Validating the accuracy of deep learning for the diagnosis of pneumonia on chest x-ray against a robust multimodal reference diagnosis: a post hoc analysis of two prospective studies.

BACKGROUND: Artificial intelligence (AI) seems promising in diagnosing pneumonia on chest x-rays (CXR), but deep learning (DL) algorithms have primarily been compared with radiologists, whose diagnosis can be not completely accurate. Therefore, we evaluated the accuracy of DL in diagnosing pneumonia on CXR using a more robust reference diagnosis. METHODS: We trained a DL convolutional neural network model to diagnose pneumonia and evaluated its accuracy in two prospective pneumonia cohorts including 430 patients, for whom the reference diagnosis was determined a posteriori by a multidisciplinary expert panel using multimodal data. The performance of the DL model was compared with that of senior radiologists and emergency physicians reviewing CXRs and that of radiologists reviewing computed tomography (CT) performed concomitantly. RESULTS: Radiologists and DL showed a similar accuracy on CXR for both cohorts (p ≥ 0.269): cohort 1, radiologist 1 75.5% (95% confidence interval 69.1-80.9), radiologist 2 71.0% (64.4-76.8), DL 71.0% (64.4-76.8); cohort 2, radiologist 70.9% (64.7-76.4), DL 72.6% (66.5-78.0). The accuracy of radiologists and DL was significantly higher (p ≤ 0.022) than that of emergency physicians (cohort 1 64.0% [57.1-70.3], cohort 2 63.0% [55.6-69.0]). Accuracy was significantly higher for CT (cohort 1 79.0% [72.8-84.1], cohort 2 89.6% [84.9-92.9]) than for CXR readers including radiologists, clinicians, and DL (all p-values < 0.001). CONCLUSIONS: When compared with a robust reference diagnosis, the performance of AI models to identify pneumonia on CXRs was inferior than previously reported but similar to that of radiologists and better than that of emergency physicians. RELEVANCE STATEMENT: The clinical relevance of AI models for pneumonia diagnosis may have been overestimated. AI models should be benchmarked against robust reference multimodal diagnosis to avoid overestimating its performance. TRIAL REGISTRATION: NCT02467192 , and NCT01574066 . KEY POINT: • We evaluated an openly-access convolutional neural network (CNN) model to diagnose pneumonia on CXRs. • CNN was validated against a strong multimodal reference diagnosis. • In our study, the CNN performance (area under the receiver operating characteristics curve 0.74) was lower than that previously reported when validated against radiologists' diagnosis (0.99 in a recent meta-analysis). • The CNN performance was significantly higher than emergency physicians' (p ≤ 0.022) and comparable to that of board-certified radiologists (p ≥ 0.269).

Mir-21 Suppression Promotes Mouse Hepatocarcinogenesis.

The microRNA 21 (miR-21) is upregulated in almost all known human cancers and is considered a highly potent oncogene and potential therapeutic target for cancer treatment. In the liver, miR-21 was reported to promote hepatic steatosis and inflammation, but whether miR-21 also drives hepatocarcinogenesis remains poorly investigated in vivo. Here we show using both carcinogen (Diethylnitrosamine, DEN) or genetically (PTEN deficiency)-induced mouse models of hepatocellular carcinoma (HCC), total or hepatocyte-specific genetic deletion of this microRNA fosters HCC development-contrasting the expected oncogenic role of miR-21. Gene and protein expression analyses of mouse liver tissues further indicate that total or hepatocyte-specific miR-21 deficiency is associated with an increased expression of oncogenes such as Cdc25a, subtle deregulations of the MAPK, HiPPO, and STAT3 signaling pathways, as well as alterations of the inflammatory/immune anti-tumoral responses in the liver. Together, our data show that miR-21 deficiency promotes a pro-tumoral microenvironment, which over time fosters HCC development via pleiotropic and complex mechanisms. These results question the current dogma of miR-21 being a potent oncomiR in the liver and call for cautiousness when considering miR-21 inhibition for therapeutic purposes in HCC.

Pelvic MRI for Endometriosis: A Diagnostic Challenge for the Inexperienced Radiologist. How Much Experience Is Enough?

PURPOSE: The purpose of this study was to investigate the impact of radiologist experience on diagnostic performance of pelvic magnetic resonance imaging (MRI) for the evaluation of endometriomas and different localisations of deep pelvic endometriosis (DPE). MATERIALS AND METHODS: In this prospective study all pelvic MRI examinations performed for pelvic endometriosis from December 2016 to August 2017 were evaluated by readers with different experience levels; junior resident (0-6 weeks of experience in female imaging), senior resident (7-24 weeks), fellow (6-24 months), and expert (10 years) in female imaging for the presence of endometriomas and DPE. Their evaluations were compared with surgery confirmed with pathology. Diagnostic performances of readers with different levels of experience were studied by the means of receiving operating characteristic curves and areas under the curve (AUC) were compared with the ones of the expert reader. RESULTS: Of 174 patients evaluated, the standard of reference was available for 59, consisting the final population of the study. The AUC for endometriomas, DPE for the posterior and anterior pelvic compartment, for rectosigmoid DPE and for overall evaluation were 0.983, 0.921, 0.615, 0.862, and 0.914 for the expert reader, 0.966 (p = 0.178), 0.805 (p = 0.001), 0.605 (p = 0.91), 0.872 (p = 0.317), and 0.849 (p = 0.0009) for the fellow level, 0.877 (p = 0.002), 0.757 (p < 0.001), 0.585 (p = 0.761), 0.744 (p = 0.239), and 0.787 (p = < 0.001) for the senior resident level and 0.861 (p = 0.177), 0.649 (p < 0.001), 0.648 (p = 0.774), 0.862 (p = 1), and 0.721 (p < 0.001) for the junior resident level. CONCLUSIONS: According to our results, interpretation of pelvic MRI for DPE should be performed by specialists as; even the performance of radiologists with up to 2 years of experience in female imaging was statistically inferior to that of experts.

Genetic Ablation of MiR-22 Fosters Diet-Induced Obesity and NAFLD Development.

miR-22 is one of the most abundant miRNAs in the liver and alterations of its hepatic expression have been associated with the development of hepatic steatosis and insulin resistance, as well as cancer. However, the pathophysiological roles of miR-22-3p in the deregulated hepatic metabolism with obesity and cancer remains poorly characterized. Herein, we observed that alterations of hepatic miR-22-3p expression with non-alcoholic fatty liver disease (NAFLD) in the context of obesity are not consistent in various human cohorts and animal models in contrast to the well-characterized miR-22-3p downregulation observed in hepatic cancers. To unravel the role of miR-22 in obesity-associated NAFLD, we generated constitutive Mir22 knockout (miR-22KO) mice, which were subsequently rendered obese by feeding with fat-enriched diet. Functional NAFLD- and obesity-associated metabolic parameters were then analyzed. Insights about the role of miR-22 in NAFLD associated with obesity were further obtained through an unbiased proteomic analysis of miR-22KO livers from obese mice. Metabolic processes governed by miR-22 were finally investigated in hepatic transformed cancer cells. Deletion of Mir22 was asymptomatic when mice were bred under standard conditions, except for an onset of glucose intolerance. However, when challenged with a high fat-containing diet, Mir22 deficiency dramatically exacerbated fat mass gain, hepatomegaly, and liver steatosis in mice. Analyses of explanted white adipose tissue revealed increased lipid synthesis, whereas mass spectrometry analysis of the liver proteome indicated that Mir22 deletion promotes hepatic upregulation of key enzymes in glycolysis and lipid uptake. Surprisingly, expression of miR-22-3p in Huh7 hepatic cancer cells triggers, in contrast to our in vivo observations, a clear induction of a Warburg effect with an increased glycolysis and an inhibited mitochondrial respiration. Together, our study indicates that miR-22-3p is a master regulator of the lipid and glucose metabolism with differential effects in specific organs and in transformed hepatic cancer cells, as compared to non-tumoral tissue.

Local rotation invariance in 3D CNNs.

Locally Rotation Invariant (LRI) image analysis was shown to be fundamental in many applications and in particular in medical imaging where local structures of tissues occur at arbitrary rotations. LRI constituted the cornerstone of several breakthroughs in texture analysis, including Local Binary Patterns (LBP), Maximum Response 8 (MR8) and steerable filterbanks. Whereas globally rotation invariant Convolutional Neural Networks (CNN) were recently proposed, LRI was very little investigated in the context of deep learning. LRI designs allow learning filters accounting for all orientations, which enables a drastic reduction of trainable parameters and training data when compared to standard 3D CNNs. In this paper, we propose and compare several methods to obtain LRI CNNs with directional sensitivity. Two methods use orientation channels (responses to rotated kernels), either by explicitly rotating the kernels or using steerable filters. These orientation channels constitute a locally rotation equivariant representation of the data. Local pooling across orientations yields LRI image analysis. Steerable filters are used to achieve a fine and efficient sampling of 3D rotations as well as a reduction of trainable parameters and operations, thanks to a parametric representations involving solid Spherical Harmonics (SH),which are products of SH with associated learned radial profiles. Finally, we investigate a third strategy to obtain LRI based on rotational invariants calculated from responses to a learned set of solid SHs. The proposed methods are evaluated and compared to standard CNNs on 3D datasets including synthetic textured volumes composed of rotated patterns, and pulmonary nodule classification in CT. The results show the importance of LRI image analysis while resulting in a drastic reduction of trainable parameters, outperforming standard 3D CNNs trained with rotational data augmentation.

Clot-Based Radiomics Predict a Mechanical Thrombectomy Strategy for Successful Recanalization in Acute Ischemic Stroke.

BACKGROUND AND PURPOSE: Mechanical thrombectomy (MTB) is a reference treatment for acute ischemic stroke, with several endovascular strategies currently available. However, no quantitative methods are available for the selection of the best endovascular strategy or to predict the difficulty of clot removal. We aimed to investigate the predictive value of an endovascular strategy based on radiomic features extracted from the clot on preinterventional, noncontrast computed tomography to identify patients with first-attempt recanalization with thromboaspiration and to predict the overall number of passages needed with an MTB device for successful recanalization. METHODS: We performed a study including 2 cohorts of patients admitted to our hospital: a retrospective training cohort (n=109) and a prospective validation cohort (n=47). Thrombi were segmented on noncontrast computed tomography, followed by the automatic computation of 1485 thrombus-related radiomic features. After selection of the relevant features, 2 machine learning models were developed on the training cohort to predict (1) first-attempt recanalization with thromboaspiration and (2) the overall number of passages with MTB devices for successful recanalization. The performance of the models was evaluated on the prospective validation cohort. RESULTS: A small subset of radiomic features (n=9) was predictive of first-attempt recanalization with thromboaspiration (receiver operating characteristic curve-area under the curve, 0.88). The same subset also predicted the overall number of passages required for successful recanalization (explained variance, 0.70; mean squared error, 0.76; Pearson correlation coefficient, 0.73; P<0.05). CONCLUSIONS: Clot-based radiomics have the ability to predict an MTB strategy for successful recanalization in acute ischemic stroke, thus allowing a potentially better selection of the MTB strategy, as well as patients who are most likely to benefit from the intervention.

Radiomics and Machine Learning Differentiate Soft-Tissue Lipoma and Liposarcoma Better than Musculoskeletal Radiologists.

Distinguishing lipoma from liposarcoma is challenging on conventional MRI examination. In case of uncertain diagnosis following MRI, further invasive procedure (percutaneous biopsy or surgery) is often required to allow for diagnosis based on histopathological examination. Radiomics and machine learning allow for several types of pathologies encountered on radiological images to be automatically and reliably distinguished. The aim of the study was to assess the contribution of radiomics and machine learning in the differentiation between soft-tissue lipoma and liposarcoma on preoperative MRI and to assess the diagnostic accuracy of a machine-learning model compared to musculoskeletal radiologists. 86 radiomics features were retrospectively extracted from volume-of-interest on T1-weighted spin-echo 1.5 and 3.0 Tesla MRI of 38 soft-tissue tumors (24 lipomas and 14 liposarcomas, based on histopathological diagnosis). These radiomics features were then used to train a machine-learning classifier to distinguish lipoma and liposarcoma. The generalization performance of the machine-learning model was assessed using Monte-Carlo cross-validation and receiver operating characteristic curve analysis (ROC-AUC). Finally, the performance of the machine-learning model was compared to the accuracy of three specialized musculoskeletal radiologists using the McNemar test. Machine-learning classifier accurately distinguished lipoma and liposarcoma, with a ROC-AUC of 0.926. Notably, it performed better than the three specialized musculoskeletal radiologists reviewing the same patients, who achieved ROC-AUC of 0.685, 0.805, and 0.785. Despite being developed on few cases, the trained machine-learning classifier accurately distinguishes lipoma and liposarcoma on preoperative MRI, with better performance than specialized musculoskeletal radiologists.

Generative Adversarial Networks Improve the Reproducibility and Discriminative Power of Radiomic Features.

PURPOSE: To assess the contribution of a generative adversarial network (GAN) to improve intermanufacturer reproducibility of radiomic features (RFs). MATERIALS AND METHODS: The authors retrospectively developed a cycle-GAN to translate texture information from chest radiographs acquired using one manufacturer (Siemens) to chest radiographs acquired using another (Philips), producing fake chest radiographs with different textures. The authors prospectively evaluated the ability of this texture-translation cycle-GAN to reduce the intermanufacturer variability of RFs extracted from the lung parenchyma. This study assessed the cycle-GAN's ability to fool several machine learning (ML) classifiers tasked with recognizing the manufacturer on the basis of chest radiography inputs. The authors also evaluated the cycle-GAN's ability to mislead radiologists who were asked to perform the same recognition task. Finally, the authors tested whether the cycle-GAN had an impact on radiomic diagnostic accuracy for chest radiography in patients with congestive heart failure (CHF). RESULTS: RFs, extracted from chest radiographs after the cycle-GAN's texture translation (fake chest radiographs), showed decreased intermanufacturer RF variability. Using cycle-GAN-generated chest radiographs as inputs, ML classifiers categorized the fake chest radiographs as belonging to the target manufacturer rather than to a native one. Moreover, cycle-GAN fooled two experienced radiologists who identified fake chest radiographs as belonging to a target manufacturer class. Finally, reducing intermanufacturer RF variability with cycle-GAN improved the discriminative power of RFs for patients without CHF versus patients with CHF (from 55% to 73.5%, P < .001). CONCLUSION: Both ML classifiers and radiologists had difficulty recognizing the chest radiographs' manufacturer. The cycle-GAN improved RF intermanufacturer reproducibility and discriminative power for identifying patients with CHF. This deep learning approach may help counteract the sensitivity of RFs to differences in acquisition.Supplemental material is available for this article.© RSNA, 2020See also the commentary by Alderson in this issue.

Accuracy of cone-beam computed tomography for syndesmosis injury diagnosis compared to conventional computed tomography.

BACKGROUND: Syndesmosis injury can lead to ankle mortise instability and early osteoarthritis. Several multiple detector computed tomography (MDCT) methods for measurement have been developed. Weight-bearing cone beam CT (WB CBCT) is an emerging technique that offers the possibility of upright-position scanning and lower doses. This study sought to assess the diagnostic accuracy of WB CBCT in syndesmose injury compared to MDCT, with instability confirmed via manual testing upon arthroscopic examination. METHODS: Three musculoskeletal radiologists with different levels of expertise prospectively analyzed 11 MDCT and eight WB CBCT scans of the same trauma-afflicted ankles with clinical suspicion of syndesmosis lesion over a period of 5 months. They evaluated 10 methods of measurement in both sides. Syndesmosis was considered pathological on arthroscopic examination in four patients. Correlation between readers was evaluated with intra-class correlation testing (p < 0.05 was considered significant). Capacity of discrimination was assessed by area under the curve (AUC) for all methods. RESULTS: Inter-observer agreement was near excellent for both WB CBCT and MDCT for the anterior tibio-fibular (TF) distance (ICC = 0.781 and 0.831, respectively), posterior TF distance (ICC = 0.841 and 0.826), minimal TF distance (ICC = 0.899 and 0.875), and TF surface (ICC = 0.93 and 0.84). AUC were better for MDCT than WB CBCT in assessing syndesmosis instability for: anterior TF distance (ROC = 0.869 vs. 0.555, p = 0.01), minimal TF distance (ROC = 0.883 vs. 0.608, p = 0.02) and antero-posterior fibular translation (ROC = 0.894 vs. 0.467, p = 0.006). CONCLUSIONS: MDCT demonstrated better ability to distinguish pathological syndesmosis than WB CBCT, with the antero-posterior fibular translation the best discriminating measurement. The physiological widening of the contralateral syndesmosis occurring with the WB CBCT upright position may explain these results.

Imaging patterns of Pneumocystis jirovecii pneumonia in HIV-positive and renal transplant patients - a multicentre study.

OBJECTIVES: To investigate differences in chest computed tomography (CT) and chest radiographs (CXRs) of Pneumocystis jirovecii pneumonia (PJP) between renal transplant recipients (RTRs) and human immunodeficiency virus (HIV)-positive patients. METHODS: From 2005 to 2012, 84 patients with PJP (RTR n = 24; HIV n = 60) were included in this retrospective multicentre study. Written informed consent was obtained. CT scans and CXRs were recorded within 2 weeks after the onset of symptoms. PJP diagnosis was confirmed either by cytology/histology or successful empirical treatment. Two blinded radiologists analysed the conventional chest films and CT images, and recorded the radiological lung parenchyma patterns, lymph node enlargement and pleural pathologies (pneumothorax, effusion). The radiological features of the two subgroups were compared. RESULTS: Consolidations and solid nodules prevailed on CT in RTRs (91.7 ± 5.6% vs 58.3 ± 6.4% with HIV, p = 0.019 and 91.7 ± 5.6% vs 51.6 ± 6.5% with HIV, p = 0.005). HIV-positive patients with PJP showed more atelectasis (41.7 ± 6.4% vs 4.2 ± 4.1% in RTRs, p = 0.017) and hilar lymph node enlargement (23.3 ± 5.5% vs 0.0 ± 0.0% in RTRs, p = 0.088). Ground glass opacification was found in all cases. Pneumothorax was a rare complication, occurring in 3% of the HIV-positive patients; no pneumothorax was found in the RTRs. On CXR, the basal lungs were more affected in HIV-positive patients as compared with RTRs (p = 0.024). CONCLUSIONS: PJP on CT differs substantially between RTRs and HIV-positive patients. Physicians should be aware of such differences in order not to delay treatment, particularly in renal transplant recipients.

Mild hyperthermia by MR-guided focused ultrasound in an ex vivo model of osteolytic bone tumour: optimization of the spatio-temporal control of the delivered temperature.

BACKGROUND: Magnetic resonance guided focused ultrasound was suggested for the induction of deep localized hyperthermia adjuvant to radiation- or chemotherapy. In this study we are aiming to validate an experimental model for the induction of uniform temperature elevation in osteolytic bone tumours, using the natural acoustic window provided by the cortical breakthrough. MATERIALS AND METHODS: Experiments were conducted on ex vivo lamb shank by mimicking osteolytic bone tumours. The cortical breakthrough was exploited to induce hyperthermia inside the medullar cavity by delivering acoustic energy from a phased array HIFU transducer. MR thermometry data was acquired intra-operatory using the proton resonance frequency shift (PRFS) method. Active temperature control was achieved via a closed-loop predictive controller set at 6 °C above the baseline. Several beam geometries with respect to the cortical breakthrough were investigated. Numerical simulations were used to further explain the observed phenomena. Thermal safety of bone heating was assessed by cross-correlating MR thermometry data with the measurements from a fluoroptic temperature sensor inserted in the cortical bone. RESULTS: Numerical simulations and MR thermometry confirmed the feasibility of spatio-temporal uniform hyperthermia (± 0.5 °C) inside the medullar cavity using a fixed focal point sonication. This result was obtained by the combination of several factors: an optimal positioning of the focal spot in the plane of the cortical breakthrough, the direct absorption of the HIFU beam at the focal spot, the "acoustic oven effect" yielded by the beam interaction with the bone, and a predictive temperature controller. The fluoroptical sensor data revealed no heating risks for the bone and adjacent tissues and were in good agreement with the PRFS thermometry from measurable voxels adjacent to the periosteum. CONCLUSION: To our knowledge, this is the first study demonstrating the feasibility of MR-guided focused ultrasound hyperthermia inside the medullar cavity of bones affected by osteolytic tumours. Our results are considered a promising step for combining adjuvant mild hyperthermia to external beam radiation therapy for sustained pain relief in patients with symptomatic bone metastases.

Treatment with direct-acting antivirals improves peripheral insulin sensitivity in non-diabetic, lean chronic hepatitis C patients.

BACKGROUND AND AIMS: Hepatitis C virus (HCV) infection is associated with insulin resistance, which may lead to type 2 diabetes and its complications. Although HCV infects mainly hepatocytes, it may impair insulin sensitivity at the level of uninfected extrahepatic tissues (muscles and adipose tissue). The aim of this study was to assess whether an interferon-free, antiviral therapy may improve HCV-associated hepatic vs. peripheral insulin sensitivity. METHODS: In a single-arm exploratory trial, 17 non-diabetic, lean chronic hepatitis C patients without significant fibrosis were enrolled, and 12 completed the study. Patients were treated with a combination of sofosbuvir/ledipasvir and ribavirin for 12 weeks, and were submitted to a 2-step euglycemic hyperinsulinemic clamp with tracers, together with indirect calorimetry measurement, to measure insulin sensitivity before and after 6 weeks of antivirals. A panel of 27 metabolically active cytokines was analyzed at baseline and after therapy-induced viral suppression. RESULTS: Clamp analysis performed in 12 patients who achieved complete viral suppression after 6 weeks of therapy showed a significant improvement of the peripheral insulin sensitivity (13.1% [4.6-36.7], p = 0.003), whereas no difference was observed neither in the endogenous glucose production, in lipolysis suppression nor in substrate oxidation. A distinct subset of hepatokines, potentially involved in liver-to-periphery crosstalk, was modified by the antiviral therapy. CONCLUSION: Pharmacological inhibition of HCV improves peripheral (but not hepatic) insulin sensitivity in non-diabetic, lean individuals with chronic hepatitis C without significant fibrosis.

Rational Use of CT-Scan for the Diagnosis of Pneumonia: Comparative Accuracy of Different Strategies.

Diagnosing pneumonia in emergency departments is challenging because the accuracy of symptoms, signs and laboratory tests is limited. As a confirmation test, chest X-ray has significant limitations and is outperformed by CT-scan. However, obtaining a CT-scan in all cases of suspected pneumonia has significant drawbacks. We used a cohort of 200 consecutive elderly patients admitted to the hospital for suspected pneumonia to build a simple prediction score, which was used to determine indication for performing a CT-scan. The reference diagnosis was adjudicated by experts considering all available data, including evolution until discharge and CT scan in all patients. Results were externally validated in a second cohort of 319 patients. Pneumonia was confirmed in 133 patients (67%). Area under the receiver operator curve (AUROC) of physician evaluation was 0.55 (0.46-0.64). The score incorporated four variables independently predicting confirmed pneumonia: male gender, acute cough, C-reactive protein >70 mg/L, and urea <7 mmol/L. AUROC of the score was 0.68 (95% confidence interval (CI) 0.60-0.76). When a CT-scan was obtained for patients at low or intermediate predicted risk (108 patients, 54% of the cohort), AUROC was 0.71 (0.63-0.80) and 0.69 (0.64-0.74) in the derivation and validation cohort, respectively. A simple prediction score for pneumonia had moderate accuracy and could guide the performance of a CT-scan.

Differentiating kidney stones from phleboliths in unenhanced low-dose computed tomography using radiomics and machine learning.

OBJECTIVES: Distinguishing between kidney stones and phleboliths can constitute a diagnostic challenge in patients undergoing unenhanced low-dose CT (LDCT) for acute flank pain. We sought to investigate the accuracy of radiomics and a machine-learning classifier in differentiating between kidney stones and phleboliths on LDCT. METHODS: Radiomics features were extracted following a semi-automatic segmentation of kidney stones and phleboliths for two independent consecutive cohorts of patients undergoing LDCT for acute flank pain. Radiomics features from the first cohort of patients (n = 369) were ultimately used to train a machine-learning model designed to distinguish kidney stones (n = 211) from phleboliths (n = 201). Classification performance was assessed on the second independent cohort (i.e., testing set) (kidney stones n = 24; phleboliths n = 23) using positive and negative predictive values (PPV and NPV), area under the receiver operating curves (AUC), and permutation testing. RESULTS: Our machine-learning classification model trained on radiomics features achieved an overall accuracy of 85.1% on the independent testing set, with an AUC of 0.902, PPV of 81.5%, and NPV of 90.0%. Classification accuracy was significantly better than chance on permutation testing (p < 0.05, permutation p value). CONCLUSION: Radiomics and machine learning enable accurate differentiation between kidney stones and phleboliths on LDCT in patients presenting with acute flank pain. KEY POINTS: • Combining a machine-learning algorithm with radiomics features extracted for abdominopelvic calcification on LDCT offers a highly accurate method for discriminating phleboliths from kidney stones. • Our radiomics and machine-learning model proved robust for CT acquisition and reconstruction protocol when tested in comparison with an external independent cohort of patients with acute flank pain. • The high performance of the radiomics-based automatic classification model in differentiating phleboliths from kidney stones indicates its potential as a future diagnostic tool for equivocal abdominopelvic calcifications in the setting of suspected renal colic.

Emphysema quantification using hybrid versus model-based generations of iterative reconstruction: SAFIRE versus ADMIRE.

To compare 2 incompatible generations of iterative reconstructions from the same raw dataset based on automatic emphysema quantification and noise reduction: a hybrid algorithm called sinogram affirmed iterative reconstruction (SAFIRE) versus a model-based algorithm called advanced modeled iterative reconstruction (ADMIRE).Raw datasets of 40 non-contrast thoracic computed tomography scanners obtained from a single acquisition on a SOMATOM Definition Flash unit (Siemens Healthcare, Forchheim) were reconstructed with 3 levels of SAFIRE and ADMIRE algorithms resulting in a total of 240 datasets. Emphysema index (EI) and image noise were compared using repeated analysis of variance (ANOVA) analysis with a P value <.05 considered statistically significant.EI and image noise were stable between both generations of IR when reconstructed with the same level (P ≥0.31 and P ≥0.06, respectively).SAFIRE and ADMIRE perform equally in terms of emphysema quantification and noise reduction.

Metallic artifact reduction by evaluation of the additional value of iterative reconstruction algorithms in hip prosthesis computed tomography imaging.

To evaluate iterative metal artifact reduction (iMAR) technique in images data of hip prosthesis on computed tomography (CT) and the added value of advanced modeled iterative reconstruction (ADMIRE) compared with standard filtered back projection (FBP).Twenty-eight patients addressed to CT examinations for hip prosthesis were included prospectively. Images were reconstructed with iMAR algorithm in addition to FBP and ADMIRE techniques. Measuring image noise assessed objective image quality and attenuation values with standardized region of interest (ROI) in 4 predefined anatomical structures (gluteus medius and rectus femoris muscles, inferior and anterior abdominal fat, and femoral vessels when contrast media was present). Subjective image quality was graded on a 5-point Likert scale, taking into account the size of artifacts, the metal-bone interface and the conspicuity of pelvic organs, and the diagnostic confidence.Improvement in overall image quality was statistically significant using iMAR (P<.001) compared with ADMIRE and FBP. ADMIRE did not show any impact in image noise, attenuation value, or global quality image. iMAR showed a significant decrease in image noise in all ROIs (Hounsfield Unit) as compared with FBP and ADMIRE. Interobserver agreement was high in all reconstructions (FBP, FBP+iMAR, ADMIRE, and ADMIRE + iMAR) more than 0.8. iMAR reconstructions showed emergence of new artifacts in bone-metal interface.iMAR algorithm allows a significant reduction of metal artifacts on CT images with unilateral or bilateral prostheses without additional value of ADMIRE. It improves the analysis of surrounding tissue but potentially generates new artifacts in bone-metal interface.

Iterative Algorithms Applied to Treated Intracranial Aneurysms.

PURPOSE: To investigate the impact of iterative metal artifact reduction (iMAR) on artifacts related to neurosurgical clips or endovascular coils when combined to filtered back projection (FBP) or advanced modelled iterative reconstruction (ADMIRE). MATERIAL AND METHODS: In this study 21 unenhanced brain computed tomography (CT) examinations were reconstructed with FBP and level 2 of ADMIRE, both techniques with and without iMAR algorithm, resulting in 4 series per acquisition. Subjective assessment of artifact reduction was performed as a double-blinded evaluation with a 5-point-scale. Objective analysis was performed by comparing central tendencies and distributions of voxel densities. The central tendency was assessed as the mean voxel density in Hounsfield units. The distribution was assessed by evaluating the shape and asymmetry of the histograms of voxels densities with measures of kurtosis and skewness, respectively. RESULTS: Inter-reader agreement was excellent (>0.8). FBP and ADMIRE without iMAR were scored 4 and with iMAR 5. Unusual artifacts were noted in all of the series reconstructed with iMAR, especially when combined with ADMIRE. Kurtosis revealed statistical differences for all reconstruction techniques (p ≤ 0.0007) except for the association of FBP with iMAR (p = 0.2211) for the coiling population and skewness demonstrated no statistical difference in any population (p ≥ 0.0558), confirming the subjective analysis results, except for the ADMIRE algorithm with or without iMAR (p ≤ 0.0342) in the coiling population. CONCLUSION: iMAR led to the reduction in artifacts due to intracranial metallic devices. However, it created a new artifact in the form of a halo of photon-starvation, especially when combined with ADMIRE. The combination of FBP and iMAR seems more suitable, combining the beneficial metal artifact reduction without the emergence of a halo of photon starvation just around the point of interest.

Low Iodine Contrast Injection for CT Acquisition Prior to Transcatheter Aortic Valve Replacement: Aorta Assessment and Screening for Coronary Artery Disease.

RATIONALE AND OBJECTIVES: To assess both the complete aorta and coronary artery disease (CAD) using low iodine contrast computed-tomography angiography before transcatheter aortic valve replacement. MATERIALS AND METHODS: 84 patients underwent computed-tomography angiography before transcatheter aortic valve replacement: 42 with standard iodine injection protocol (P1:120 mL); 42 with a low dose iodine injection protocol (P2:60 mL). Mean attenuation and subjective image quality were rated at different levels of the aorta, iliac and coronary arteries. Sensitivity, specificity, negative and positive predictive values for depiction of CAD were calculated according to the coronary angiography. RESULTS: Mean attenuation was significantly higher in P1 for the ascending aorta (p < 0.001). No significant difference was observed regarding image quality of the aortic valve (p = 0.876), the ascending aorta (p = 0.306), or the abdominal aorta (p = 1.0). Diagnostic image quality of coronary arteries was excellent for P1 and P2 (94.6% vs 96.5%, p = 0.08). Sensitivity, specificity, negative and positive predictive values, and accuracy for depiction of CAD were excellent for P1 and P2 (100% vs 100%; 79% vs 86%, 70% vs 87%, 100% vs 100% and 86% vs 93%) without significant differences (p = 0.93; p = 0.58; p = 0.90; p = 1.0; p = 0.74), respectively. CONCLUSION: Despite a difference in aortic mean attenuation, a reduced iodine injection protocol showed similar image quality and detection of CAD in comparison with a standard injection protocol.

What is the diagnostic performance of 18-FDG-PET/MR compared to PET/CT for the N- and M- staging of breast cancer?

PURPOSE: To compare the diagnostic performance of 18-FDG-PET/MR and PET/CT for the N- and M- staging of breast cancer. METHODS AND MATERIALS: Two independent readers blinded to clinical/follow-up data reviewed PET/MR and PET/CT examinations performed for initial or recurrent breast cancer staging in 80 consecutive patients (mean age = 48 ± 12.9 years). The diagnostic confidence for lesions in the contralateral breast, axillary/internal mammary nodes, bones and other distant sites were recorded. Sensitivity, specificity, positive (PPV) and negative predictive values (NPV) were calculated. The standard of reference included pathology and/or follow-up > 12 months. RESULTS: Nine of 80 patients had bone metastases; 13/80 had other distant metastases, 44/80 had axillary, 9/80 had internal mammary and 3/80 had contralateral breast tumours. Inter-reader agreement for lesions was excellent (weighted kappa = 0.833 for PET/CT and 0.823 for PET/MR) with similar reader confidence for the two tests (ICC = 0.875). In the patient-per-patient analysis, sensitivity and specificity of PET/MRI and PET/CT were similar (p > 0.05). In the lesion-per-lesion analysis, the sensitivity of PET/MR and PET/CT for bone metastases, other metastases, axillary and internal mammary nodes, contralateral tumours and all lesions together was 0.924 and 0.6923 (p = 0.0034), 0.923 and 0.923 (p = 1), 0.854 and 0.812 (p = 0.157), 0.9 and 0.9 (p = 1), 1 and 0.25 (p = 0.083), and 0.89 and 0.77 (p = 0.0013) respectively. The corresponding specificity was 0.953 and 1 (p = 0.0081), 1 and 1 (p = 1), 0.893 and 0.92 (p = 0.257), 1 and 1 (p = 1), 0.987 and 0.99 (p = 1) and 0.96 and 0.98 (p = 0.0075) respectively. CONCLUSIONS: Reader confidence, inter-reader agreement and diagnostic performance per patient were similar with PET/MR and PET/CT. However, for all lesions together, PET/MR had a superior sensitivity and lower specificity in the lesion-per-lesion analysis. KEY POINTS: • N and M breast cancer staging performance of PET/MR and PET/CT is similar per patient. • In a lesion-per-lesion analysis PET/MR is more sensitive than PET/CT especially for bone metastasis. • Readers' diagnostic confidence is similar for both tests.

Automatic Annotation Tool to Support Supervised Machine Learning for Scaphoid Fracture Detection.

The aim of this work is to develop and validate an automatic annotation tool for the detection and bone localization of scaphoid fractures in radiology reports. To achieve this goal, a rule-based method using a Natural Language Processing (NLP) tool was applied. Finite state automata were constructed to detect, classify and annotate reports. An evaluation of the method on a manually annotated dataset has shown 96,8% of total match.