCS8x can maintain the image quality compared to SENSE4x and CS4x. CLINICAL RELEVANCE STATEMENT: Deep learning-constrained compressed sensing can improve the image quality or accelerate acquisition of 3D MRI of the brachial plexus, which should be benefit in evaluating the brachial plexus and its branches in clinical practice. KEY POINTS: •Deep learning-constrained compressed sensing showed higher aSNR, aCNR, and qualitative scores for the brachial plexus than SENSE and CS at the same acceleration factor with similar scanning time. •Deep learning-constrained compressed sensing at acceleration factor of 8 had comparable aSNR, aCNR, and qualitative scores to SENSE4x and CS4x with approximately half the examination time. •Deep learning-constrained compressed sensing may be helpful in clinical practice for improving image quality and acquisition time in three-dimensional MRI of the brachial plexus.

[Application of Deep Learning Reconstruction Algorithm in Low-Dose Thin-Slice Liver CT of Healthy Volunteers].

OBJECTIVE: To explore the clinical feasibility of applying deep learning (DL) reconstruction algorithm in low-dose thin-slice liver CT examination of healthy volunteers by comparing the reconstruction algorithm based on DL, filtered back projection (FBP) reconstruction algorithm and iterative reconstruction (IR) algorithm. METHODS: A standard water phantom with a diameter of 180 mm was scanned, using the 160 slice multi-detector CT scanning of United Imaging Healthcare, to compare the noise power spectrums of DL, FBP and IR algorithms. 100 healthy volunteers were prospectively enrolled, with 50 assigned to the normal dose group (ND) and 50 to the low dose group (LD). IR algorithm was used in the ND group to reconstruct images, while DL, FBP and IR algorithms were used in the LD group to reconstruct images. One-way analysis of variance was used to compare the liver CT values, the liver noise, liver signal-to-noise ratio (SNR), contrast noise ratio (CNR) and figure of merit (FOM) of the images of ND-IR, LD-FBP, LD-IR and LD-DL. The Kruskal-Wallis test was used to analyse subjective scores of anatomical structures. RESULTS: The DL algorithm had the lowest average peak value of noise power spectrum, and its shape was similar to that of medium-level IR algorithm. Liver CT values of ND-IR, LD-FBP, LD-IR and LD-DL did not show statistically significant difference. The noise of LD-DL was lower than that of LD-FBP, LD-IR and ND-IR ( P<0.05), and the SNR, CNR and FOM of LD-DL were higher than those of LD-FBP, LD-IR and ND-IR ( P<0.05). The subjective scores of anatomical structures of LD-DL did not show significant difference compared to those of ND-IR ( P >0.05), and were higher than those of LD-FBP and LD-IR. The radiation dose of the LD group was reduced by about 50.2% compared with that of the ND group. CONCLUSION: The DL algorithm with noise shape similar to the medium iterative grade IR commonly used in clinical practice showed higher noise reduction ability than IR did. Compared with FBP, the DL algorithm had smoother noise shape, but much better noise reduction ability. The application of DL algorithm in low-dose thin-slice liver CT of healthy volunteers can help achieve the standard image quality of liver CT.

[Application of Automated Machine Learning Based on Radiomics Features of T2WI and RS-EPI DWI to Predict Preoperative T Staging of Rectal Cancer].

OBJECTIVE: To explore the radiomics features of T2 weighted image (T2WI) and readout-segmented echo-planar imaging (RS-EPI) plus difusion-weighted imaging (DWI), to develop an automated mahchine-learning model based on the said radiomics features, and to test the value of this model in predicting preoperative T staging of rectal cancer. METHODS: The study retrospectively reviewed 131 patients who were diagnosed with rectal cancer confirmed by the pathology results of their surgical specimens at West China Hospital of Sichuan University between October, 2017 and December, 2018. In addition, these patients had preoperative rectal MRI. Tumor regions from preoperative MRI were manually segmented by radiologists with the ITK-SNAP software from T2WI and RS-EPI DWI images. PyRadiomics was used to extract 200 features-100 from T2WI and 100 from the apparent diffusion coefficient (ADC) calculated from the RS-EPI DWI. MWMOTE and NEATER were used to resample and balance the dataset, and 13 cases of T 1-2 stage simulation cases were added. The overall dataset was divided into a training set (111 cases) and a test set (37 cases) by a ratio of 3∶1. Tree-based Pipeline Optimization Tool (TPOT) was applied on the training set to optimize model parameters and to select the most important radiomics features for modeling. Five independent T stage models were developed accordingly. Accuracy and the area under the curve ( AUC) of receiver operating characteristic (ROC) were used to pick out the optimal model, which was then applied on the training set and the original dataset to predict the T stage of rectal cancer. RESULTS: The performance of the the five T staging models recommended by automated machine learning were as follows: The accuracy for the training set ranged from 0.802 to 0.838, sensitivity, from 0.762 to 0.825, specificity, from 0.833 to 0.896, AUC, from 0.841 to 0.893, and average precision (AP) from 0.870 to 0.901. After comparison, an optimal model was picked out, with sensitivity, specificity and AUC for the training set reaching 0.810, 0.875, and 0.893, respectively. The sensitivity, specificity and AUC for the test set were 0.810, 0.813, and 0.810, respectively. The sensitivity, specificity and AUC for the original dataset were 0.810, 0.830, and 0.860, respectively. CONCLUSION: Based on the radiomics data of T2WI and RS-EPI DWI, the model established by automated machine learning showed a fairly high accuracy in predicting rectal cancer T stage.

[Noise Reduction Effect of Deep-learning-based Image Reconstruction Algorithms in Thin-section Chest CT].

OBJECTIVE: To evaluate the noise reduction effect of deep learning-based reconstruction algorithms in thin-section chest CT images by analyzing images reconstructed with filtered back projection (FBP), adaptive statistical iterative reconstruction (ASIR), and deep learning image reconstruction (DLIR) algorithms. METHODS: The chest CT scan raw data of 47 patients were included in this study. Images of 0.625 mm were reconstructed using six reconstruction methods, including FBP, ASIR hybrid reconstruction (ASIR50%, ASIR70%), and deep learning low, medium and high modes (DL-L, DL-M, and DL-H). After the regions of interest were outlined in the aorta, skeletal muscle and lung tissue of each group of images, the CT values, SD values and signal-to-noise ratio (SNR) of the regions of interest were measured, and two radiologists evaluated the image quality. RESULTS: CT values, SD values and SNR of the images obtained by the six reconstruction methods showed statistically significant difference ( P<0.001). There were statistically significant differences in the image quality scores of the six reconstruction methods ( P<0.001). Images reconstruced with DL-H have the lowest noise and the highest overall quality score. CONCLUSION: The model based on deep learning can effectively reduce the noise of thin-section chest CT images and improve the image quality. Among the three deep-learning models, DL-H showed the best noise reduction effect.

[The Application Value of Artificial Intelligence-based Filtering and Interpolated Image Reconstruction Algorithm in Abdominal Magnetic Resonance Image Denoising].

OBJECTIVE: To compare the noise reduction performance of conventional filtering and artificial intelligence-based filtering and interpolation (AIFI) and to explore for optimal parameters of applying AIFI in the noise reduction of abdominal magnetic resonance imaging (MRI). METHODS: Sixty patients who underwent upper abdominal MRI examination in our hospital were retrospectively included. The raw data of T1-weighted image (T1WI), T2-weighted image (T2WI), and dualecho sequences were reconstructed with two image denoising techniques, conventional filtering and AIFI of different levels of intensity. The difference in objective image quality indicators, peak signal-to-noise ratio (pSNR) and image sharpness, of the different denoising techniques was compared. Two radiologists evaluated the image noise, contrast, sharpness, and overall image quality. Their scores were compared and the interobserver agreement was calculated. RESULTS: Compared with the original images, improvement of varying degrees were shown in the pSNR and the sharpness of the images of the three sequences, T1W1, T2W2, and dual echo sequence, after denoising filtering and AIFI were used (all P<0.05). In addition, compared with conventional filtering, the objective quality scores of the reconstructed images were improved when conventional filtering was combined with AIFI reconstruction methods in T1WI sequence, AIFI level≥3 was used in T2WI and echo1 sequence, and AIFI level≥4 was used in echo2 sequence (all P<0.05). The subjective scores given by the two radiologists for the image noise, contrast, sharpness, and overall image quality in each sequence of conventional filtering reconstruction, AIFI reconstruction (except for AIFI level=1), and two-method combination reconstruction were higher than those of the original images (all P<0.05). However, the image contrast scores were reduced for AIFI level=5. There was good interobserver agreement between the two radiologists (all r>0.75, P<0.05). After multidimensional comparison, the optimal parameters of using AIFI technique for noise reduction in abdominal MRI were conventional filtering+AIFI level=3 in the T1WI sequence and AIFI level=4 in the T2WI and dualecho sequences. CONCLUSION: AIFI is superior to filtering in imaging denoising at medium and high levels. It is a promising noise reduction technique. The optimal parameters of using AIFI for abdominal MRI are Filtering+AIFI level=3 in the T1WI sequence and AIFI level=4 in T2WI and dualecho sequences.

[Application of MRI-based Radiomics Models in the Assessment of Hepatic Metastasis of Rectal Cancer].

OBEJECTIVE: To explore the clinical value of using radiomics models based on different MRI sequences in the assessment of hepatic metastasis of rectal cancer. METHODS: 140 patients with pathologically confirm edrectal cancer were included in the study. They underwent baseline magnetic resonance imaging (MRI) between April 2015 and May 2018 before receiving any treatment. According to the results of liver biopsy, surgical pathology, and imaging, patients were put into two groups, the patients with hepatic metastasis and those without. T2 weighted images (T2WI), diffusion weighted images (DWI) and apparent diffusion coefficient (ADC) images were used to draw the region of interest (ROI) of primary lesions on consecutive slices on ITK-SNAP. 3-D ROIs were generated and loaded into Artificial Intelligent Kit for extraction of radiomics features and 396 features were extracted for each sequence. The feature data were preprocessed on Python and the samples were oversampled, using Support Vector Machine-Synthetic Minority Over-Sampling Technique (SVM-SMOTE) to balance the number of samples in the group with liver metastasis and the group with no liver metastasis at the end of the follow-up. Then, the samples were divided into the training cohort and the test cohort at a ratio of 2∶1. The logistic regression models were developed with selected radionomic features on R software. The receiver operating characteristics (ROC) curves and calibration curves were used to evaluate the performance of the models. RESULTS: In total, 52 patients with liver metastasis and 88 patients without liver metastasis at the end of follow-up were enrolled. Carcinoembryonic antigen (CEA) and T stage and N stage evaluated on the MRI images showed statistically significant difference between the two groups ( P<0.05). After data preprocessing and selecting, except for 17 non-radiomic features, the model combining T2WI, DWI and ADC features, the model of T2WI features alone, the model of DWI features alone and the model of ADC features alone were developed with 32 features, 10 features, 30 features and 15 features, respectively. The combined model (T2WI+DWI+ADC), the T2WI model, and the ADC model can assess hepatic metastasis accurately, with the area under curve ( AUC) on the train set reaching 93.5%, 89.2%, 90.6% and that of the test set reaching 80.8%, 80.5%, 81.4%, respectively. The combined model did not show a higher AUC than those of the T2WI and ADC alone models. Model based on DWI features has a slightly insufficient AUC of 90.3% in the train set and 75.1% in the test set. The calibration curve showed the smallest fluctuation in the combined model, which is closest fit to the diagonal reference line. The fluctuation in the three independent data set models were similar. The calibration curves of all the four models showed that as the risk increased, the prediction of the models turned from an underestimation to an overestimating the risk. In brief, the combined model showed the best performance, with the best fit to the diagonal reference line in calibration curve and high AUC comparable to the AUC of the T2WI model and ADC model. The performance of T2WI and ADC alone models were second to that of the combined model, while the DWI alone model showed relatively poor performance. CONCLUSION: Radiomics models based on MRI could be effectively used in assessing liver metastasis in rectal cancer, which may help determine clinical staging and treatment.

[Application of 3.0T Time-of-flight Magnetic Resonance Angiography with Sparse Undersampling and Iterative Reconstruction in the Diagnosis of Unruptured Intracranial Aneurysms].

OBJECTIVE: To evaluate the diagnostic value of 3.0T time-of-flight MR angiography with sparse undersampling and iterative reconstruction (TOFu-MRA) for unruptured intracranial aneurysms (UIAs) on the basis of using digital subtraction angiography (DSA) as the reference standard. METHODS: A total of 65 patients with suspected UIAs were prospectively enrolled and all patients underwent TOFu-MRA and DSA. Relying on DSA as the reference standard, the sensitivity (SEN), specificity (SPE), positive predictive value (PPV) and negative predictive value (NPV) of using TOFu-MRA in UIA diagnosis were calculated, and the inter-observer agreement between two doctors was determined. Comparison of maximum intensity projection (MIP) and volume rendering (VR) image datasets was made to evaluate the agreement between DSA results and TOFu-MRA in the measurement of UIA morphological parameters, including the neck width (D neck), height (H) , and width (D width) of UIAs. RESULTS: The study covered 55 UIAs from 46 patients. The SEN, SPE, PPV and NPV of the two doctors using TOFu-MRA in UIA diagnosis were as follows: (95.7%, 95.7%), (94.7%, 94.7%), (97.8%, 97.8%) and (90.0%, 90.0%), respectively for patient-based assessment; (96.4%, 94.5%), (94.7%, 94.7%), (98.1%, 98.1%) and (90.0%, 85.7%), respectively, for aneurysm-based assessment. There is a strong inter-observer agreement (Kappa=0.93 for patient-based assessment and 0.96 for aneurysm-based assessment) between the two doctors. Moreover, Bland-Altman analysis showed that more than 95% points fell within the limits of agreement (LoA), suggesting strong agreement between the two examination methods for the measurement of UIAs morphological parameters. CONCLUSION: TOFu-MRA showed good diagnostic efficacy for UIAs and the results were in good agreement with those of DSA, the reference standard, for assessing UIA morphological parameter. TOFu-MRA can be used as a first choice for noninvasive diagnostic evaluation of UIAs.

[The Application Value of One-stop CT Myocardial Perfusion Imaging in the Evaluation of Patients with Severe Coronary Artery Stenosis].

ObstractPurpose "One-stop" CT myocardial perfusion imaging (CT-MPI) was compared with cardiac magnetic resonance(CMR) to investigate its application value in evaluating patients with severe coronary artery stenosis.MethodsFifty patients with coronary artery stenosis≥90% of at least one major coronary arteries comfirmed by coronary angiography (CAG) in the department of cardiology in our hospital, who referred for coronary artery stent implantation were prospectively enrolled. All the patients underwent "One-stop" CT-MPI within a week before surgery, among which 22 patients underwent CMR examination simultaneously. The postprocessing software Ziostation2 was used to obatin and compare the perfusion parameters of patients with normal and perfusion defect myocardium, including blood flow (BF), blood volume (BV), peak time (TTP), and mean transit time (MTT). Pearson correlation analysis was used to compare the correlation of relative perfusion parameters (defect/normal myocardium) between CT and CMR. Bland-Altman analysis was used to analyze the consistency between CT and CMR in left ventricular (LV) function parameters measurements.ResultsCompared with normal myocardium, BV and BF of perfusion defect myocardium were significantly decreased, while MTT and TTP were significantly prolonged (all P < 0.05). The rBV, rBF, rMTT and rTTP were medium to high positive correlated between CT and CMR (r=0.685, 0.641, 0.871, 0.733, respectively, all P < 0.05). Bland-Altman analysis showed that 95% (21/22) points were within the 95% limits of agreement (LoA), suggesting the LV function parameters measurements between two methods were highly consistent.Conclusion"One-stop" CT-MPI can simultaneously obtain the information about coronary anatomy, myocardial perfusion and LV function. It is of great value in the evaluation of patients with severe coronary artery stenosis, with shorter scan time and less contraindications compared with CMR.

[Automated Estimation of Stenosis Severity in Coronary Computed Tomography Angiography].

OBJECTIVE: To determine the value of automated detection in computed tomography angiography (CTA) for cases with greater than 70% coronary stenosis. METHODS: Fifty-seven patients who had both coronary CTA and digital subtraction angiography (DSA) were retrospectively recruited in this study. The patients were categorized into two groups using a cutoff value of 70% stenosis in DSA. The AW4.6 software was used to estimate the diameter and square values from the data obtained from CTA. The sensitivity (SE), specificity (SPE), positive predictive value (PPV) and negative predictive value (NPV) of the automated CTA estimations were calculated. RESULTS: A total of 178 vessels from the 57 patients were analyzed. The automated CTA estimations had moderate to high levels of agreements (Kappa value: 0.716-0.804, P < 0.001) with the DSA diagnoses, compared with low to moderate levels of agreements (Kappa value: 0.385-0.533, P < 0.001) in manual interpretations. The square estimations generated high SE (100%) and NPV (100%) for patient diagnoses (P < 0.016 7 vs. manual interpretations). The diameter estimations generated high SPE (90.48%) and PPV (94.12%) for patient diagnoses (P < 0.016 7, vs. manual interpretations). Similarly, high SE (96.92%) and NPV (97.89%) were found for square estimations in vessel diagnoses, while high SPE (94.69%) and PPV (90.16%) were found for diameter estimations in vessel diagnoses. CONCLUSIONS: Both automated diameter and square algorithms have high accuracy for diagnosing patients with greater than 70% coronary artery stenosis. The AW4.6 can improve the detection of severe stenosis that needs stent interventions.