Purpose: Perfusion-weighted non-contrast enhanced proton lung MRI during free breathing is maturing as a novel technique for assessment of regional lung perfusion, but has yet not been validated in COPD patients. The goal of this study was to determine if phase-resolved functional lung (PREFUL-) MRI correlates with dynamic contrast enhanced (DCE)-MRI as established reference and lung function testing.
Methods and Materials: Forty-seven patients with stable COPD were included as a single center subgroup analysis nested in the German COPD cohort study COSYCONET - a prospective, observational, multicenter cohort. For PREFUL-MRI a spoiled gradient echo sequence and for DCE-MRI a contrast-enhanced three-dimensional time-resolved spoiled gradient echo sequence was used at 1.5T. Slices of PREFUL and DCE-MRI were matched using a landmark based approach and co-registered. Perfusion defect percentages (QDP) were calculated for both methods and correlated with each other and lung function testing using Spearman’s correlation coefficient (r), spatial overlap metrics and Bland-Altman plot analysis.
Results: PREFUL-MRI and DCE-MRI matched spatially with a spatial overlap of 62.2(57.2-67.2)%. QDP were comparable on a global 39.3(31.8-45.5)% vs. 44.7(35.4-50.0)%) as well as on a lobar level, although a systematic overestimation of PREFUL-QDP compared to PBF-QDP mainly in the lower lobes resulted in an overall overestimation for the whole lung with a mean difference of 5%(95% confidence interval [CI]: 3.0%;7.0%). Significant correlations with lung function test parameters was found for both methods, such as PREFUL-QDP vs. FEV1, r=-0.75, P<0.0001.
Conclusion: PREFUL-MRI is a promising noninvasive, radiation-free tool for quantification of regional perfusion in COPD patients.
Purpose: To develop and validate a fully automated 3D quantification of emphysema extent in COPD patients using 3D-MRI with ultrashort echo time (UTE).
Methods and Materials: Twenty-nine COPD patients (21 males/ 8 females, age 70+-7 years) and 11 healthy volunteers (10 males/1 female, age 64+-4 years) were prospectively enrolled between 2014 and 2017. All patients underwent pulmonary function test, 3D-UTE MRI at millimetre spatial resolution (voxel size=1mm3) and MDCT scan the same day. Two readers experienced in chest imaging performed the evaluation. 3D automatic quantification of relative low signal area (LSA%) and relative low attenuation area (LAA%<-950HU) were measured as a surrogate for emphysema using 3D-UTE MRI and MDCT, respectively. Skewness of lung MR frequency distribution histogram was also determined. Comparison of means was performed using t-test and correlation using Pearson test. Reproducibility was assessed using Lin's concordance correlation coefficient.
Results: Using 3D-UTE, LSA% was significantly higher in COPD patients than in healthy volunteers (p<0.001). Skewness of lung MR frequency distribution histogram was significantly right tailed towards the lowest signal values in COPD (p<0.001). In 29 COPD patients, both LSA% and skewness correlated well with LAA%<-950HU (r=0.77, p<0.001; r=0.87, p<0.001, respectively) and FEV1 (r=-0.43, p=0.01; r=-0.55, p<0.001, respectively). Intra and inter-observer reproducilbilty of both LSA% and skewness were found perfect using Lin's concordance correlation coefficient (ρc>0.99).
Conclusion: 3D automatic quantification of the lung emphysema is feasible and reproducible using non-contrast enhanced 3D-UTE MRI and may prove useful in the follow-up of COPD patients without radiation exposure.
Purpose: To directly and prospectively compare the quantitative capability for diagnosis of solitary pulmonary nodules (SPNs) among diffusion-weighted imaging (DWI) with fast advanced spin-echo (FASE) and echo planar imaging (EPI) sequences in a 3T system, DWI with EPI sequence at a 1.5T system and FDG-PET/CT.
Methods and Materials: 97 consecutive patients with 129 SPNs underwent DWIs with FASE and/or EPI sequences at 3T and 1.5T systems, FDG-PET/CT, and pathological and/or follow-up examinations. According to final diagnoses, all SPNs were divided into malignant (n=87) and benign (n=42) SPNs. In each lesion, apparent diffusion coefficients (ADCs) from all DWIs (ADC3TFASE, ADC3TEPI and ADC1.5TEPI) and SUVmax were assessed by ROI measurements. To compare all indexes between two groups, Student’s t test was performed. Then ROC analyses were performed to compare diagnostic performance among all indexes. Finally, sensitivity, specificity and accuracy were compared among all methods by McNemar’s test.
Results: There was significant difference of all indexes between two SPN groups (p<0.0001). ROC analyses showed area under the curve (Az) of ADC3TFASE was significantly larger than that of others (p<0.05). Accuracy (AC) of ADC3TFASE (90.7%) was significantly higher than that of others (ADC1.5TEPI: 84.4%, p=0.008; ADC3TEPI: 82.9%, p=0.002; SUVmax 75.2%, p<0.0001).
Conclusion: DWIs with FASE sequence has a better potential for quantitative diagnosis of SPNs than DWI with EPI sequence at 1.5T and 3T systems and FDG-PET/CT. FASE sequence would be better to be applied for DWI at 3T system to improve diagnostic performance of SPNs.
Purpose: To differentiate fibrotic and inflammatory GGO using T2-weighted (T2-w) and contrast-enhanced MRI (CEMRI).
Methods and Materials: After consent, 3 patients with IPF/NSIP underwent CT and MRI on the same day, which were repeated after 6-month-therapy. MRI consisted of 3D-SPGR-PD-w, 2D-PROPELLER, 3D-CUBE-T2-w and CEMRI acquired at 5-10-20 minutes. Two main radiological interstitial lung disease (ILD) features were quantitatively scored with PRAGMA-ILD on MRI: Fibrotic(FL) and Normal Lung(NL) tissue. FL included all fibrosis-related changes (honeycombing, reticulation, GGO). Data were expressed as ml and % total lung volume. MRI data were compared with CT data scored with CALIPER-software and clinical data. Based on T2-w signal intensity (SI) and CEMRI, FL was classified as inflammatory. Statistic included Pearson correlation and Mann-Whitney-U test.
Results: MRI-score had excellent correlation with CALIPER (r=0.97, p<0.00001). MRI-score overestimated fibrosis (mean-difference 15.3 % SD 7.5 %) and underestimated normal lung (mean-difference 17.8 SD 7.5) compared to CT. 4/9 ROIs showed significant difference in T2-w SI and were labelled as inflammatory. The mean SI differences FL were 106.8 (SD 32.1) and 2.1 (SD 34.4) for the inflammatory and the stable ROI, respectively. Inflammatory ROIs had higher SI difference than stable ROI between baseline and follow-up (p<0.0001). Two inflammatory areas, as expected in the NSIP, evolved in fibrotic tissue. CEMRI showed rapid peak of enhancement in the FL suspicious for active inflammation.
Conclusion: In our pilot study ILD morphological changes can be reliably be quantified with MRI. T2-w SI and CEMRI can identify inflammation area and their evolution during ILD therapy.
Purpose: The aim of our study was to evaluate the availability of Magnetic Resonance Spectroscopy (MRS), which is a non-invasive method for the differentiation of benign or malignant pulmonary masses.
Methods and Materials: A total of 59 patients (45 male, 14 female) with pulmonary masses were included in this prospective study. MRS was applied to the pulmonary lesions of the patients. Cholin (Cho), lipid and H2O parameters were determined in TE 50, 135 and 270 bands in MRS. Afterwards CT-guided percutaneous needle biopsy was performed. According to the biopsy results, pulmonary masses in 25 patients were benign and pulmonary masses in 34 patients were malignant.
Results: Cho, Cho/H2O and Cho/lipid levels were significantly higher in TE 50 band in malignant group compared to benign group (p <0.001, p = 0.011 and p <0.001). Cho, Cho/H2O and Cho/lipid levels were significantly higher in TE 135 band in malignant group compared to benign group (p <0.001, p = 0.012 and p = 0.012). Cho and lipid levels were significantly higher in TE 270 band in malignant group compared to benign group (p = 0.050 and p = 0.013).
Conclusion: When the other conditions were kept constant, the probability of malignancy was increased significantly 17.381 times (95% CI: 3.779-79.932) in those with Cho levels above 1.65 compared to those with Cho levels below 1.65 in the TE 50 band (p<0.001). In conclusion, MRS can be used in the differential diagnosis of pulmonary masses.
Purpose: To compare imaging parameters of Dual energy CT (DECT) with MRI for differentiating Thymoma from Thymic Hyperplasia (TH) and correlate the imaging findings of thymoma with the WHO grade & Masaoka Koga (MK) stage among patients with Myasthenia Gravis (MG).
Methods and Materials: Fifty-four patients of MG were recruited and underwent a single venous phase DECT (n=29), a Dynamic Contrast Enhanced MRI (1.5T, n=41) or both (n=23). Based on the histopathology they were divided into two groups (i) Thymoma (n=27) & (ii) Thymic hyperplasia (n=18), while patients with normal thymus glands (n=9) were excluded. Statistical analysis was performed between the two groups for DECT and MRI.
Results: Quantitative MRI parameters including Signal intensity index (SII) and Chemical Shift Ratio (CSR) showed a sensitivity & specificity of 47%/75% and 65%/86% respectively for detecting thymoma, at present published cutoff values (SII<7.77, CSR>0.849). Using ROC analysis, we propose a cutoff for SII<38% & CSR>0.75 which improves the sensitivity and specificity to 65%/68% & 89%/86% respectively. Quantitative DECT parameters using our proposed cutoffs for attenuation on Virtual Non-Contrast image (>33.3 HU) and Venous Phase (>51.3 HU), & fat fraction (<17.8%) showed a sensitivity and specificity of 100%/100%, 89.9%/90.9%, & 100%/100% respectively. DECT was also superior to MRI in predicting WHO grade, however no quantitative parameter on either DECT or MRI could differentiate the MK stages.
Conclusion: DECT has the potential to provide equal if not greater information than MRI for characterizing thymic lesions in patients with MG, however further validation is required.
Purpose: A substudy of 400 whole-body MRI scans from the KORA-FF4 cohort showed good correlation between MRI-derived lung volumes and residual volume derived by pulmonary function tests and chronic obstructive pulmonary disease (COPD). As cardiac dysfunction often accompanies COPD, we aimed to evaluate this relationship using whole-body MRI.
Methods and Materials: 400 subjects without cardiovascular diseases from the KORA-FF4 cohort study underwent whole-body MRI. Pulmonary volumes were derived from coronal T1w-sequences. Cardiac function was assessed from cine-SSFP sequences using cvi42, and LV filling rates were assessed using pyHeart (in-house developed). Cardiac function parameters were standardized to body surface area, and its association to lung volume was analyzed using Pearson correlation and multivariate linear regression models.
Results: MRI parameters were available for 356 subjects (56.4±9.2 years). Cardiac measurements were within the normal physiologic range, and the mean lung volume was 4.0±1.1L. In a univariate model, left ventricle (LV) and right ventricle (RV) stroke volume were negatively correlated to lung volume. After multivariate adjustment, stroke volume as well as end-diastolic volume from LV (ß=-2.75, p=0.001; ß=-1.71, p=0.001) and RV (ß=-2.14, p=0.02; ß=-1.45, p=0.004) were negatively associated with lung volume. Myocardial mass, peak ejection rate and ejection fraction were not associated with lung volumes. For LV, the early diastolic filling rate was negatively associated with lung volume (ß=-17.3, p=0.006; ß=-11, p=0.08).
Conclusion: Using whole-body MRI, we observed an association between lung volume and biventricular end-diastolic and stroke volume, as well as LV filling rates, indicating potential for early detection of subclinical cardiac impairment.
Purpose: To determine the capability of multi-parametric approach among chemical exchange saturation transfer (CEST) imaging, diffusion-weighted imaging (DWI), and FDG-PET/CT for diagnosis of solitary pulmonary nodules (SPNs).
Methods and Materials: 113 consecutive patients with 122 SPNs underwent CEST imaging and DWI at a 3T MR system, FDG-PET/CT, and pathological and/or follow-up examinations. According to final diagnoses, all SPNs were divided into malignant (n=76) and benign (n=46) SPNs. In each patient, magnetization transfer ratio asymmetry (MTRasym) was calculated from z-spectra at 3.5ppm in each pixel, and MTRasym map was computationally generated from CEST data. In each lesion, MTRasym, apparent diffusion coefficient (ADC) and SUVmax were assessed by ROI measurements. To compare all indexes between two groups, Student’s t test was performed. Then ROC analyses were performed to compare diagnostic performance among all indexes as well as combined methods. Finally, sensitivity, specificity and accuracy were compared among all methods by McNemar’s.
Results: Each index had significant difference between malignant and benign SPNs (p<0.05). ROC analyses showed area under the curve (Az) of combined method (Az=0.92) was significantly larger than that of SUVmax (Az=0.77, p<0.05). Sensitivity (SE) and accuracy (AC) of combined methods (SE: 85.5 [65/76]%, AC: 85.2 [104/122]%) were significantly higher than those of SUVmax (SE: 64.5 [49/76]%, p<0.05; AC: 71.3 [87/122]%, p<0.05).
Conclusion: Multiparametric approach among CEST, DWI and PET/CT has a significantly better potential for diagnosis of SPNs than PET/CT.
Purpose: To evaluate the feasibility of machine learning (ML) for detection of dose optimisation potential in quality assurance of chest CT in a retrospective, cross-sectional study.
Methods and Materials: 3,199 chest CTs were used for training and testing of the neural network (01/2016-12/2017, 61% male, 62±15 years, 80/20 split). The model was optimised and trained to predict the volumetric computed tomography dose index (CTDIvol) on scan and patient metrics (scanner, study description, protocol, patient age, sex and water-equivalent diameter (DW)). The root mean-squared error (RMSE) was calculated as performance measurement. A validation dataset of 100 separate, consecutive examinations (01/2018, 60% male, 63±16 years) was independently reviewed by two blinded radiologists for dose optimisation potential and used to define an optimal cutoff.
Results: RMSE was 1.71, 1.45 and 1.52 for the training, test and validation datasets, respectively. CT scanner and DW were the most important features. The radiologists found dose optimisation potential in 7/100 of the validation cases. A percentage deviation of 18.3% between predicted and actual CTDIvol was found to be the optimal cutoff. 8/100 cases were flagged by the model (range 18.3-53.2%), including all cases flagged by the radiologists. In the false-positive case, the model predicted a 33% lower CTDIvol, which was due to a severe soft tissue emphysema leading to a very small DW.
Conclusion: ML can comprehensively detect chest CT examinations with dose optimisation potential; however, a finale human review is mandatory. A deviation of >=18.3% between the predicted and actual CTDIvol seems adequate for CT quality assurance.
Purpose: To validate the performance of deep convolutional neural networks optimised for the detection of pulmonary embolism (PE) on CT pulmonary angiograms (CTPAs).
Methods and Materials: We downloaded all CTPAs performed in 2017 along with the corresponding reports (n = 1,499) from our RIS/PACS archive using an in-house-developed search engine. The reports were manually reviewed by a radiologist. CTPAs with other clinical questions than PE or poor diagnostic quality were excluded. The remaining exams were then classified into positive (n = 232) and negative (n = 1,204) for PE. All emboli in positive exams were labeled by a radiologist using bounding boxes. The data served as ground truth for the validation of prototype algorithms (Aidoc, Tel Aviv, Israel) that had previously been trained on 28.000 independent CTPAs from other centres.
Results: Four trained prototype algorithms were tested on our CTPA dataset. The best performing algorithm was a fully convolutional neural network with a backbone based on the Resnet architecture. It achieved a sensitivity of 93% and a specificity of 95%. This corresponds to a positive predictive value of 77%.
Conclusion: The best-performing AI algorithm we validated is capable of detecting pulmonary embolism in CTPAs with a high sensitivity and specificity. In a clinical setting, this can complement conventional workflows with a worklist prioritisation and has the potential to improve the quality of healthcare by accelerating the diagnostic process and communication. We plan to further test the algorithm and finally implement it in the clinical routine to perform prospective evaluations.