Supervised learning based multimodal MRI brain tumour segmentation using texture features from supervoxels.

BACKGROUND: Accurate segmentation of brain tumour in magnetic resonance images (MRI) is a difficult task due to various tumour types. Using information and features from multimodal MRI including structural MRI and isotropic (p) and anisotropic (q) components derived from the diffusion tensor imaging (DTI) may result in a more accurate analysis of brain images. METHODS: We propose a novel 3D supervoxel based learning method for segmentation of tumour in multimodal MRI brain images (conventional MRI and DTI). Supervoxels are generated using the information across the multimodal MRI dataset. For each supervoxel, a variety of features including histograms of texton descriptor, calculated using a set of Gabor filters with different sizes and orientations, and first order intensity statistical features are extracted. Those features are fed into a random forests (RF) classifier to classify each supervoxel into tumour core, oedema or healthy brain tissue. RESULTS: The method is evaluated on two datasets: 1) Our clinical dataset: 11 multimodal images of patients and 2) BRATS 2013 clinical dataset: 30 multimodal images. For our clinical dataset, the average detection sensitivity of tumour (including tumour core and oedema) using multimodal MRI is 86% with balanced error rate (BER) 7%; while the Dice score for automatic tumour segmentation against ground truth is 0.84. The corresponding results of the BRATS 2013 dataset are 96%, 2% and 0.89, respectively. CONCLUSION: The method demonstrates promising results in the segmentation of brain tumour. Adding features from multimodal MRI images can largely increase the segmentation accuracy. The method provides a close match to expert delineation across all tumour grades, leading to a faster and more reproducible method of brain tumour detection and delineation to aid patient management.

Automatic image quality assessment and measurement of fetal head in two-dimensional ultrasound image.

Owing to the inconsistent image quality existing in routine obstetric ultrasound (US) scans that leads to a large intraobserver and interobserver variability, the aim of this study is to develop a quality-assured, fully automated US fetal head measurement system. A texton-based fetal head segmentation is used as a prerequisite step to obtain the head region. Textons are calculated using a filter bank designed specific for US fetal head structure. Both shape- and anatomic-based features calculated from the segmented head region are then fed into a random forest classifier to determine the quality of the image (e.g., whether the image is acquired from a correct imaging plane), from which fetal head measurements [biparietal diameter (BPD), occipital-frontal diameter (OFD), and head circumference (HC)] are derived. The experimental results show a good performance of our method for US quality assessment and fetal head measurements. The overall precision for automatic image quality assessment is 95.24% with 87.5% sensitivity and 100% specificity, while segmentation performance shows 99.27% ([Formula: see text]) of accuracy, 97.07% ([Formula: see text]) of sensitivity, 2.23 mm ([Formula: see text]) of the maximum symmetric contour distance, and 0.84 mm ([Formula: see text]) of the average symmetric contour distance. The statistical analysis results using paired [Formula: see text]-test and Bland-Altman plots analysis indicate that the 95% limits of agreement for inter observer variability between the automated measurements and the senior expert measurements are 2.7 mm of BPD, 5.8 mm of OFD, and 10.4 mm of HC, whereas the mean differences are [Formula: see text], [Formula: see text], and [Formula: see text], respectively. These narrow 95% limits of agreements indicate a good level of consistency between the automated and the senior expert's measurements.

Automated brain tumour detection and segmentation using superpixel-based extremely randomized trees in FLAIR MRI.

PURPOSE: We propose a fully automated method for detection and segmentation of the abnormal tissue associated with brain tumour (tumour core and oedema) from Fluid- Attenuated Inversion Recovery (FLAIR) Magnetic Resonance Imaging (MRI). METHODS: The method is based on superpixel technique and classification of each superpixel. A number of novel image features including intensity-based, Gabor textons, fractal analysis and curvatures are calculated from each superpixel within the entire brain area in FLAIR MRI to ensure a robust classification. Extremely randomized trees (ERT) classifier is compared with support vector machine (SVM) to classify each superpixel into tumour and non-tumour. RESULTS: The proposed method is evaluated on two datasets: (1) Our own clinical dataset: 19 MRI FLAIR images of patients with gliomas of grade II to IV, and (2) BRATS 2012 dataset: 30 FLAIR images with 10 low-grade and 20 high-grade gliomas. The experimental results demonstrate the high detection and segmentation performance of the proposed method using ERT classifier. For our own cohort, the average detection sensitivity, balanced error rate and the Dice overlap measure for the segmented tumour against the ground truth are 89.48 %, 6 % and 0.91, respectively, while, for the BRATS dataset, the corresponding evaluation results are 88.09 %, 6 % and 0.88, respectively. CONCLUSIONS: This provides a close match to expert delineation across all grades of glioma, leading to a faster and more reproducible method of brain tumour detection and delineation to aid patient management.

DCE-MRI Perfusion and Permeability Parameters as predictors of tumor response to CCRT in Patients with locally advanced NSCLC.

In this prospective study, 36 patients with stage III non-small cell lung cancers (NSCLC), who underwent dynamic contrast-enhanced MRI (DCE-MRI) before concurrent chemo-radiotherapy (CCRT) were enrolled. Pharmacokinetic analysis was carried out after non-rigid motion registration. The perfusion parameters [including Blood Flow (BF), Blood Volume (BV), Mean Transit Time (MTT)] and permeability parameters [including endothelial transfer constant (Ktrans), reflux rate (Kep), fractional extravascular extracellular space volume (Ve), fractional plasma volume (Vp)] were calculated, and their relationship with tumor regression was evaluated. The value of these parameters on predicting responders were calculated by receiver operating characteristic (ROC) curve. Multivariate logistic regression analysis was conducted to find the independent variables. Tumor regression rate is negatively correlated with Ve and its standard variation Ve_SD and positively correlated with Ktrans and Kep. Significant differences between responders and non-responders existed in Ktrans, Kep, Ve, Ve_SD, MTT, BV_SD and MTT_SD (P < 0.05). ROC indicated that Ve < 0.24 gave the largest area under curve of 0.865 to predict responders. Multivariate logistic regression analysis also showed Ve was a significant predictor. Baseline perfusion and permeability parameters calculated from DCE-MRI were seen to be a viable tool for predicting the early treatment response after CCRT of NSCLC.

A supervised texton based approach for automatic segmentation and measurement of the fetal head and femur in 2D ultrasound images.

This paper presents a supervised texton based approach for the accurate segmentation and measurement of ultrasound fetal head (BPD, OFD, HC) and femur (FL). The method consists of several steps. First, a non-linear diffusion technique is utilized to reduce the speckle noise. Then, based on the assumption that cross sectional intensity profiles of skull and femur can be approximated by Gaussian-like curves, a multi-scale and multi-orientation filter bank is designed to extract texton features specific to ultrasound fetal anatomic structure. The extracted texton cues, together with multi-scale local brightness, are then built into a unified framework for boundary detection of ultrasound fetal head and femur. Finally, for fetal head, a direct least square ellipse fitting method is used to construct a closed head contour, whilst, for fetal femur a closed contour is produced by connecting the detected femur boundaries. The presented method is demonstrated to be promising for clinical applications. Overall the evaluation results of fetal head segmentation and measurement from our method are comparable with the inter-observer difference of experts, with the best average precision of 96.85%, the maximum symmetric contour distance (MSD) of 1.46 mm, average symmetric contour distance (ASD) of 0.53 mm; while for fetal femur, the overall performance of our method is better than the inter-observer difference of experts, with the average precision of 84.37%, MSD of 2.72 mm and ASD of 0.31 mm.

Reproducibility of Dynamic Contrast-Enhanced MRI in Renal Cell Carcinoma: A Prospective Analysis on Intra- and Interobserver and Scan-Rescan Performance of Pharmacokinetic Parameters.

The objective of this study was to investigate the intra- and interobserver as well as scan-rescan reproducibility of quantitative parameters of renal cell carcinomas (RCCs) with dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI). A total of 21 patients with clear cell RCCs (17 men, 4 woman; age 37-69 years, mean age 54.6 years, mean size, 5.0 ±â€Š2.2 cm) were prospectively recruited from September 2012 to November 2012. Patients underwent paired DCE-MRI studies on a 3.0 T MR system with an interval of 48 to 72 hours. The extended-Tofts model and population-based arterial input function were used to calculate kinetic parameters. Three observers defined the 2-dimensional whole-tumor region of interest at the slice with the maximum diameter of the RCC. Intraobserver and scan-rescan differences were assessed using paired t tests, whereas interobserver differences using two-way analysis of variance. Intra- and interobserver reproducibility and scan-rescan reproducibility were evaluated using within-subject coefficient of variation (wCoV) and intraclass correlation coefficient (ICC). There were no significant intra-, interobserver, or scan-rescan differences in parameters (all P > 0.05). All ICCs for intra- and interobserver agreements were >0.75 (P < 0.05), whereas the scan-rescan agreement was moderate to good; V(e) (0.764, 95% confidence interval [CI]: 0.378-0.925) and K(ep) (0.906, 95% CI: 0.710-0.972) had higher ICC than K(trans) (0.686; 95% CI: 0.212-0.898) and V(p) (0.657; 95% CI: 0.164-0.888). In intra- and interobserver variability analyses, all parameters except V(p) had low wCoV values. K(trans) and V(e) had slightly lower intraobserver wCoV (1.2% and 0.9%) compared with K(ep) (3.7%), whereas all 3 of these parameters had similar interobserver wCoV values (2.5%, 3.1%, and 2.9%, respectively). Regarding scan-rescan variability, K(trans) and K(ep) showed slightly higher variation (15.6% and 15.4%) than V(e) (10.1%). V(p) had the largest wCoV in all variability analyses (all >30%). DCE-MRI demonstrated good intra- and interobserver reproducibility and moderate to good scan-rescan performance in the assessment of RCC using K(trans), K(ep), and V(e) as parameters under noncontinuous scanning mode. V(p) showed poor reproducibility, and thus may not be suitable for this scanning protocol.

Areas of normal pulmonary parenchyma on HRCT exhibit increased FDG PET signal in IPF patients.

PURPOSE: Patients with idiopathic pulmonary fibrosis (IPF) show increased PET signal at sites of morphological abnormality on high-resolution computed tomography (HRCT). The purpose of this investigation was to investigate the PET signal at sites of normal-appearing lung on HRCT in IPF. METHODS: Consecutive IPF patients (22 men, 3 women) were prospectively recruited. The patients underwent (18)F-FDG PET/HRCT. The pulmonary imaging findings in the IPF patients were compared to the findings in a control population. Pulmonary uptake of (18)F-FDG (mean SUV) was quantified at sites of morphologically normal parenchyma on HRCT. SUVs were also corrected for tissue fraction (TF). The mean SUV in IPF patients was compared with that in 25 controls (patients with lymphoma in remission or suspected paraneoplastic syndrome with normal PET/CT appearances). RESULTS: The pulmonary SUV (mean ± SD) uncorrected for TF in the controls was 0.48 ± 0.14 and 0.78 ± 0.24 taken from normal lung regions in IPF patients (p < 0.001). The TF-corrected mean SUV in the controls was 2.24 ± 0.29 and 3.24 ± 0.84 in IPF patients (p < 0.001). CONCLUSION: IPF patients have increased pulmonary uptake of (18)F-FDG on PET in areas of lung with a normal morphological appearance on HRCT. This may have implications for determining disease mechanisms and treatment monitoring.

An in vivo subject-specific 3D functional knee joint model using combined MR imaging.

PURPOSE: We aim to quantitatively characterise the knee joint function in vivo under body-weight-bearing conditions via subject-specific models extracted from magnetic resonance (MR) data, in order to better understand the knee joint kinematic function in 3D. METHODS: Six healthy volunteers without any record of knee abnormality were scanned using a combined MR imaging strategy to record quasi-squatting motion and 3D knee anatomy. After a semi-automatic segmentation to delineate tibio-femoral articulation components, motion data were mapped to the anatomical data using a bi-rigid registration in order to achieve six degrees of freedom. The individual knee joint function was characterised by analysing the tibio-femoral articulation contact mechanism based on the reconstructed models in 3D and MR images in 2D. Contact points were extracted and their trajectory was plotted on the tibia plateau. RESULTS: The 3D models clearly show the relative rotation and gliding between tibia and femur during global flexion. Within the measured flexion arc, the contact points move less between 30[Formula: see text] and 100[Formula: see text] on both tibial plateaux as compared to that on the rest of the flexion arc. Four out of the six volunteers showed a global pattern of less moving extent of contact points on the medial tibial plateau than on the lateral tibial plateau in both 3D and 2D. CONCLUSION: The proposed subject-specific model is able to characterise knee joint kinematic function. It provides a way to describe knee joint surface kinematics quantitatively, which may help to better understand the knee function and joint derangements.

18F-Fluorodeoxyglucose positron emission tomography pulmonary imaging in idiopathic pulmonary fibrosis is reproducible: implications for future clinical trials.

PURPOSE: Noninvasive markers of disease activity in patients with idiopathic pulmonary fibrosis (IPF) are lacking. We performed this study to investigate the reproducibility of pulmonary (18)F-FDG PET/CT in patients with IPF. METHODS: The study group comprised 13 patients (11 men, 2 women; mean age 71.1 ± 9.9 years) with IPF recruited for two thoracic (18)F-FDG PET/CT studies performed within 2 weeks of each other. All patients were diagnosed with IPF in consensus at multidisciplinary meetings as a result of typical clinical, high-resolution CT and pulmonary function test features. Three methods for evaluating pulmonary (18)F-FDG uptake were used. The maximal (18)F-FDG pulmonary uptake (SUVmax) in the lungs was determined using manual region-of-interest placement. An (18)F-FDG uptake intensity histogram was automatically constructed from segmented lungs to evaluate the distribution of SUVs. Finally, mean SUV was determined for volumes-of-interest in pulmonary regions with interstitial lung changes identified on CT scans. Processing included correction for tissue fraction effects. Bland-Altman analysis was performed and interclass correlation coefficients (ICC) were determined to assess the reproducibility between the first and second PET scans, as well as the level of intraobserver and interobserver agreement. RESULTS: The mean time between the two scans was 6.3 ± 4.3 days. The interscan ICCs for pulmonary SUVmax analysis and mean SUV corrected for tissue fraction effects were 0.90 and 0.91, respectively. Intensity histograms were different in only 1 of the 13 paired studies. Intraobserver agreement was also excellent (0.80 and 0.85, respectively). Some bias was observed between observers, suggesting that serial studies would benefit from analysis by the same observer. CONCLUSION: This study demonstrated that there is excellent short-term reproducibility in pulmonary (18)F-FDG uptake in patients with IPF.

The importance of correction for tissue fraction effects in lung PET: preliminary findings.

PURPOSE: It has recently been recognized that PET/CT may play a role in diffuse parenchymal lung disease. However, interpretation can be confounded due to the variability in lung density both within and between individuals. To address this issue a novel correction method is proposed. METHODS: A CT scan acquired during shallow breathing is registered to a PET study and smoothed so as to match the PET resolution. This is used to derive voxel-based tissue fraction correction factors for the individual. The method was evaluated in a lung phantom study in which the lung was simulated by a Styrofoam/water mixture. The method was further evaluated using (18)F-FDG in 12 subjects free from pulmonary disease where ranges before and after correction were considered. RESULTS: Correction resulted in similar activity concentrations for the lung and background regions, consistent with the experimental phantom set-up. Correction resulted in reduced inter- and intrasubject variability in the estimated SUV. The possible application of the method was further demonstrated in five subjects with interstitial lung changes where increased SUV was demonstrated. Single study pre- and post-treatment studies were also analysed to further illustrate the utility of the method. CONCLUSION: The proposed tissue fraction correction method is a promising technique to account for variability of density in interpreting lung PET studies.

2D/3D fetal cardiac dataset segmentation using a deformable model.

PURPOSE: To segment the fetal heart in order to facilitate the 3D assessment of the cardiac function and structure. METHODS: Ultrasound acquisition typically results in drop-out artifacts of the chamber walls. The authors outline a level set deformable model to automatically delineate the small fetal cardiac chambers. The level set is penalized from growing into an adjacent cardiac compartment using a novel collision detection term. The region based model allows simultaneous segmentation of all four cardiac chambers from a user defined seed point placed in each chamber. RESULTS: The segmented boundaries are automatically penalized from intersecting at walls with signal dropout. Root mean square errors of the perpendicular distances between the algorithm's delineation and manual tracings are within 2 mm which is less than 10% of the length of a typical fetal heart. The ejection fractions were determined from the 3D datasets. We validate the algorithm using a physical phantom and obtain volumes that are comparable to those from physically determined means. The algorithm segments volumes with an error of within 13% as determined using a physical phantom. CONCLUSIONS: Our original work in fetal cardiac segmentation compares automatic and manual tracings to a physical phantom and also measures inter observer variation.

A novel method for incorporating respiratory-matched attenuation correction in the motion correction of cardiac PET-CT studies.

Mismatches between PET and CT datasets due to respiratory effects can lead to artefactual perfusion defects. To overcome this, we have proposed a method of aligning a single CT with each frame of a gated PET study in a semi-automatic manner, incorporating a statistical shape model of the diaphragm and a rigid registration of the heart. This ensures that the structures that could influence the appearance of the reconstructed cardiac activity are correctly matched between emission and transmission datasets. When tested on two patient studies, it was found in both cases that attenuation correction using the proposed technique resulted in PET images that were closer to the gold standard of attenuation correction with a gated CT, compared with scenarios where only heart matching was considered (and not the diaphragm) or where no transformation was performed (i.e. where a single CT frame was used to attenuation-correct all PET frames). These preliminary results suggest that diaphragm matching between PET and CT improves the quantitative accuracy of reconstructed PET images and that the proposed method of using a statistical shape model to describe the diaphragm shape and motion, in combination with a rigid registration to determine respiratory-induced heart motion, is a feasible method of achieving this.

Novel approaches to the measurement of arterial blood flow from dynamic digital X-ray images.

We have developed two new algorithms for the measurement of blood flow from dynamic X-ray angiographic images. Both algorithms aim to improve on existing techniques. First, a model-based (MB) algorithm is used to constrain the concentration-distance curve matching approach. Second, a weighted optical flow algorithm (OP) is used to improve on point-based optical flow methods by averaging velocity estimates along a vessel with weighting based on the magnitude of the spatial derivative. The OP algorithm was validated using a computer simulation of pulsatile blood flow. Both the OP and the MB algorithms were validated using a physiological blood flow circuit. Dynamic biplane digital X-ray images were acquired following injection of iodine contrast medium into a variety of simulated arterial vessels. The image data were analyzed using our integrated angiographic analysis software SARA to give blood flow waveforms using the MB and OP algorithms. These waveforms were compared to flow measured using an electromagnetic flow meter (EMF). In total 4935 instantaneous measurements of flow were made and compared to the EMF recordings. It was found that the new algorithms showed low measurement bias and narrow limits of agreement and also out-performed the concentration-distance curve matching algorithm (ORG) and a modification of this algorithm (PA) in all studies.