Yaniv Gal

Cerebral palsy (CP) describes a group of permanent disorders of posture and movement caused by disturbances in the developing brain. Accurate diagnosis and prognosis, in terms of motor type and severity, is difficult to obtain due to the heterogeneous appearance of brain injury and large anatomical distortions commonly observed in children with CP. There is a need to optimise treatment strategies for individual patients in order to lead to lifelong improvements in function and capabilities. Magnetic resonance imaging (MRI) is critical to non-invasively visualizing brain lesions, and is currently used to assist the diagnosis and qualitative classification in CP patients. Although such qualitative approaches under-utilise available data, the quantification of MRIs is not automated and therefore not widely performed in clinical assessment. Automated brain lesion segmentation techniques are necessary to provide valid and reproducible quantifications of injury. Such techniques have been ...

Accurate knowledge of O(6)-methylguanine methyltransferase (MGMT) gene promoter subtype in patients with glioblastoma (GBM) is important for treatment. However, this test is not always available. Pre-operative diffusion MRI (dMRI) can be used to probe tumour biology using the apparent diffusion coefficient (ADC); however, its ability to act as a surrogate to predict MGMT status has shown mixed results. We investigated whether this was due to variations in the method used to analyse ADC. We undertook a retrospective study of 32 patients with GBM who had MGMT status measured. Matching pre-operative MRI data were used to calculate the ADC within contrast enhancing regions of tumour. The relationship between ADC and MGMT was examined using two published ADC methods. A strong trend between a measure of 'minimum ADC' and methylation status was seen. An elevated minimum ADC was more likely in the methylated compared to the unmethylated MGMT group (U = 56, P = 0.0561). In contrast, utilising a two-mixture model histogram approach, a significant reduction in mean measure of the 'low ADC' component within the histogram was associated with an MGMT promoter methylation subtype (P < 0.0246). This study shows that within the same patient cohort, the method selected to analyse ADC measures has a significant bearing on the use of that metric as a surrogate marker of MGMT status. Thus for dMRI data to be clinically useful, consistent methods of data analysis need to be established prior to establishing any relationship with genetic or epigenetic profiling.

Abstract We present a novel method for the segmentation of enhancing breast tissue, suspicious of malignancy, in dynamic contrast-enhanced (DCE) MR images. The method is based on seeded region growing and merging using criteria based on both the original ...

In this paper, six existing image fusion algorithms, and their application to medical images are studied. Quantitative measurements used in the literature were tested on synthetic data to verify their consistency with qualitative analysis. Each of these algorithms was optimized for quantitative measures which our experiments found consistent with visual qualitative analysis. Finally, algorithms with the optimal parameters were tested on medical data, and corresponding quantitative analysis is provided. Our experiments suggest that, for medical image fusion, the use of Contour let transform achieves the best results. Although, the stationary wavelet transform provided similar quantitative result with substantially less computational complexity.

Despite radical treatment therapies, glioma continues to carry with it a uniformly poor prognosis. Patients diagnosed with WHO Grade IV glioma (glioblastomas; GBM) generally succumb within two years, even those with WHO Grade III anaplastic gliomas and WHO Grade II gliomas carry prognoses of 2-5 and 2years, respectively. PET imaging with (18)F-FDOPA allows in vivo assessment of the metabolism of glioma relative to surrounding tissues. The high sensitivity of (18)F-DOPA imaging grants utility for a number of clinical applications. A collection of published work about (18)F-FDOPA PET was made and a critical review was discussed and written. A number of research papers have been published demonstrating that in conjunction with MRI, (18)F-FDOPA PET provides greater sensitivity and specificity than these modalities in detection, grading, prognosis and validation of treatment success in both primary and recurrent gliomas. In further comparisons with (11)C-MET, (18)F-FLT, (18)F-FET and MRI, (18)F-FDOPA has shown similar or better efficacy. Recently synthesis cassettes have become available, making (18)F-FDOPA more accessible. According to the available data, (18)F-FDOPA PET is a viable radiotracer for imaging and treatment planning of gliomas. ADVANCES IN KNOWLEDGE AND IMPLICATION FOR PATIENT CARE: (18)F-FDOPA PET appears to be a viable radiopharmaceutical for the diagnosis and treatment planning of gliomas cases, improving on that of MRI and (18)F-FDG PET.

ABSTRACT Arterial input function (AIF) is important for the determination of cerebral blood flow and the analysis of related disease. Detection of artery voxels in dynamic contrast-enhanced (DCE) MRI is the key challenge in estimating the AIF. In the presence of tumour tissue, automatic detection of arteries becomes as even more challenging task. In this paper we propose a supervised machine-learning based method for the detection of artery voxels in DCE-MRI of the brain. The method utilises a set of kinetic and local structural features with a logistic regression classifier in order to detect arterial voxels in the image. The performance of the method is evaluated on 11 DCE-MRI datasets, of patients with diagnosed brain cancer, in terms of area under the ROC curve and in terms of correlation with an ideal AIF. The results of the evaluation suggest that the proposed method has the potential to be used as a tool for accurate estimation of AIF in DCE-MRI of the brain.

Metabolic imaging using positron emission tomography (PET) has found increasing clinical use for the management of infiltrating tumours such as glioma. However, the heterogeneous biological nature of tumours and intrinsic treatment resistance in some regions means that knowledge of multiple biological factors is needed for effective treatment planning. For example, the use of (18)F-FDOPA to identify infiltrative tumour and (18)F-FMISO for localizing hypoxic regions. Performing multiple PET acquisitions is impractical in many clinical settings, but previous studies suggest multiplexed PET imaging could be viable. The fidelity of the two signals is affected by the injection interval, scan timing and injected dose. The contribution of this work is to propose a framework to explicitly trade-off signal fidelity with logistical constraints when designing the imaging protocol. The particular case of estimating (18)F-FMISO from a single frame prior to injection of (18)F-FDOPA is considered. Theoretical experiments using simulations for typical biological scenarios in humans demonstrate that results comparable to a pair of single-tracer acquisitions can be obtained provided protocol timings are carefully selected. These results were validated using a pre-clinical data set that was synthetically multiplexed. The results indicate that the dual acquisition of (18)F-FMISO and (18)F-FDOPA could be feasible in the clinical setting. The proposed framework could also be used to design protocols for other tracers.

There is significant interest in the development of improved image-guided therapy for neuro-oncology applications. Glioblastomas (GBM) in particular present a considerable challenge because of their pervasive nature, propensity for recurrence, and resistance to conventional therapies. MRI is routinely used as a guide for planning treatment strategies. However, this imaging modality is not able to provide images that clearly delineate tumor boundaries and affords only indirect information about key tumor pathophysiology. With the emergence of PET imaging with new oncology radiotracers, mapping of tumor infiltration and other important molecular events such as hypoxia is now feasible within the clinical setting. In particular, the importance of imaging hypoxia levels within the tumoral microenvironment is gathering interest, as hypoxia is known to play a central role in glioma pathogenesis and resistance to treatment. One of the hypoxia radiotracers known for its clinical utility is (18)F-fluoromisodazole ((18)F-FMISO). In this review, we highlight the typical causes of treatment failure in gliomas that may be linked to hypoxia and outline current methods for the detection of hypoxia. We also provide an overview of the growing body of studies focusing on the clinical translation of (18)F-FMISO PET imaging, strengthening the argument for the use of (18)F-FMISO hypoxia imaging to help optimize and guide treatment strategies for patients with glioblastoma.

ABSTRACT In clinically applicable structural magnetic resonance images (MRI), bone and air have similarly low signal intensity, making the differentiation between them a very challenging task. MRI-based bone/air segmentation, however, is a critical step in some emerging applications, such as skull atlas building, MRI-based attenuation correction for Positron Emission Tomography (PET), and MRI-based radiotherapy planning. In view of the availability of hybrid PET-MRI machines, we propose a voxel-wise classification method for bone/air segmentation. The method is based on random forest theory and features extracted from structural MRI and attenuation uncorrected PET. The Dice Similarity Score (DSC) score between the segmentation result and the 'ground truth' obtained by thresholding Computed Tomography images was calculated for validation. Images from 10 subjects were used for validation, achieving a DSC of 0.83±0.08 and 0.98±0.01 for air and bone, respectively. The results suggest that structural MRI and uncorrected PET can be used to reliably differentiate between air and bone.

Abstract The clinical interpretation of breast MRI remains largely subjective, and the reported findings qualitative. Although the sensitivity of the method for detecting breast cancer is high, its specificity is poor. Computerised interpretation offers the possibility of improving ...

ABSTRACT Multi-voxel pattern analysis is an approach to investigating brain activity measured by functional Magnetic Resonance Imaging (fMRI) in response to given stimuli. The signal acquired using fMRI is spatiotemporal, and can be used to predict the stimuli causing brain activation. Existing prediction methods suffer from the ‘curse of dimensionality’ by embedding all time points of the experiment in feature space. Although this problem can be alleviated by feature selection in spatial domain so that informative voxels are selected, feature selection in temporal domain has not been attempted. Henceforth, it is unclear which spatiotemporal combination of fMRI data gives the best prediction. In this study, we investigate the effect of using different combinations of fMRI time points on the prediction accuracy of visual stimuli, using support vector machine and random forest as classification methods. Using a publicly available fMRI dataset, we demonstrate that classification using multiple concatenated time points significantly outperforms a single time point based classification. Our results highlight the necessity of considering both temporal and spatial patterns to achieve better prediction of visual stimuli from fMRI data.

ABSTRACT A new method for image thresholding of two or more images that are acquired in different modalities or acquisition protocols is proposed. The method is based on measures from information theory and has no underlying free parameters nor does it require training or calibration. The method is based on finding an optimal set of global thresholds, one for each image, by maximising the mutual information above the thresholds while minimising the mutual information below the thresholds. Although some assumptions on the nature of images are made, no assumptions are made by the method on the intensity distributions or on the shape of the image histograms. The effectiveness of the method is demonstrated both on synthetic images and medical images from clinical practice. It is then compared against three other thresholding methods.

This paper presents an empirical evaluation of the goodness-of-fit (GOF) of four parametric models of contrast enhancement for dynamic resonance imaging of the breast: the Tofts, Brix, and Hayton pharmacokinetic models, and a novel empiric model. The goodness-of-fit of each model was evaluated with respect to: (i) two model-fitting algorithms (Levenberg-Marquardt and Nelder-Mead) and two fitting tolerances; and (ii) temporal resolution. In the first case the GOF was measured using data from three dynamic contrast-enhanced (DCE) MRI data sets from routine clinical examinations: one case with benign enhancement, one with malignant enhancement, and one with normal findings. Results are presented for fits to both the whole breast volume and to a selected region of interest. In the second case the GOF was measured by first fitting the models to several temporally sub-sampled versions of a custom high temporal resolution data set (subset of the breast volume containing a malignant lesion)...

This paper presents a new algorithm for denoising dynamic contrast-enhanced (DCE) MR images. The algorithm is called Dynamic Non-Local Means and is a novel variation on the Non-Local Means (NL-Means) algorithm. It exploits the redundancy of information in the DCE-MRI sequence of images. An evaluation of the performance of the algorithm relative to six other denoising algorithms-Gaussian filtering, the original NL-Means algorithm, bilateral filtering, anisotropic diffusion filtering, the wavelets adaptive multiscale products threshold method, and the traditional wavelet thresholding method-is also presented. The evaluation was performed by two groups of expert observers-18 signal/image processing experts, and 9 clinicians (8 radiographers and 1 radiologist)-using real DCE-MRI data. The results of the evaluation provide evidence, at the alpha=0.05 level of significance, that both groups of observers deem the DNLM algorithm to perform visually better than all of the other algorithms.