Deep Learning to Classify AL versus ATTR Cardiac Amyloidosis MR Images.

The aim of this work was to compare the classification of cardiac MR-images of AL versus ATTR amyloidosis by neural networks and by experienced human readers. Cine-MR images and late gadolinium enhancement (LGE) images of 120 patients were studied (70 AL and 50 TTR). A VGG16 convolutional neural network (CNN) was trained with a 5-fold cross validation process, taking care to strictly distribute images of a given patient in either the training group or the test group. The analysis was performed at the patient level by averaging the predictions obtained for each image. The classification accuracy obtained between AL and ATTR amyloidosis was 0.750 for cine-CNN, 0.611 for Gado-CNN and between 0.617 and 0.675 for human readers. The corresponding AUC of the ROC curve was 0.839 for cine-CNN, 0.679 for gado-CNN (p < 0.004 vs. cine) and 0.714 for the best human reader (p < 0.007 vs. cine). Logistic regression with cine-CNN and gado-CNN, as well as analysis focused on the specific orientation plane, did not change the overall results. We conclude that cine-CNN leads to significantly better discrimination between AL and ATTR amyloidosis as compared to gado-CNN or human readers, but with lower performance than reported in studies where visual diagnosis is easy, and is currently suboptimal for clinical practice.

Radiomics in the evaluation of lung nodules: Intrapatient concordance between full-dose and ultra-low-dose chest computed tomography.

PURPOSE: The purpose of this study was to retrospectively evaluate the quantitative and qualitative intrapatient concordance of pulmonary nodule risk assessment by commercially available radiomics software between full-dose (FD) chest-CT and ultra-low-dose (ULD) chest CT. MATERIALS AND METHODS: Between July 2013 and September 2015, 68 patients (52 men and16 women; mean age, 65.5±10.6 [SD] years; range: 35-87 years) with lung nodules≥5mm and<30mm who underwent the same day FD chest CT (helical acquisition; 120kV; automated tube current modulation) and ULD chest CT (helical acquisition; 135kV; 10mA fixed) were retrospectively included. Each nodule on each acquisition was assessed by a commercial radiomics software providing a similarity malignancy index (mSI), classifying it as "benign-like" (mSI<0.1); "malignant-like" (mSI>0.9) or "undetermined" (0.1≤mSI≤0.9). Intrapatient qualitative agreement was evaluated with weighted Cohen-Kappa test and quantitative agreement with intraclass correlation coefficient (ICC). RESULTS: Ninety-nine lung nodules with a mean size of 9.14±4.3 (SD) mm (range: 5-25mm) in 68 patients (mean 1.46 nodule per patient; range: 1-5) were assessed; mean mSI was 0.429±0.331 (SD) (range: 0.001-1) with FD chest CT (22/99 [22%] "benign-like", 67/99 [68%] "undetermined" and 10/99 [10%] "malignant-like") and mean mSI was 0.487±0.344 (SD) (range: 0.002-1) with ULD chest CT (20/99 [20%] "benign-like", 59/99 [60%] "undetermined" and 20/99 [20%] "malignant-like"). Qualitative and quantitative agreement of FD chest CT with ULD chest CT were "good" with Kappa value of 0.60 (95% CI: 0.46-0.74) and ICC of 0.82 (95% CI: 0.73-0.87), respectively. CONCLUSION: A good agreement in malignancy similarity index can be obtained between ULD chest CT and FD chest CT using radiomics software. However, further studies must be done with more case material to confirm our results and elucidate the diagnostic capabilities of radiomics software using ULD chest CT for lung nodule characterization by comparison with FD chest CT.

Deep Learning Supplants Visual Analysis by Experienced Operators for the Diagnosis of Cardiac Amyloidosis by Cine-CMR.

BACKGROUND: Diagnosing cardiac amyloidosis (CA) from cine-CMR (cardiac magnetic resonance) alone is not reliable. In this study, we tested if a convolutional neural network (CNN) could outperform the visual diagnosis of experienced operators. METHOD: 119 patients with cardiac amyloidosis and 122 patients with left ventricular hypertrophy (LVH) of other origins were retrospectively selected. Diastolic and systolic cine-CMR images were preprocessed and labeled. A dual-input visual geometry group (VGG ) model was used for binary image classification. All images belonging to the same patient were distributed in the same set. Accuracy and area under the curve (AUC) were calculated per frame and per patient from a 40% held-out test set. Results were compared to a visual analysis assessed by three experienced operators. RESULTS: frame-based comparisons between humans and a CNN provided an accuracy of 0.605 vs. 0.746 (p < 0.0008) and an AUC of 0.630 vs. 0.824 (p < 0.0001). Patient-based comparisons provided an accuracy of 0.660 vs. 0.825 (p < 0.008) and an AUC of 0.727 vs. 0.895 (p < 0.002). CONCLUSION: based on cine-CMR images alone, a CNN is able to discriminate cardiac amyloidosis from LVH of other origins better than experienced human operators (15 to 20 points more in absolute value for accuracy and AUC), demonstrating a unique capability to identify what the eyes cannot see through classical radiological analysis.

Wide-volume versus helical acquisition in unenhanced chest CT: prospective intra-patient comparison of diagnostic accuracy and radiation dose in an ultra-low-dose setting.

OBJECTIVES: Diagnostic performance and potential radiation dose reduction of wide-area detector CT sequential acquisition ("wide-volume" acquisition (WV)) in unenhanced chest examination are unknown. This study aims to assess the image quality, the diagnostic performance, and the radiation dose reduction of WV mode compared with the classical helical acquisition for lung parenchyma analysis in an ultra-low-dose (ULD) protocol. METHODS: After Institutional Review Board Approval and written informed consent, 64 patients (72% men; 67.6 ± 9.7 years old; BMI 26.1 ± 5.3 kg/m2) referred for a clinically indicated unenhanced chest CT were prospectively included. All patients underwent, in addition to a standard helical acquisition (120 kV, automatic tube current modulation), two ULD acquisitions (135 kV, fixed tube current at 10 mA): one in helical mode and one in WV mode. Image noise, subjective image quality (5-level Likert scale), and diagnostic performance for the detection of 9 predetermined parenchymal abnormalities were assessed by two radiologists and compared using the chi-square or Fisher non-parametric tests. RESULTS: Subjective image quality (4.2 ± 0.7 versus 4.2 ± 0.8, p = 0.56), image noise (41.7 ± 8 versus 40.9 ± 8.7, p = 0.3), and diagnostic performance were equivalent between ULD WV and ULD helical. Radiation dose was significantly lower for the ULD WV acquisition (mean dose-length product 14.1 ± 1.3 mGy cm versus 15.8 ± 1.3, p < 0.0001). CONCLUSION: An additional 11% dose reduction is achieved with the WV mode in ULD chest CT with fixed tube current, with equivalent image quality and diagnostic performance when compared with the helical acquisition. KEY POINTS: • Image quality and diagnostic performance of ultra-low-dose unenhanced chest CT are identical between wide-volume mode and the reference helical acquisition. • Wide-volume mode allows an additional radiation dose reduction of 11% (mean dose-length product 14.1 ± 1.3 mGy cm versus 15.8 ± 1.3, p < 0.0001).

Prospective evaluation of ultra-low-dose contrast-enhanced 100-kV abdominal computed tomography with tin filter: effect on radiation dose reduction and image quality with a third-generation dual-source CT system.

OBJECTIVES: To investigate the radiation dose exposure, image quality, and diagnostic performance of enhanced 100-kVp abdominopelvic single-energy CT protocol with tin filter (TF). METHODS: Ninety-three consecutive patients referred for a single-phase enhanced abdominopelvic CT were prospectively included after informed consent. They underwent in addition to a standard protocol (SP) an acquisition with TF. Both examinations were performed on a third-generation dual-source CT system (DSCT), in single energy, using automatic tube current modulation, identical pitch, and identical level of iterative reconstruction. Radiation metrics were compared. Size-specific dose estimates (SSDE), contrast to noise ratio (CNR), and figure of merit (FOM) were calculated. Diagnostic confidence for the assessment of a predetermined list of abdominal lesions was rated by two independent readers. RESULTS: The mean dose of the TF protocol was significantly lower (CDTI 1.56 ± 0.43 mGy vs. 8.13 ± 3.32, p < 0.001; SSDE 9.94 ± 3.08 vs. 1.93 ± 0.39, p < 0.001), with an effective dose close to 1 mSv (1.14 mSv ± 0.34; p < 0.001). TF group exhibited non-significant lower liver CNR (2.76 vs. 3.03, p = 0.56) and was more dose efficient (FOM 10.6 vs. 2.49/mSv, p < 0.001) in comparison to SP. The mean diagnostic confidence for visceral, bone, and peritoneal tumors was equivalent between both groups. CONCLUSIONS: Enhanced 100-kVp abdominopelvic CT acquired after spectral shaping with tin filtration can achieve similar diagnostic performance and CNR compared to a standard CT protocol, while reducing the radiation dose by 81%. KEY POINTS: • 100-kVp spectral filtration enables enhanced abdominal CT with high-dose efficiency. • The radiation dose reaches the 1-mSv range. • Predetermined abdominopelvic lesions can be assessed without impairing on diagnostic confidence.

320-row CT transcatheter aortic valve replacement planning with a single reduced contrast media bolus injection.

OBJECTIVE: To reduce the iodine load required for CT Transcatheter Aortic Valve Replacement (TAVR) planning on a 320-row scanner by acquiring the two CT TAVR steps (ECG-gated aortic root CTA and non-gated aorto-ilio-femoral CTA) within a single contrast media bolus injection. METHODS: 50 consecutive patients (82.6±6.9 years; 56% female) were prospectively enrolled and underwent a TAVR planning using a 320-row CT, with ECG-gated aortic root CTA immediately followed by a non-gated aorto-iliac acquisition, all within a single bolus of 40-70mL of Iohexol 350mgI/mL. The Iodine load, image quality, SNR, CNR and radiation dose were compared using a Mann-Whitney test to that of 24 consecutive patients (84.3±4.8 years, 58% female) previously imaged on a 64-row scanner with a conventional two-step protocol. RESULTS: Iodine load was reduced by 44%. All examinations were of diagnostic quality, with improvement of the aortic root CTA image quality (4.9±0.3 versus 4.6±0.5, p<0.01) and a non-significant decrease of the aorto-iliac CTA image quality (4.7±0.6 versus 4.9±0.3, p = 0.07). SNR and CNR were significantly improved in the aortic root CTA (14.0±5.3 and 10.4±4.5 versus 10.3±4.2 and 6.8±3.3, p<0.01 for both) and non-significantly higher in the aorto-iliac CTA (16.5±8.0 and 14.1±7.9 versus 14.7±5.5 and 12.5±5.0, p = 0.42 and p = 0.66). Total radiation dose was reduced by 32%. CONCLUSION: 320-row CT scanner enables a 44% reduction of iodine load in TAVR planning, while maintaining excellent aorto-ilio-femoral arterial enhancement and lowering radiation dose.

Computed Tomography Perfusion Measurements in Renal Lesions Obtained by Bayesian Estimation, Advanced Singular-Value Decomposition Deconvolution, Maximum Slope, and Patlak Models: Intermodel Agreement and Diagnostic Accuracy of Tumor Classification.

OBJECTIVES: The aims of this study were to evaluate the agreement of computed tomography (CT)-perfusion parameter values of the normal renal cortex and various renal tumors, which were obtained by different mathematical models, and to evaluate their diagnostic accuracy. MATERIALS AND METHODS: Perfusion imaging was performed prospectively in 35 patients to analyze 144 regions of interest of the normal renal cortex and 144 regions of interest of renal tumors, including 21 clear-cell renal cell carcinomas (RCC), 6 papillary RCCs, 5 oncocytomas, 1 chromophobe RCC, 1 angiomyolipoma with minimal fat, and 1 tubulocystic RCC. Identical source data were postprocessed and analyzed on 2 commercial software applications with the following implemented mathematical models: maximum slope, Patlak plot, standard singular-value decomposition (SVD), block-circulant SVD, oscillation-limited block-circulant SVD, and Bayesian estimation technique. Results for blood flow (BF), blood volume (BV), and mean transit time (MTT) were recorded. Agreement and correlation between pairs of models and perfusion parameters were assessed. Diagnostic accuracy was evaluated by receiver operating characteristic (ROC) analysis. RESULTS: Significant differences and poor agreement of BF, BV, and MTT values were noted for most of model comparisons in both the normal renal cortex and different renal tumors. The correlations between most model pairs and perfusion parameters ranged between good and perfect (Spearman ρ = 0.79-1.00), except for BV values obtained by Patlak method (ρ = 0.61-0.72). All mathematical models computed BF and BV values, which differed significantly between clear cell RCCs, papillary RCCs, and oncocytomas, which introduces them as useful diagnostic tests to differentiate between different histologic subgroups (areas under ROC curve, 0.83-0.99). The diagnostic accuracy to discriminate between clear-cell RCCs and the renal cortex was the lowest based on the Patlak plot model (area under ROC curve, 0.76); BF and BV values obtained by other algorithms did not differ significantly in their diagnostic accuracy. CONCLUSIONS: Quantitative perfusion parameters obtained from different mathematical models cannot be used interchangeably. Based on BF and BV estimates, all models are a useful tool in the differential diagnosis of kidney tumors, with the Patlak plot model yielding a significantly lower diagnostic accuracy.

Magnetic resonance evaluation of cardiac thrombi and masses by T1 and T2 mapping: an observational study.

The purpose of this work was to evaluate CMR T1 and T2 mapping sequences in patients with intracardiac thrombi and masses in order to assess T1 and T2 relaxometry usefulness and to allow better etiological diagnosis. This observational study of patients scheduled for routine CMR was performed from September 2014 to August 2015. All patients referred to our department for a 1.5 T CMR were screened to participate. T1 mapping were acquired before and after Gadolinium injection; T2 mapping images were obtained before injection. 41 patients were included. 22 presented with cardiac thrombi and 19 with cardiac masses. The native T1 of thrombi was 1037 ± 152 ms (vs 1032 ± 39 ms for myocardium, p = 0.88; vs 1565 ± 88 ms for blood pool, p < 0.0001). T2 were 74 ± 13 ms (vs 51 ± 3 ms for myocardium, p < 0.0001; vs 170 ± 32 ms for blood pool, p < 0.0001). Recent thrombi had a native T1 shorter than old thrombi (911 ± 177 vs 1169 ± 107 ms, p = 0.01). The masses having a shorter T1 than the myocardium were lipomas (278 ± 29 ms), calcifications (621 ± 218 ms), and melanoma (736 ms). All other masses showed T1 values higher than myocardial T1, with T2 consistently >70 ms. T1 and T2 mapping CMR sequences can be useful and represent a new approach for the evaluation of cardiac thrombi and masses.

Prostate cancer diagnosis: Efficacy of a simple electromagnetic MRI-TRUS fusion method to target biopsies.

OBJECTIVE: To assess that transrectal ultrasound guidance (TRUS) targeted biopsies (TB) aimed with an easy to use electronic real-time fusion registration device have a higher rate of prostate cancer (PC) detection than standard biopsies (SB). MATERIAL AND METHODS: This prospective study included 130 patients referred for TRUS biopsies after suspicious MRI. They underwent 16-core SB and 2 to 3 cores in each MRI suspicious area, using a fusion software. We noted SB and TB positivity for PC and Gleason score (GS). We used the McNemar test to compare SB and TB, with a statistical significance p<0.05. RESULTS: Among 130 patients, 68.5% had PC. Additional time due to the fusion registration was 3.3 min. One hundred fifteen patients (88.4%) had pathological findings on the histological analysis (prostate cancer n=89, others n=26). TB diagnosed PC in 75 patients with negative SB. Positivity rate for PC was significantly higher for TB than SB (p=0.03). Among highly suspicious MRI lesions, detection rate of histological abnormalities using SB and TB was 96% with 79.7% of PC. Most PC that TB diagnosed alone were clinically significant (86.3%). CONCLUSION: TRUS biopsies performed with a simple MRI and US electronic fusion is an unrestrainedly method to increase PC diagnosis.

Diagnostic Performance of Ultra-Low-Dose Computed Tomography for Detecting Asbestos-Related Pleuropulmonary Diseases: Prospective Study in a Screening Setting.

OBJECTIVE: To evaluate the diagnostic performance of Ultra-Low-Dose Chest CT (ULD CT) for the detection of any asbestos-related lesions (primary endpoint) and specific asbestos-related abnormalities, i.e. non-calcified and calcified pleural plaques, diffuse pleural thickening, asbestosis and significant lung nodules (secondary endpoints). MATERIAL AND METHODS: 55 male patients (55.7±8.1 years old) with occupational asbestos exposure for at least 15 years and where CT screening was indicated were prospectively included. They all underwent a standard unenhanced chest CT (120kV, automated tube current modulation), considered as the reference, and an ULD CT (135kV, 10mA), both with iterative reconstruction. Two chest radiologists independently and blindly read the examinations, following a detailed protocol. Sensitivity, specificity, positive predictive value, negative predictive value, accuracy and error rate of ULD CT were calculated using the exact method of Pearson with a confidence interval of 95%. RESULTS: Radiation dose was 17.9±1.2mGy.cm (0.25mSv) for the ULD-CT versus 288.8 ±151mGy.cm (4mSv); p <2.2e-16. Prevalence of abnormalities was 20%. The ULD CT's diagnostic performance in joint reading was high for the primary endpoint (sensitivity = 90.9%, specificity = 100%, positive predictive value = 100%, negative predictive value = 97.8%), high for lung nodules, diffuse pleural thickening and calcified pleural plaques (sensitivity, specificity, PPV and NPV = 100%) and fair for asbestosis (sensitivity = 75%, specificity = 100%, PPV = 00%, NPV = 98.1%). Intra-reader accuracy between the ULD CT and the reference CT for the primary endpoint was 98% for the senior and 100% for the junior radiologist. Inter-reader agreement for the primary endpoint was almost perfect (Cohen's Kappa of 0.81). CONCLUSION: ULD CT in the screening of asbestos exposure related diseases has 90.9% sensitivity and 100% specificity, and could therefore be proposed as a first line examination.

DWI in the Etiologic Diagnosis of Excretory Upper Urinary Tract Lesions: Can It Help in Differentiating Benign From Malignant Tumors? A Retrospective Study of 98 Patients.

OBJECTIVE: The objective of this study was to evaluate the diagnostic performance of high-field DWI in distinguishing benign from malignant lesions of the upper urinary tract (UUT). MATERIALS AND METHODS: Ninety-eight patients who underwent 3-T DWI (b = 1000 s/mm(2)) for assessment of a UUT lesion were retrospectively included in the study. Data on the size, location, and mean apparent diffusion coefficient (ADC) value of the lesions were collected and correlated with the final diagnosis. ROC curve analysis was used to determine the best threshold value of the mean ADC for characterization of the lesions. RESULTS: A total of 66 carcinomas and 33 benign lesions were identified. The mean (± SD) ADC value was statistically significantly lower for the malignant lesions than for the benign lesions (0.984 ± 0.048 vs 1.536 ± 0.067 × 10(-3) mm(2)/s; p = 0.000007). For characterization of malignant lesions, the best cutoff ADC value on the ROC curve was found to be less than or equal to 0.996 10(-3) mm(2)/s. On the basis of this value, the sensitivity and specificity of 3-T DWI for the identification of malignant UUT lesions were 78.3% and 95.5%, respectively, with a positive predictive value of 94.7% and a negative predictive value of 80.8% (p = 0.0001). The thickness of the smallest lesion was 3 mm. The mean ADC value of high-grade tumors was not statistically significantly lower than that of low-grade tumors. The best cutoff value for predicting benignancy was greater than or equal to 1.100 × 10(-3) mm(2)/s with sensitivity of 90.9%, specificity of 82.6%, positive predictive value of 83.3%, and negative predictive value of 90.5%. CONCLUSION: The ADC measurement is a useful additional parameter in the differentiation of benign from malignant UUT lesions.

[Ultra-low dose chest CT: The end of chest radiograph?]. / Scanner thoracique ultra-basse dose : la mort de la radiographie thoracique ?

Ultra-low dose chest CT (ULD-CT) is acquired at a radiation dose lowered to that of a PA and lateral chest X-ray. Its image quality is degraded, yet remains diagnostic in many clinical indications. Technological improvements, with iterative reconstruction at the foreground, allowed a strong increase in the image quality obtained with this examination, which is achievable on most recent (<5 years) scanner. Established clinical indications of ULD-CT are increasing, and its non-inferiority compared to the reference "full dose" chest CT are currently demonstrated for the detection of solid nodules, for asbestos-related pleural diseases screening and for the monitoring of infectious pneumonia. Its current limitations are the obese patients (BMI>35) and the interstitial pneumonia, situations in which their performances are insufficient.