•Proposes methods for modelling different types of uncertainty that arise in deep learning (DL) applications for image enhancement problems.•Demonstrates in dMRI super-resolution tasks that modelling ...uncertainty enhances the safety of DL-based enhancement system by bringing two categories of practical benefits:(1) “performance improvement”: e.g., the generalisation to out-of-distribution data, robustness to noise and outliers (Section 4.3)(2) “reliability assessment of prediction”: e.g., certification of performance based on uncertainty-thresholding (Section 4.4.1); detection of unfamiliar structures and understanding the sources of uncertainty (Section 4.4.2).•Provide a comprehensive set of experiments in a diverse set of datasets, which vary in demographics, scanner types, acquisition protocols or pathology.•The methods are in theory applicable to many other imaging modalities and data enhancement applications.•Codes will be available on Github.
Deep learning (DL) has shown great potential in medical image enhancement problems, such as super-resolution or image synthesis. However, to date, most existing approaches are based on deterministic models, neglecting the presence of different sources of uncertainty in such problems. Here we introduce methods to characterise different components of uncertainty, and demonstrate the ideas using diffusion MRI super-resolution. Specifically, we propose to account for intrinsic uncertainty through a heteroscedastic noise model and for parameter uncertainty through approximate Bayesian inference, and integrate the two to quantify predictive uncertainty over the output image. Moreover, we introduce a method to propagate the predictive uncertainty on a multi-channelled image to derived scalar parameters, and separately quantify the effects of intrinsic and parameter uncertainty therein. The methods are evaluated for super-resolution of two different signal representations of diffusion MR images—Diffusion Tensor images and Mean Apparent Propagator MRI—and their derived quantities such as mean diffusivity and fractional anisotropy, on multiple datasets of both healthy and pathological human brains. Results highlight three key potential benefits of modelling uncertainty for improving the safety of DL-based image enhancement systems. Firstly, modelling uncertainty improves the predictive performance even when test data departs from training data (“out-of-distribution” datasets). Secondly, the predictive uncertainty highly correlates with reconstruction errors, and is therefore capable of detecting predictive “failures”. Results on both healthy subjects and patients with brain glioma or multiple sclerosis demonstrate that such an uncertainty measure enables subject-specific and voxel-wise risk assessment of the super-resolved images that can be accounted for in subsequent analysis. Thirdly, we show that the method for decomposing predictive uncertainty into its independent sources provides high-level “explanations” for the model performance by separately quantifying how much uncertainty arises from the inherent difficulty of the task or the limited training examples. The introduced concepts of uncertainty modelling extend naturally to many other imaging modalities and data enhancement applications.
Proton MRS (1H MRS) provides noninvasive, quantitative metabolite profiles of tissue and has been shown to aid the clinical management of several brain diseases. Although most modern clinical MR ...scanners support MRS capabilities, routine use is largely restricted to specialized centers with good access to MR research support. Widespread adoption has been slow for several reasons, and technical challenges toward obtaining reliable good‐quality results have been identified as a contributing factor. Considerable progress has been made by the research community to address many of these challenges, and in this paper a consensus is presented on deficiencies in widely available MRS methodology and validated improvements that are currently in routine use at several clinical research institutions. In particular, the localization error for the PRESS localization sequence was found to be unacceptably high at 3 T, and use of the semi‐adiabatic localization by adiabatic selective refocusing sequence is a recommended solution. Incorporation of simulated metabolite basis sets into analysis routines is recommended for reliably capturing the full spectral detail available from short TE acquisitions. In addition, the importance of achieving a highly homogenous static magnetic field (B0) in the acquisition region is emphasized, and the limitations of current methods and hardware are discussed. Most recommendations require only software improvements, greatly enhancing the capabilities of clinical MRS on existing hardware. Implementation of these recommendations should strengthen current clinical applications and advance progress toward developing and validating new MRS biomarkers for clinical use.
Sporadic Creutzfeldt–Jakob disease (sCJD) is a transmissible brain proteinopathy. Five main clinicopathological subtypes (sCJD-MM(V)1, -MM(V)2C, -MV2K, -VV1, and -VV2) are currently distinguished. ...Histopathological evidence suggests that the localisation of prion aggregates and spongiform lesions varies among subtypes. Establishing whether there is an initial site with detectable imaging abnormalities (epicentre) and an order of lesion propagation would be informative for disease early diagnosis, patient staging, management and recruitment in clinical trials. Diffusion magnetic resonance imaging (MRI) is the most-used and most-sensitive test to detect spongiform degeneration. This study was designed to identify, in vivo and for the first time, subtype-dependent epicentre and lesion propagation in the brain using diffusion-weighted images (DWI), in the largest known cross-sectional dataset of autopsy-proven subjects with sCJD. We estimate lesion propagation by cross-sectional DWI using event-based modelling, a well-established data-driven technique. DWI abnormalities of 594 autopsy-diagnosed subjects (448 patients with sCJD) were scored in 12 brain regions by 1 neuroradiologist blind to the diagnosis. We used the event-based model to reconstruct sequential orderings of lesion propagation in each of five pure subtypes. Follow-up data from 151 patients validated the estimated sequences. Results showed that epicentre and ordering of lesion propagation are subtype specific. The two most common subtypes (-MM1 and -VV2) showed opposite ordering of DWI abnormality appearance: from the neocortex to subcortical regions, and vice versa, respectively. The precuneus was the most likely epicentre also in -MM2 and -VV1 although at variance with -MM1, abnormal signal was also detected early in cingulate and insular cortices. The caudal-rostral sequence of lesion propagation that characterises -VV2 was replicated in -MV2K. Combined, these data-driven models provide unprecedented dynamic insights into subtype-specific epicentre at onset and propagation of the pathologic process, which may also enhance early diagnosis and enable disease staging in sCJD.
Magnetic resonance spectroscopic imaging (MRSI) offers considerable promise for monitoring metabolic alterations associated with disease or injury; however, to date, these methods have not had a ...significant impact on clinical care, and their use remains largely confined to the research community and a limited number of clinical sites. The MRSI methods currently implemented on clinical MRI instruments have remained essentially unchanged for two decades, with only incremental improvements in sequence implementation. During this time, a number of technological developments have taken place that have already greatly benefited the quality of MRSI measurements within the research community and which promise to bring advanced MRSI studies to the point where the technique becomes a true imaging modality, while making the traditional review of individual spectra a secondary requirement. Furthermore, the increasing use of biomedical MR spectroscopy studies has indicated clinical areas where advanced MRSI methods can provide valuable information for clinical care. In light of this rapidly changing technological environment and growing understanding of the value of MRSI studies for biomedical studies, this article presents a consensus from a group of experts in the field that reviews the state‐of‐the‐art for clinical proton MRSI studies of the human brain, recommends minimal standards for further development of vendor‐provided MRSI implementations, and identifies areas which need further technical development.
MR spectroscopic imaging enables noninvasive mapping of metabolites within the body and offers considerable potential for clinical and biomedical research studies. Recent technological advances, in high‐field MRI instrumentation, spatial‐spectral sampling and image reconstruction methods, and data analysis, have greatly improved spatial resolution and the extent of the brain over which metabolites can be mapped. This consensus statement summarizes the state‐of‐the‐art for clinical MRSI studies of the brain and provides a set of implementation standards matched to different clinical applications.
Summary Human prion diseases can be sporadic, inherited, or acquired by infection. Distinct clinical and pathological characteristics separate sporadic diseases into three phenotypes: ...Creutzfeldt-Jakob disease (CJD), fatal insomnia, and variably protease-sensitive prionopathy. CJD accounts for more than 90% of all cases of sporadic prion disease; it is commonly categorised into five subtypes that can be distinguished according to leading clinical signs, histological lesions, and molecular traits of the pathogenic prion protein. Three subtypes affect prominently cognitive functions whereas the other two impair cerebellar motor activities. An accurate and timely diagnosis depends on careful clinical examination and early performance and interpretation of diagnostic tests, including electroencephalography, quantitative assessment of the surrogate markers 14-3-3, tau, and of the prion protein in the CSF, and neuroimaging. The reliability of CSF tests is improved when these tests are interpreted alongside neuroimaging data.
A large body of published work shows that proton (hydrogen 1 (1)H) magnetic resonance (MR) spectroscopy has evolved from a research tool into a clinical neuroimaging modality. Herein, the authors ...present a summary of brain disorders in which MR spectroscopy has an impact on patient management, together with a critical consideration of common data acquisition and processing procedures. The article documents the impact of (1)H MR spectroscopy in the clinical evaluation of disorders of the central nervous system. The clinical usefulness of (1)H MR spectroscopy has been established for brain neoplasms, neonatal and pediatric disorders (hypoxia-ischemia, inherited metabolic diseases, and traumatic brain injury), demyelinating disorders, and infectious brain lesions. The growing list of disorders for which (1)H MR spectroscopy may contribute to patient management extends to neurodegenerative diseases, epilepsy, and stroke. To facilitate expanded clinical acceptance and standardization of MR spectroscopy methodology, guidelines are provided for data acquisition and analysis, quality assessment, and interpretation. Finally, the authors offer recommendations to expedite the use of robust MR spectroscopy methodology in the clinical setting, including incorporation of technical advances on clinical units.
Purpose To evaluate the feasibility of a standardized protocol for acquisition and analysis of dynamic contrast material-enhanced (DCE) and dynamic susceptibility contrast (DSC) magnetic resonance ...(MR) imaging in a multicenter clinical setting and to verify its accuracy in predicting glioma grade according to the new World Health Organization 2016 classification. Materials and Methods The local research ethics committees of all centers approved the study, and informed consent was obtained from patients. One hundred patients with glioma were prospectively examined at 3.0 T in seven centers that performed the same preoperative MR imaging protocol, including DCE and DSC sequences. Two independent readers identified the perfusion hotspots on maps of volume transfer constant (K
), plasma (v
) and extravascular-extracellular space (v
) volumes, initial area under the concentration curve, and relative cerebral blood volume (rCBV). Differences in parameters between grades and molecular subtypes were assessed by using Kruskal-Wallis and Mann-Whitney U tests. Diagnostic accuracy was evaluated by using receiver operating characteristic curve analysis. Results The whole protocol was tolerated in all patients. Perfusion maps were successfully obtained in 94 patients. An excellent interreader reproducibility of DSC- and DCE-derived measures was found. Among DCE-derived parameters, v
and v
had the highest accuracy (are under the receiver operating characteristic curve A
= 0.847 and 0.853) for glioma grading. DSC-derived rCBV had the highest accuracy (A
= 0.894), but the difference was not statistically significant (P > .05). Among lower-grade gliomas, a moderate increase in both v
and rCBV was evident in isocitrate dehydrogenase wild-type tumors, although this was not significant (P > .05). Conclusion A standardized multicenter acquisition and analysis protocol of DCE and DSC MR imaging is feasible and highly reproducible. Both techniques showed a comparable, high diagnostic accuracy for grading gliomas.
RSNA, 2018 Online supplemental material is available for this article.
To determine the values of iron accumulation in the basal ganglia of healthy volunteers of different ages with R2* and raw signal intensity measurements from T1-weighted magnetic resonance (MR) ...images, supported by voxel-based relaxometry (VBR), and to compare them with previously reported iron concentrations found in autopsy material.
The ethics committee approved the study, and the participants or their parents gave written informed consent. Eighty subjects (41 female and 39 male subjects; age range, 1-80 years) were examined at 1.5 T. For each subject, R2* values were calculated. Curves for R2* versus age were obtained for globus pallidus (GP), putamen, caudate nucleus, substantia nigra (SN), and frontal white matter (FWM). To highlight possible differences in iron concentration among the age decades, VBR was applied. Signal intensity values were estimated on T1-weighted fast low-angle shot images, and regions of interest were drawn in each nucleus. R2* values were also compared with iron concentrations reported in a postmortem study. Statistical analysis was performed (t test), and a difference with P < .05 (FDR corrected) was significant.
The curves for R2* versus age showed an exponential increase with increasing age in all the basal ganglia. VBR demonstrated significant differences (P < .05, corrected) in the comparison between the 2nd and the following decades for lenticular nuclei. Good correlation coefficients were found for GP (R(2) = 0.64), putamen (R(2) = 0.51), and SN (R(2) = 0.53) when compared with findings in the postmortem study. Signal intensity curves were similar to the R2* curves.
R2* measurements can be used to quantify brain iron accumulation and thus may allow better evaluation of neurodegenerative diseases associated with iron deposition.
To compare the respective efficiency of CSF tau (quantitative) and CSF 14-3-3 protein (qualitative) in the diagnosis of prion disease.
We made measurements on 420 live subjects, who subsequently ...underwent a postmortem neuropathology examination, including protein chemistry, immunohistochemistry, and histology. We performed tau by ELISA. We detected 14-3-3 protein by Western blot. Both assays were optimized for maximum efficiency (accuracy).
We found tau and 14-3-3 proteins to be closely correlated, but tau had a significantly better ability to predict disease status than 14-3-3 protein. Also, tau distinguished disease status at least as well as when both assays' results are combined in a variety of ways. Importantly, the area under the receiver operating characteristic curve for tau (0.82) was significantly larger than that for 14-3-3 protein (0.68) (p < 0.001). Diagnostic test statistics are provided for the study subjects with 58.3% prevalence, and for a more typical, nonselected, 7.5% prevalence as received by our center.
In this study, tau is superior to 14-3-3 protein as a marker in the diagnosis of Creutzfeldt-Jakob disease, and is as efficient singly compared to a variety of combinations with 14-3-3 protein. This is the first study of this magnitude to examine prion disease diagnostic tests in a carefully characterized patient population with detailed statistical evaluation.