TIST-Net: style transfer in dynamic contrast enhanced MRI using spatial and temporal information.

Objective.Training deep learning models for image registration or segmentation of dynamic contrast enhanced (DCE) MRI data is challenging. This is mainly due to the wide variations in contrast enhancement within and between patients. To train a model effectively, a large dataset is needed, but acquiring it is expensive and time consuming. Instead, style transfer can be used to generate new images from existing images. In this study, our objective is to develop a style transfer method that incorporates spatio-temporal information to either add or remove contrast enhancement from an existing image.Approach.We propose a temporal image-to-image style transfer network (TIST-Net), consisting of an auto-encoder combined with convolutional long short-term memory networks. This enables disentanglement of the content and style latent spaces of the time series data, using spatio-temporal information to learn and predict key structures. To generate new images, we use deformable and adaptive convolutions which allow fine grained control over the combination of the content and style latent spaces. We evaluate our method, using popular metrics and a previously proposed contrast weighted structural similarity index measure. We also perform a clinical evaluation, where experts are asked to rank images generated by multiple methods.Main Results.Our model achieves state-of-the-art performance on three datasets (kidney, prostate and uterus) achieving an SSIM of 0.91 ± 0.03, 0.73 ± 0.04, 0.88 ± 0.04 respectively when performing style transfer between a non-enhanced image and a contrast-enhanced image. Similarly, SSIM results for style transfer from a contrast-enhanced image to a non-enhanced image were 0.89 ± 0.03, 0.82 ± 0.03, 0.87 ± 0.03. In the clinical evaluation, our method was ranked consistently higher than other approaches.Significance.TIST-Net can be used to generate new DCE-MRI data from existing images. In future, this may improve models for tasks such as image registration or segmentation by allowing small training datasets to be expanded.

Multiparametric magnetic resonance imaging for quantitation of liver disease: a two-centre cross-sectional observational study.

LiverMultiScan is an emerging diagnostic tool using multiparametric MRI to quantify liver disease. In a two-centre prospective validation study, 161 consecutive adult patients who had clinically-indicated liver biopsies underwent contemporaneous non-contrast multiparametric MRI at 3.0 tesla (proton density fat fraction (PDFF), T1 and T2* mapping), transient elastography (TE) and Enhanced Liver Fibrosis (ELF) test. Non-invasive liver tests were correlated with gold standard histothological measures. Reproducibility of LiverMultiScan was investigated in 22 healthy volunteers. Iron-corrected T1 (cT1), TE, and ELF demonstrated a positive correlation with hepatic collagen proportionate area (all p < 0·001). TE was superior to ELF and cT1 for predicting fibrosis stage. cT1 maintained good predictive accuracy for diagnosing significant fibrosis in cases with indeterminate ELF, but not for cases with indeterminate TE values. PDFF had high predictive accuracy for individual steatosis grades, with AUROCs ranging from 0.90-0.94. T2* mapping diagnosed iron accumulation with AUROC of 0.79 (95% CI: 0.67-0.92) and negative predictive value of 96%. LiverMultiScan showed excellent test/re-test reliability (coefficients of variation ranging from 1.4% to 2.8% for cT1). Overall failure rates for LiverMultiScan, ELF and TE were 4.3%, 1.9% and 15%, respectively. LiverMultiScan is an emerging point-of-care diagnostic tool that is comparable with the established non-invasive tests for assessment of liver fibrosis, whilst at the same time offering a superior technical success rate and contemporaneous measurement of liver steatosis and iron accumulation.

Vascular and plaque imaging with ultrasmall superparamagnetic particles of iron oxide.

Cardiovascular Magnetic Resonance (CMR) has become a primary tool for non-invasive assessment of cardiovascular anatomy, pathology and function. Existing contrast agents have been utilised for the identification of infarction, fibrosis, perfusion deficits and for angiography. Novel ultrasmall superparamagnetic particles of iron oxide (USPIO) contrast agents that are taken up by inflammatory cells can detect cellular inflammation non-invasively using CMR, potentially aiding the diagnosis of inflammatory medical conditions, guiding their treatment and giving insight into their pathophysiology. In this review we describe the utilization of USPIO as a novel contrast agent in vascular disease.

Ultrasmall superparamagnetic particles of iron oxide in patients with acute myocardial infarction: early clinical experience.

BACKGROUND: Inflammation following acute myocardial infarction (MI) has detrimental effects on reperfusion, myocardial remodelling, and ventricular function. Magnetic resonance imaging using ultrasmall superparamagnetic particles of iron oxide can detect cellular inflammation in tissues, and we therefore explored their role in acute MI in humans. METHODS AND RESULTS: Sixteen patients with acute ST-segment elevation MI were recruited to undergo 3 sequential magnetic resonance scans within 5 days of admission at baseline, 24 and 48 hours following no infusion (controls; n=6) or intravenous infusion of ultrasmall superparamagnetic particles of iron oxide (n=10; 4 mg/kg). T2*-weighted multigradient-echo sequences were acquired and R2* values were calculated for specific regions of interest. In the control group, R2* values remained constant in all tissues across all scans with excellent repeatability (bias of -0.208 s(-1), coefficient of repeatability of 26.96 s(-1); intraclass coefficient 0.989). Consistent with uptake by the reticuloendothelial system, R2* value increased in the liver (84±49.5 to 319±70.0 s(-1); P<0.001) but was unchanged in skeletal muscle (54±8.4 to 67.0±9.5 s(-1); P>0.05) 24 hours after administration of ultrasmall superparamagnetic particles of iron oxide. In the myocardial infarct, R2* value increased from 41.0±12.0 s(-1) (baseline) to 155±45.0 s(-1) (P<0.001) and 124±35.0 s(-1) (P<0.05) at 24 and 48 hours, respectively. A similar but lower magnitude response was seen in the remote myocardium, where it increased from 39±3.2 s(-1) (baseline) to 80±14.9 s(-1) (P<0.001) and 67.0±15.7 s(-1) (P<0.05) at 24 and 48 hours, respectively. CONCLUSIONS: Following acute MI, uptake of ultrasmall superparamagnetic particles of iron oxide occurs with the infarcted and remote myocardium. This technique holds major promise as a potential method for assessing cellular myocardial inflammation and left ventricular remodelling, which may have a range of applications in patients with MI and other inflammatory cardiac conditions.

In vivo mononuclear cell tracking using superparamagnetic particles of iron oxide: feasibility and safety in humans.

BACKGROUND: Cell therapy is an emerging and exciting novel treatment option for cardiovascular disease that relies on the delivery of functional cells to their target site. Monitoring and tracking cells to ensure tissue delivery and engraftment is a critical step in establishing clinical and therapeutic efficacy. The study aims were (1) to develop a Good Manufacturing Practice-compliant method of labeling competent peripheral blood mononuclear cells with superparamagnetic particles of iron oxide (SPIO), and (2) to evaluate its potential for magnetic resonance cell tracking in humans. METHODS AND RESULTS: Peripheral blood mononuclear cells 1-5 × 10(9) were labeled with SPIO. SPIO-labeled cells had similar in vitro viability, migratory capacity, and pattern of cytokine release to unlabeled cells. After intramuscular administration, up to 10(8) SPIO-labeled cells were readily identifiable in vivo for at least 7 days using magnetic resonance imaging scanning. Using a phased-dosing study, we demonstrated that systemic delivery of up to 10(9) SPIO-labeled cells in humans is safe, and cells accumulating in the reticuloendothelial system were detectable on clinical magnetic resonance imaging. In a healthy volunteer model, a focus of cutaneous inflammation was induced in the thigh by intradermal injection of tuberculin. Intravenously delivered SPIO-labeled cells tracked to the inflamed skin and were detectable on magnetic resonance imaging. Prussian blue staining of skin biopsies confirmed iron-laden cells in the inflamed skin. CONCLUSIONS: Human peripheral blood mononuclear cells can be labeled with SPIO without affecting their viability or function. SPIO labeling for magnetic resonance cell tracking is a safe and feasible technique that has major potential for a range of cardiovascular applications including monitoring of cell therapies and tracking of inflammatory cells. Clinical Trial Registration- URL: http://www.clinicaltrials.gov; Unique identifier: NCT00972946, NCT01169935.

Quantification techniques to minimize the effects of native T1 variation and B1 inhomogeneity in dynamic contrast-enhanced MRI of the breast at 3 T.

The variation of the native T(1) (T(10)) of different tissues and B(1) transmission-field inhomogeneity at 3 T are major contributors of errors in the quantification of breast dynamic contrast-enhanced MRI. To address these issues, we have introduced new enhancement indices derived from saturation-recovery snapshot-FLASH (SRSF) images. The stability of the new indices, i.e., the SRSF enhancement factor (EF(SRSF)) and its simplified version (EF'(SRSF)) with respect to differences in T(10) and B(1) inhomogeneity was compared against a typical index used in breast dynamic contrast-enhanced MRI, i.e., the enhancement ratio (ER), by using computer simulations. Imaging experiments with Gd-DTPA-doped gel phantoms and a female volunteer were also performed. A lower error was observed in the new indices compared to enhancement ratio in the presence of typical T(10) variation and B(1) inhomogeneity. At changes of relaxation rate (ΔR(1)) of 8 s(-1), the differences between a T(10) of 1266 and 566 ms are <1, 12, and 58%, respectively, for EF(SRSF), EF'(SRSF), and ER, whereas differences of 20, 8, and 51%, respectively, result from a 50% B(1) field reduction at the same ΔR(1). These quantification techniques may be a solution to minimize the effect of T(10) variation and B(1) inhomogeneity on dynamic contrast-enhanced MRI of the breast at 3 T.

B1 transmission-field inhomogeneity and enhancement ratio errors in dynamic contrast-enhanced MRI (DCE-MRI) of the breast at 3T.

PURPOSE: To quantify B(1) transmission-field inhomogeneity in breast imaging of normal volunteers at 3T using 3D T(1)-weighted spoiled gradient echo and to assess the resulting errors in enhancement ratio (ER) measured in dynamic contrast-enhanced MRI (DCE-MRI) studies of the breast. MATERIALS AND METHODS: A total of 25 volunteers underwent breast imaging at 3T and the B(1) transmission-fields were mapped. Gel phantoms that simulate pre- and postcontrast breast tissue T(1) were developed. The effects of B(1)-field inhomogeneity on ER, as measured using a 3D spoiled gradient echo sequence, were investigated by computer simulation and experiments on gel phantoms. RESULTS: It was observed that by using the patient orientation and MR scanner employed in this study, the B(1) transmission-field field is always reduced toward the volunteer's right side. The median B(1)-field in the right breast is reduced around 40% of the expected B(1)-field. For some volunteers the amplitude was reduced by more than 50%. Computer simulation and experiment showed that a reduction in B(1)-field decreases ER. This reduction increases with both B(1)-field error and contrast agent uptake. CONCLUSION: B(1) transmission-field inhomogeneity is a critical issue in breast imaging at 3T and causes errors in quantifying ER. These errors would be sufficient to reduce the conspicuity of a malignant lesion and could result in reduced sensitivity for cancer detection.

A combined pharmacokinetic and radiologic assessment of dynamic contrast-enhanced magnetic resonance imaging predicts response to chemoradiation in locally advanced cervical cancer.

PURPOSE: To investigate the combination of pharmacokinetic and radiologic assessment of dynamic contrast-enhanced magnetic resonance imaging (MRI) as an early response indicator in women receiving chemoradiation for advanced cervical cancer. METHODS AND MATERIALS: Twenty women with locally advanced cervical cancer were included in a prospective cohort study. Dynamic contrast-enhanced MRI was carried out before chemoradiation, after 2 weeks of therapy, and at the conclusion of therapy using a 1.5-T MRI scanner. Radiologic assessment of uptake parameters was obtained from resultant intensity curves. Pharmacokinetic analysis using a multicompartment model was also performed. General linear modeling was used to combine radiologic and pharmacokinetic parameters and correlated with eventual response as determined by change in MRI tumor size and conventional clinical response. A subgroup of 11 women underwent repeat pretherapy MRI to test pharmacokinetic reproducibility. RESULTS: Pretherapy radiologic parameters and pharmacokinetic K(trans) correlated with response (p < 0.01). General linear modeling demonstrated that a combination of radiologic and pharmacokinetic assessments before therapy was able to predict more than 88% of variance of response. Reproducibility of pharmacokinetic modeling was confirmed. CONCLUSIONS: A combination of radiologic assessment with pharmacokinetic modeling applied to dynamic MRI before the start of chemoradiation improves the predictive power of either by more than 20%. The potential improvements in therapy response prediction using this type of combined analysis of dynamic contrast-enhanced MRI may aid in the development of more individualized, effective therapy regimens for this patient group.

Monitoring primary breast cancer throughout chemotherapy using FDG-PET.

UNLABELLED: We have compared 2-deoxy-2-[(18)F]-fluoro-D-glucose positron emission tomography (FDG-PET) images of large or locally advanced breast cancers (LABC) acquired during Anthracycline-based chemotherapy. The purpose was to determine whether there is an optimal method for defining tumour volume and an optimal imaging time for predicting pathologic chemotherapy response. METHOD: PET data were acquired before the first and second cycles, at the midpoint and at the endpoint of neoadjuvant chemotherapy. FDG uptake was quantified using the mean and maximum standardized uptake values (SUV) and the coefficient of variation within a region of interest. Receiver-operator characteristic (ROC) analysis was used to determine the discrimination between tumours demonstrating a high pathological response (i.e. those with greater than 90% reduction in viable tumour cells) and low pathological response. RESULTS: Only tumours with an initial tumour to background ratio (TBR) of greater than five showed a difference between response categories. In terms of response discrimination, there was no statistically significant advantage of any of the methods used for image quantification or any of the time points. The best discrimination was measured for mean SUV at the midpoint of therapy, which identified 77% of low responding tumours whilst correctly identifying 100% of high responding tumours and had an ROC area of 0.93. CONCLUSION: FDG-PET is efficacious for predicting the pathologic response of most primary breast tumours throughout the duration of a neoadjuvant chemotherapy regimen. However, this technique is ineffective for tumours with low image contrast on pre-therapy PET scans.

The relationship between vascular and metabolic characteristics of primary breast tumours.

The objective of this study was to investigate the relationship between vascular and metabolic characteristics of breast tumours in vivo, using contrast-enhanced dynamic MRI and 2-[(18)F] fluoro-2-deoxy- d-glucose (FDG) PET imaging. Twenty patients with large or locally advanced primary breast cancers were imaged prior to therapy. MRI data were acquired using a dynamic gradient echo sequence and analysed using two pharmacokinetic models. Static PET data were acquired in 2D mode. A significant association ( P<0.05) was observed between the calculated exchange rate constants of both pharmacokinetic models and calculated PET FDG dose uptake ratios (DUR). Statistical analysis showed that the exchange rate constants can explain between 27 and 44% of the variance observed in the PET FDG uptake ratios. A relationship was demonstrated between the vascular and metabolic characteristics of primary breast tumours showing that any assessment of tumour metabolic activity using PET may be controlled at least in part by delivery of uptake agent due to the vascular characteristics of the tumour. MRI and PET provide methods of assessing breast tumour vascularity and metabolism in vivo using the exchange rate constants of dynamic MRI, and DUR of PET, respectively, these measures being related but not equivalent.