Adaptive model initialization and deformation for automatic segmentation of T1-weighted brain MRI data

Red-light excitation of protoporphyrin IX fluorescence for subsurface tumor detection

OBJECTIVE The objective of this study was to detect 5-aminolevulinic acid (ALA)-induced tumor fluorescence from glioma below the surface of the surgical field by using red-light illumination. METHODS To overcome the shallow tissue penetration of blue light, which maximally excites the ALA-induced fluorophore protoporphyrin IX (PpIX) but is also strongly absorbed by hemoglobin and oxyhemoglobin, a system was developed to illuminate the surgical field with red light (620-640 nm) matching a secondary, smaller absorption peak of PpIX and detecting the fluorescence emission through a 650-nm longpass filter. This wide-field spectroscopic imaging system was used in conjunction with conventional blue-light fluorescence for comparison in 29 patients undergoing craniotomy for resection of high-grade glioma, low-grade glioma, meningioma, or metastasis. RESULTS Although, as expected, red-light excitation is less sensitive to PpIX in exposed tumor, it did reveal tumor at a depth up to 5 mm below the resection bed in 22 of 24 patients who also exhibited PpIX fluorescence under blue-light excitation during the course of surgery. CONCLUSIONS Red-light excitation of tumor-associated PpIX fluorescence below the surface of the surgical field can be achieved intraoperatively and enables detection of subsurface tumor that is not visualized under conventional blue-light excitation. Clinical trial registration no.: NCT02191488 (clinicaltrials.gov).

Robust generative asymmetric GMM for brain MR image segmentation

BACKGROUND AND OBJECTIVES

Accurate segmentation of brain tissues from magnetic resonance (MR) images based on the unsupervised statistical models such as Gaussian mixture model (GMM) has been widely studied during last decades. However, most GMM based segmentation methods suffer from limited accuracy due to the influences of noise and intensity inhomogeneity in brain MR images. To further improve the accuracy for brain MR image segmentation, this paper presents a Robust Generative Asymmetric GMM (RGAGMM) for simultaneous brain MR image segmentation and intensity inhomogeneity correction.

METHOD

First, we develop an asymmetric distribution to fit the data shapes, and thus construct a spatial constrained asymmetric model. Then, we incorporate two pseudo-likelihood quantities and bias field estimation into the model's log-likelihood, aiming to exploit the neighboring priors of within-cluster and between-cluster and to alleviate the impact of intensity inhomogeneity, respectively. Finally, an expectation maximization algorithm is derived to iteratively maximize the approximation of the data log-likelihood function to overcome the intensity inhomogeneity in the image and segment the brain MR images simultaneously.

RESULTS

To demonstrate the performances of the proposed algorithm, we first applied the proposed algorithm to a synthetic brain MR image to show the intermediate illustrations and the estimated distribution of the proposed algorithm. The next group of experiments is carried out in clinical 3T-weighted brain MR images which contain quite serious intensity inhomogeneity and noise. Then we quantitatively compare our algorithm to state-of-the-art segmentation approaches by using Dice coefficient (DC) on benchmark images obtained from IBSR and BrainWeb with different level of noise and intensity inhomogeneity. The comparison results on various brain MR images demonstrate the superior performances of the proposed algorithm in dealing with the noise and intensity inhomogeneity.

CONCLUSION

In this paper, the RGAGMM algorithm is proposed which can simply and efficiently incorporate spatial constraints into an EM framework to simultaneously segment brain MR images and estimate the intensity inhomogeneity. The proposed algorithm is flexible to fit the data shapes, and can simultaneously overcome the influence of noise and intensity inhomogeneity, and hence is capable of improving over 5% segmentation accuracy comparing with several state-of-the-art algorithms.

Intraoperative image updating for brain shift following dural opening

OBJECTIVE Preoperative magnetic resonance images (pMR) are typically coregistered to provide intraoperative navigation, the accuracy of which can be significantly compromised by brain deformation. In this study, the authors generated updated MR images (uMR) in the operating room (OR) to compensate for brain shift due to dural opening, and evaluated the accuracy and computational efficiency of the process. METHODS In 20 open cranial neurosurgical cases, a pair of intraoperative stereovision (iSV) images was acquired after dural opening to reconstruct a 3D profile of the exposed cortical surface. The iSV surface was registered with pMR to detect cortical displacements that were assimilated by a biomechanical model to estimate whole-brain nonrigid deformation and produce uMR in the OR. The uMR views were displayed on a commercial navigation system and compared side by side with the corresponding coregistered pMR. A tracked stylus was used to acquire coordinate locations of features on the cortical surface that served as independent positions for calculating target registration errors (TREs) for the coregistered uMR and pMR image volumes. RESULTS The uMR views were visually more accurate and well aligned with the iSV surface in terms of both geometry and texture compared with pMR where misalignment was evident. The average misfit between model estimates and measured displacements was 1.80 ± 0.35 mm, compared with the average initial misfit of 7.10 ± 2.78 mm between iSV and pMR, and the average TRE was 1.60 ± 0.43 mm across the 20 patients in the uMR image volume, compared with 7.31 ± 2.82 mm on average in the pMR cases. The iSV also proved to be accurate with an average error of 1.20 ± 0.37 mm. The overall computational time required to generate the uMR views was 7-8 minutes. CONCLUSIONS This study compensated for brain deformation caused by intraoperative dural opening using computational model-based assimilation of iSV cortical surface displacements. The uMR proved to be more accurate in terms of model-data misfit and TRE in the 20 patient cases evaluated relative to pMR. The computational time was acceptable (7-8 minutes) and the process caused minimal interruption of surgical workflow.

Intraoperative fiducial-less patient registration using volumetric 3D ultrasound: a prospective series of 32 neurosurgical cases

OBJECT

Fiducial-based registration (FBR) is used widely for patient registration in image-guided neurosurgery. The authors of this study have developed an automatic fiducial-less registration (FLR) technique to find the patient-to-image transformation by directly registering 3D ultrasound (3DUS) with MR images without incorporating prior information. The purpose of the study was to evaluate the performance of the FLR technique when used prospectively in the operating room and to compare it with conventional FBR.

METHODS

In 32 surgical patients who underwent conventional FBR, preoperative T1-weighted MR images (pMR) with attached fiducial markers were acquired prior to surgery. After craniotomy but before dural opening, a set of 3DUS images of the brain volume was acquired. A 2-step registration process was executed immediately after image acquisition: 1) the cortical surfaces from pMR and 3DUS were segmented, and a multistart sum-of-squared-intensity-difference registration was executed to find an initial alignment between down-sampled binary pMR and 3DUS volumes; and 2) the alignment was further refined by a mutual information-based registration between full-resolution grayscale pMR and 3DUS images, and a patient-to-image transformation was subsequently extracted.

RESULTS

To assess the accuracy of the FLR technique, the following were quantified: 1) the fiducial distance error (FDE); and 2) the target registration error (TRE) at anterior commissure and posterior commissure locations; these were compared with conventional FBR. The results showed that although the average FDE (6.42 ± 2.05 mm) was higher than the fiducial registration error (FRE) from FBR (3.42 ± 1.37 mm), the overall TRE of FLR (2.51 ± 0.93 mm) was lower than that of FBR (5.48 ± 1.81 mm). The results agreed with the intent of the 2 registration techniques: FBR is designed to minimize the FRE, whereas FLR is designed to optimize feature alignment and hence minimize TRE. The overall computational cost of FLR was approximately 4-5 minutes and minimal user interaction was required.

CONCLUSIONS

Because the FLR method directly registers 3DUS with MR by matching internal image features, it proved to be more accurate than FBR in terms of TRE in the 32 patients evaluated in this study. The overall efficiency of FLR in terms of the time and personnel involved is also improved relative to FBR in the operating room, and the method does not require additional image scans immediately prior to surgery. The performance of FLR and these results suggest potential for broad clinical application.

Stereovision to MR image registration for cortical surface displacement mapping to enhance image-guided neurosurgery

PURPOSE

A surface registration method is presented to align intraoperative stereovision (iSV) with preoperative magnetic resonance (pMR) images, which utilizes both geometry and texture information to extract tissue displacements as part of the overall process of compensating for intraoperative brain deformation in order to maintain accurate neuronavigational image guidance during surgery.

METHODS

A sum-of-squared-difference rigid image registration was first executed to detect lateral shift of the cortical surface and was followed by a mutual-information-based block matching method to detect local nonrigid deformation caused by distention or collapse of the cortical surface. Ten (N = 10) surgical cases were evaluated in which an independent point measurement of a dominant cortical surface feature location was recorded with a tracked stylus in each case and compared to its surface-registered counterpart. The full three-dimensional (3D) displacement field was also extracted to drive a biomechanical brain deformation model, the results of which were reconciled with the reconstructed iSV surface as another form of evaluation.

RESULTS

Differences between the tracked stylus coordinates of cortical surface features and their surface-registered locations were 1.94 ± 0.59 mm on average across the ten cases. When the complete displacement map derived from surface registration was utilized, the resulting images generated from mechanical model updates were consistent in terms of both geometry (1-2 mm of model misfit) and texture, and were generated with less than 10 min of computational time. Analysis of the surface-registered 3D displacements indicate that the magnitude of motion ranged from 4.03 to 9.79 mm in the ten patient cases, and the amount of lateral shift was not related statistically to the direction of gravity (p = 0.73 ≫ 0.05) or the craniotomy size (p = 0.48 ≫ 0.05) at the beginning of surgery.

CONCLUSIONS

The iSV-pMR surface registration method utilizes texture and geometry information to extract both global lateral shift and local nonrigid movement of the cortical surface in 3D. The results suggest small differences exist in surface-registered locations when compared to positions measured independently with a coregistered stylus and when the full iSV surface was aligned with model-updated MR. The effectiveness and efficiency of the registration method is also minimally disruptive to surgical workflow.

A new parametric model-based technique in bone tumour analysis

The study presents a new statistical model based segmentation technique dedicated to inhomogeneous bone tumours structure analysis. The presented 3-D segmentation procedure applies a statistic description of the structure based on Gaussian mixture model and an adaptive model-based relative fuzzy connectedness technique. It has been tested on 94 different MR series of 38 young patients. The final segmentation results have been evaluated using two different verification techniques and compared with other segmentation methods. The developed technique yields higher bone tumours segmentation accuracy compared to results obtained with conventional fuzzy connectedness approach and different segmentation methods presented in the literature, and based on active contour models or statistical analysis.

Intraoperative patient registration using volumetric true 3D ultrasound without fiducials

PURPOSE

Accurate patient registration is crucial for effective image-guidance in open cranial surgery. Typically, it is accomplished by matching skin-affixed fiducials manually identified in the operating room (OR) with their counterparts in the preoperative images, which not only consumes OR time and personnel resources but also relies on the presence (and subsequent fixation) of the fiducials during the preoperative scans (until the procedure begins). In this study, the authors present a completely automatic, volumetric image-based patient registration technique that does not rely on fiducials by registering tracked (true) 3D ultrasound (3DUS) directly with preoperative magnetic resonance (MR) images.

METHODS

Multistart registrations between binary 3DUS and MR volumes were first executed to generate an initial starting point without incorporating prior information on the US transducer contact point location or orientation for subsequent registration between grayscale 3DUS and MR via maximization of either mutual information (MI) or correlation ratio (CR). Patient registration was then computed through concatenation of spatial transformations.

RESULTS

In ten (N = 10) patient cases, an average fiducial (marker) distance error (FDE) of 5.0 mm and 4.3 mm was achieved using MI or CR registration (FDE was smaller with CR vs MI in eight of ten cases), which are comparable to values reported for typical fiducial- or surface-based patient registrations. The translational and rotational capture ranges were found to be 24.0 mm and 27.0° for binary registrations (up to 32.8 mm and 36.4°), 12.2 mm and 25.6° for MI registrations (up to 18.3 mm and 34.4°), and 22.6 mm and 40.8° for CR registrations (up to 48.5 mm and 65.6°), respectively. The execution time to complete a patient registration was 12-15 min with parallel processing, which can be significantly reduced by confining the 3DUS transducer location to the center of craniotomy in MR before registration (an execution time of 5 min is achievable).

CONCLUSIONS

Because common features deep in the brain and throughout the surgical volume of interest are used, intraoperative fiducial-less patient registration is possible on-demand, which is attractive in cases where preoperative patient registration is compromised (e.g., from loss∕movement of skin-affixed fiducials) or not possible (e.g., in cases of emergency when external fiducials were not placed in time). CR registration was more robust than MI (capture range about twice as big) and appears to be more accurate, although both methods are comparable to or better than fiducial-based registration in the patient cases evaluated. The results presented here suggest that 3DUS image-based patient registration holds promise for clinical application in the future.

Estimation of brain deformation for volumetric image updating in protoporphyrin IX fluorescence-guided resection

INTRODUCTION

Fluorescence-guided resection (FGR) of brain tumors is an intuitive, practical and emerging technology for visually delineating neoplastic tissue exposed intraoperatively. Image guidance is the standard technique for producing 3-dimensional spatially coregistered information for surgical decision making. Both technologies together are synergistic: the former detects surface fluorescence as a biomarker of the current surgical margin while the latter shows coregistered volumetric neuroanatomy but can be degraded by intraoperative brain shift. We present the implementation of deformation modeling for brain shift compensation in protoporphyrin IX FGR, integrating these two sources of information for maximum surgical benefit.

METHODS

Two patients underwent FGR coregistered with conventional image guidance. Histopathological analysis, intraoperative fluorescence and image space coordinates were recorded for biopsy specimens acquired during surgery. A biomechanical brain deformation model driven by intraoperative ultrasound data was used to generate updated MR images.

RESULTS

Combined use of fluorescence signatures and updated MR image information showed substantially improved accuracy compared to fluorescence or the original (i.e., nonupdated) MR images, detecting only true positives and true negatives, and no instances of false positives or false negatives.

CONCLUSION

Implementation of brain deformation modeling in FGR shows promise for increasing the accuracy of neurosurgical guidance in the delineation and resection of brain tumors.

Data assimilation using a gradient descent method for estimation of intraoperative brain deformation

Biomechanical models that simulate brain deformation are gaining attention as alternatives for brain shift compensation. One approach, known as the "forced-displacement method", constrains the model to exactly match the measured data through boundary condition (BC) assignment. Although it improves model estimates and is computationally attractive, the method generates fictitious forces and may be ill-advised due to measurement uncertainty. Previously, we have shown that by assimilating intraoperatively acquired brain displacements in an inversion scheme, the Representer algorithm (REP) is able to maintain stress-free BCs and improve model estimates by 33% over those without data guidance in a controlled environment. However, REP is computationally efficient only when a few data points are used for model guidance because its costs scale linearly in the number of data points assimilated, thereby limiting its utility (and accuracy) in clinical settings. In this paper, we present a steepest gradient descent algorithm (SGD) whose computational complexity scales nearly invariantly with the number of measurements assimilated by iteratively adjusting the forcing conditions to minimize the difference between measured and model-estimated displacements (model-data misfit). Solutions of full linear systems of equations are achieved with a parallelized direct solver on a shared-memory, eight-processor Linux cluster. We summarize the error contributions from the entire process of model-updated image registration compensation and we show that SGD is able to attain model estimates comparable to or better than those obtained with REP, capturing about 74-82% of tumor displacement, but with a computational effort that is significantly less (a factor of 4-fold or more reduction relative to REP) and nearly invariant to the amount of sparse data involved when the number of points assimilated is large. Based on five patient cases, an average computational cost of approximately 2 min for estimating whole-brain deformation has been achieved with SGD using 100 sparse data points, suggesting the new algorithm is sufficiently fast with adequate accuracy for routine use in the operating room (OR).

Brain-skull contact boundary conditions in an inverse computational deformation model

Biomechanical models simulating brain motion under loading and boundary conditions in the operating room (OR) are gaining attention as alternatives for brain shift compensation during open cranial neurosurgeries. Although the significance of brain-skull boundary conditions (BCs) in these models has been explored in dynamic simulations, it has not been fully investigated in models representing the quasi-static brain motion that prevails during neurosurgery. In this study, we extend the application of a brain-skull contact BC by incorporating it into an inversion estimation scheme for the deformation field using the steepest gradient descent (SGD) framework. The technique allows parenchymal surface motion normal to the skull while maintaining stress-free BCs at the craniotomy and minimizing the effect of measurement noise. Application of the algorithm in five clinical cases using sparse data generated at the tumor boundary confirms the significance of brain-skull BCs in the model response. Specifically, the results demonstrate that the contact BC enhances model flexibility and achieves improved or comparable performance at the tumor boundary (recovering about 85% of the deformation) relative to that obtained when normal motion of the parenchymal surface is not allowed. It also significantly improves model estimation accuracy at the craniotomy (1.6mm on average), especially when the normal motion is large. The importance of the method is that model performance significantly improves when brain-skull contact influences the deformation field but does not degrade when the contact is less critical and simpler BCs would suffice. The computational cost of the technique is currently 3.9 min on average, but may be further reduced by applying an iterative solver to the linear systems of equations involved and/or by local refinement of the mesh in regions of interest.

A Scalable Framework For Segmenting Magnetic Resonance Images

A fast, accurate and fully automatic method of segmenting magnetic resonance images of the human brain is introduced. The approach scales well allowing fast segmentations of fine resolution images. The approach is based on modifications of the soft clustering algorithm, fuzzy c-means, that enable it to scale to large data sets. Two types of modifications to create incremental versions of fuzzy c-means are discussed. They are much faster when compared to fuzzy c-means for medium to extremely large data sets because they work on successive subsets of the data. They are comparable in quality to application of fuzzy c-means to all of the data. The clustering algorithms coupled with inhomogeneity correction and smoothing are used to create a framework for automatically segmenting magnetic resonance images of the human brain. The framework is applied to a set of normal human brain volumes acquired from different magnetic resonance scanners using different head coils, acquisition parameters and field strengths. Results are compared to those from two widely used magnetic resonance image segmentation programs, Statistical Parametric Mapping and the FMRIB Software Library (FSL). The results are comparable to FSL while providing significant speed-up and better scalability to larger volumes of data.

Virtual definition of neuronal tissue by cluster analysis of multi-parametric imaging (virtual-dot-com imaging)

Individual mapping of cerebral, morphological, functionally related structures using MRI was carried out using a new multi-contrast acquisition and analysis framework, called virtual-dot-com imaging. So far, conventional anatomical MRI has been able to provide gross segmentation of gray/white matter boundaries and a few sub-cortical structures. By combining a handful of imaging contrasts mechanisms (T1, T2, magnetization transfer, T2* and proton density), we were able to further segment sub-cortical tissue to its sub-nuclei arrangement, a segmentation that is difficult based on conventional, single-contrast MRI. Using an automatic four-step image and signal processing algorithm, we segmented the thalamus to at least 7 sub-nuclei with high similarity across subjects and high statistical significance within subjects (p<0.0001). The identified sub-nuclei resembled the known anatomical arrangement of the thalamus given in various atlases. Each cluster was characterized by a unique MRI contrast fingerprint. With this procedure, the weighted proportions of the different cellular compartments could be estimated, a property available to date only by histological analysis. Each sub-nucleus could be characterized in terms of normalized MRI contrast and compared to other sub-nuclei. The different weights of the contrasts (T1/T2/T2*/PD/MT, etc.) for each sub-nuclei cluster might indicate the intra-cluster morphological arrangement of the tissue that it represents. The implications of this methodology are far-ranging, from non-invasive, in vivo, individual mapping of histologically distinct brain areas to automatic identification of pathological processes.