Registration of head CT images to physical space using a weighted combination of points and surfaces

Most previously reported registration techniques that align three-dimensional image volumes by matching geometrical features such as points or surfaces use a single type of feature. We recently reported a hybrid registration technique that uses a weighted combination of multiple geometrical feature shapes. In this study we use the weighted geometrical feature (WGF) algorithm to register computed tomography (CT) images of the head to physical space using the skin surface only, the bone surface only, and various weighted combinations of these surfaces and one fiducial point (centroid of a bone-implanted marker). We use data acquired from 12 patients that underwent temporal lobe craniotomies for the resection of cerebral lesions. We evaluate and compare the accuracy of the registrations obtained using these various approaches by using as a reference gold standard the registration obtained using three bone-implanted markers. The results demonstrate that a combination of geometrical features can improve the accuracy of CT-to-physical space registration. Point-based registration requires a minimum of three noncolinear points. The position of a bone-implanted marker can be determined much more accurately than that of a skin-affixed marker or an anatomic landmark. A major disadvantage of using bone-implanted markers is that an invasive procedure is required to implant each marker. By combining surface information, the WGF algorithm allows registration to be performed using only one or two such markers. One important finding is that the use of a single very accurate point (a bone-implanted marker) allows very accurate surface-based registration to be achieved using very few surface points. Finally, the WGF algorithm, which not only allows the combination of multiple types of geometrical information but also handles point-based and surface-based registration as degenerate cases, could form the foundation of a "flexible" surgical navigation system that allows the surgeon to use what he considers the method most appropriate for an individual clinical situation.     

Effects of coregistration of MR to CT images on MR stereotactic accuracy

Coregistration of different modality imaging serves to increase the ease and accuracy of stereotactic procedures. In many cases, magnetic resonance (MR) stereotaxis is supplanting computerized tomography (CT). The advantages of increased anatomical detail and multiplanar imaging afforded by MR, however, are offset by its potential inaccuracy as well as the more cumbersome and less available nature of its hardware. A system has been developed by one of the authors by which MR imaging can be performed separately without a stereotactic fiducial headring. Then, immediately prior to surgery, a stereotactic CT scan is obtained and software is used to coregister CT and MR images anatomically by matching cranial landmarks in the two scans. The authors examined this system in six patients as well as with the use of a lucite phantom. After initially coregistering CT and MR images, six separate anatomical (for the patients) and eight artificial (for the phantom) targets were compared. With coregistration, in comparison to CT fiducial scans, errors in each axis are less than or equal to 1 mm using the Cosman-Roberts-Wells system. In fact, the coregistered images are more accurate than MR fiducial images, in the anteroposterior (p = 0.001), lateral (p < 0.05), and vertical (p < 0.03) planes. Three-dimensional error was significantly less in the coregistered scans than the MR fiducial images (p < 0.005). The coregistration procedure therefore not only increases the case of MR stereotaxis but also increases its accuracy.     

Visual assessment of the accuracy of retrospective registration of MR and CT images of the brain

In a previous study we demonstrated that automatic retrospective registration algorithms can frequently register magnetic resonance (MR) and computed tomography (CT) images of the brain with an accuracy of better than 2 mm, but in that same study we found that such algorithms sometimes fail, leading to errors of 6 mm or more. Before these algorithms can be used routinely in the clinic, methods must be provided for distinguishing between registration solutions that are clinically satisfactory and those that are not. One approach is to rely on a human observer to inspect the registration results and reject images that have been registered with insufficient accuracy. In this paper, we present a methodology for evaluating the efficacy of the visual assessment of registration accuracy. Since the clinical requirements for level of registration accuracy are likely to be application dependent, we have evaluated the accuracy of the observer's estimate relative to six thresholds: 1-6 mm. The performance of the observers was evaluated relative to the registration solution obtained using external fiducial markers that are screwed into the patient's skull and that are visible in both MR and CT images. This fiducial marker system provides the gold standard for our study. Its accuracy is shown to be approximately 0.5 mm. Two experienced, blinded observers viewed five pairs of clinical MR and CT brain images, each of which had each been misregistered with respect to the gold standard solution. Fourteen misregistrations were assessed for each image pair with misregistration errors distributed between 0 and 10 mm with approximate uniformity. For each misregistered image pair each observer estimated the registration error (in millimeters) at each of five locations distributed around the head using each of three assessment methods. These estimated errors were compared with the errors as measured by the gold standard to determine agreement relative to each of the six thresholds, where agreement means that the two errors lie on the same side of the threshold. The effect of error in the gold standard itself is taken into account in the analysis of the assessment methods. The results were analyzed by means of the Kappa statistic, the agreement rate, and the area of receiver-operating-characteristic (ROC) curves. No assessment performed well at 1 mm, but all methods performed well at 2 mm and higher. For these five thresholds, two methods agreed with the standard at least 80% of the time and exhibited mean ROC areas greater than 0.84. One of these same methods exhibited Kappa statistics that indicated good agreement relative to chance (Kappa > 0.6) between the pooled observers and the standard for these same five thresholds. Further analysis demonstrates that the results depend strongly on the choice of the distribution of misregistration errors presented to the observers.     

Automated 3-D registration of MR and CT images of the head

This paper discusses the application of voxel similarity measures in the automated registration of clinically acquired MR and CT data of the head. We describe a novel single-start multi-resolution approach to the optimization of these measures, and the issues involved in applying this to data having a range of different fields of view and sampling resolution. We compare four proposed measures of voxel similarity using the same optimization scheme when presented with 10 pairs of images with a range of initial misregistrations. The registration estimates are compared with those provided by manual point-based registration and evaluated by visual inspection to give an assessment of the robustness and accuracy of the different measures. One full-volume CT image set is used to investigate the performance of each measure when used to align truncated images from different regions in the head. The soft tissue correlation and mutual information measures were found to provide the most robust measures of misregistration, providing results comparable to or better than those from manual point-based registration for all but the most truncated image volumes.     

Frame slippage verification in stereotactic radiosurgery

PURPOSE: To develop a method for detecting frame slippage in stereotactic radiosurgery by interactively matching in three dimensions Digitally Reconstructed Radiographs (DRRs) to portal images. METHODS AND MATERIALS: DRRs are superimposed over orthogonal edge-detected portal image pairs obtained prior to treatment. By interactively manipulating the CT data in three dimensions (rotations and translations) new DRRs are generated and overlaid with the orthogonal portal images. This method of matching is able to account for ambiguities due to rotations and translations outside of the imaging plane. The matching procedure is performed with anatomical structures, and is used in tandem with a fiducial marker array attached to the stereotactic frame. The method is evaluated using portal images simulated from patient CT data and then tested using a radiographic head phantom. RESULTS: For simulation tests a mean radial alignment error of 0.82 mm was obtained with the 3D matching method compared to a mean error of 3.52 mm when using conventional matching techniques. For the head phantom tests the mean alignment displacement error for each of the stereotactic coordinates was found to be delta(x) = 0.95 mm, delta(y) = 1.06 mm, delta(z) = 0.99 mm, with a mean error radial of 1.94 mm (SD = 0.61 mm). CONCLUSION: Results indicate that the accuracy of the system is appropriate for stereotactic radiosurgery, and is therefore an effective tool for verification of frame slippage.     

Interleukin-2-induces development of denditric cells from cord blood CD34+ cells

Fiducial markers are reference points used in the registration of image space(s) with physical (patient) space. As applied to interactive, image-guided surgery, the registration of image space with physical space allows the current location of a surgical tool to be indicated on a computer display of patient-specific preoperative images. This intrasurgical guidance information is particularly valuable in surgery within the brain, where visual feedback is limited. The accuracy of the mapping between physical and image space depends upon the accuracy with which the fiducial markers were located in each coordinate system. To effect accurate space registration for interactive, image-guided neurosurgery, the use of permanent fiducial markers implanted into the surface of the skull is proposed in this paper. These small cylindrical markers are composed of materials that make them visible in the image sets. The challenge lies in locating the subcutaneous markers in physical space. This paper presents an ultrasonic technique for transcutaneously detecting the location of these markers. The technique incorporates an algorithm based on detection of characteristic properties of the reflected A-mode ultrasonic waveform. The results demonstrate that ultrasound is an appropriate technique for accurate transcutaneous marker localization. The companion paper to this article describes an automatic, enhanced implementation of the marker-localization theory described in this article.     

Interactive image directed neurosurgery: patient registration employing the Laitinen stereo-adapter

Fiducial markers visible on both imaging scans and the patient are a convenient and accurate method to register the patients head in stereotactic space preceeding an interactive image directed neurosurgical procedure. We have been using the Laitinen non-invasive relocatable stereo-adapter to carry the fidcucial markers both during the imaging and patient registration process. Since the adapter can be accurately remounted the surgical procedure can take place at any time interval after the imaging study. To compare registration accuracy studies were performed using a phantom carrying both surface fiducial markers and fiducial markers mounted on the stereo adapter. We have found that total system accuracy using either surface or adapter mounted markers and an optical tracking system is in the range of 3-5 mm and is acceptable for a broad range of neurosurgical applications.     

Accuracy evaluation of fusion of CT, MR, and spect images using commercially available software packages (SRS PLATO and IFS)

PURPOSE: A problem for clinicians is to mentally integrate information from multiple diagnostic sources, such as computed tomography (CT), magnetic resonance (MR), and single photon emission computed tomography (SPECT), whose images give anatomic and metabolic information. METHODS AND MATERIALS: To combine this different imaging procedure information, and to overlay correspondent slices, we used commercially available software packages (SRS PLATO and IFS). The algorithms utilize a fiducial-based coordinate system (or frame) with 3 N-shaped markers, which allows coordinate transformation of a clinical examination data set (9 spots for each transaxial section) to a stereotactic coordinate system. The N-shaped markers were filled with fluids visible in each modality (gadolinium for MR, calcium chloride for CT, and 99mTc for SPECT). The frame is relocatable, in the different acquisition modalities, by means of a head holder to which a face mask is fixed so as to immobilize the patient. Position errors due to the algorithms were obtained by evaluating the stereotactic coordinates of five sources detectable in each modality. RESULTS: SPECT and MR position errors due to the algorithms were evaluated with respect to CT: deltax was < or = 0.9 mm for MR and < or = 1.4 mm for SPECT, deltay was < or = 1 mm and < or = 3 mm for MR and SPECT, respectively. Maximal differences in distance between estimated and actual fiducial centers (geometric mismatch) were in the order of the pixel size (0.8 mm for CT, 1.4 mm for MR, and 1.8 mm for SPECT). In an attempt to distinguish necrosis from residual disease, the image fusion protocol was studied in 35 primary or metastatic brain tumor patients. CONCLUSIONS: The image fusion technique has a good degree of accuracy as well as the potential to improve the specificity of tissue identification and the precision of the subsequent treatment planning.     

Automatic registration of pelvic computed tomography data and magnetic resonance scans including a full circle method for quantitative accuracy evaluation

The purpose of this study is to develop a method for registration of CT and MR scans of the pelvis with minimal user interaction and to obtain a means for objective quantification of the registration accuracy of clinical data without markers. CT scans were registered with proton density MR scans using chamfer matching on automatically segmented bone. A fixed threshold was used to segment CT, while morphological filters were used to segment MR. The method was tested with transverse and coronal MR scans of 18 patients and sagittal MR scans of 8 patients. The registration accuracy was estimated by comparing (triangulating) registrations of a single CT scan with MR in different orientations in a "full circle." For example, CT is first matched on transverse MR, next transverse MR is matched independently on coronal MR, and finally coronal MR is matched independently on CT. The product of the three transformations is the identity if all matching steps are perfect. Deviations from identity occur both due to random errors and due to some types of systematic errors. MR was registered on MR (to close the "circle") by minimization of rms voxel value differences. CT-MR registration takes about 1 min, including user interaction. The random error for CT-MR registration with transverse or coronal MR was 0.5 mm in translation and 0.4 degree in rotation (standard deviation) for each axis. A systematic registration error of about 1 mm was demonstrated along the MR frequency encoding direction, which is attributed to the chemical shift. In conclusion, the presented algorithm efficiently and accurately registers pelvic CT and MR scans on bone. The "full circle" method provides an estimate of the registration accuracy on clinical data.     

Magnetic resonance imaging based digitally reconstructed radiographs, virtual simulation, and three-dimensional treatment planning for brain neoplasms

Currently, patients with brain neoplasms must undergo both computed tomography (CT) and magnetic resonance (MR) imaging to take advantage of CT's density information and MR's soft tissue imaging capabilities. A method has been developed that allows virtual simulation, digitally reconstructed radiographs (DRRs), and 3-D treatment planning of patients with brain neoplasms to be generated using only one T1-weighted MR data set. DRRs of an anthropomorphic RANDO head phantom were generated using MR and CT imaging. The MR based DRRs provided structural information equivalent to CT based DRRs. The spatial linearity of CT and MR image sets was evaluated by measuring the percent distortion and spatial error. There was no statistical difference in spatial linearity or accuracy between the CT and MR image sets. MR and CT based treatment planning were compared using a variety of different treatment accessories, field sizes, photon energies, and gantry positions. Doses at various points throughout the head phantom were used as comparison points between CT based heterogeneous, CT based homogenous, and MR based homogenous treatment planning of the head phantom. Lithium fluoride thermoluminescent dosimeters were used to verify the dosimetric accuracy of MR based treatment planning by taking measurements at these points. For treatment plans with fields that pass through large air cavities, such as the maxillary sinus, homogenous treatment planning produces unacceptable dosimetric error (2%-4%). For treatment plans with fields that pass through the skull, MR homogenous treatment planning can be used with a dosimetric accuracy of +/- 2%.     

Reduced in vitro clot lysis and release of more active platelet PAI-1 in polycythemia vera and essential thrombocythemia

The accuracy of image to patient registration is a critical issue in the intraoperative use of frameless stereotaxic instruments for surgical guidance. This study was performed to assess the accuracy of image to head phantom registration using several standard registration techniques and a clinical frameless stereotaxic instrument. Two types of radioopaque fiducial markers were fixed to a plastic head phantom, and a computed tomography scan of the phantom was performed in the routine fashion. Image to phantom registration was carried out using fiducial markers, fiducial markers plus surface fit, anatomic landmarks, and anatomic landmarks plus surface fit. After each registration, linear inaccuracy measurements were performed for each of 32 markers. Each registration was performed 10 times, and the overall mean error measurements and anterior and posterior error were computed and compared. The overall mean error was smallest for the fiducial registration alone (2.07 mm). The magnitude of error increased significantly for posterior locations for all other registration techniques, but it was not significantly increased for the fiducial registration method. For this hardware configuration, registration with surface-applied fiducial markers is measurably more accurate than registration with surface anatomic landmarks. The addition of surface points to the fiducial registration does not increase the registration accuracy but, in fact, increases the degree of error.     

A multiresolution approach for contour extraction from brain images

Many image registration methods use head surface, brain surface, or inner/outer surface of the skull to estimate rotation and translation parameters. The inner surface of the skull is also used for intracranial volume segmentation which is considered the first step in segmentation and analysis of brain images. The surface is usually characterized by a set of edge or contour points extracted from cross-sectional images. Automatic extraction of contour points is complicated by discontinuity of edges in the back of the eyes and ears and sometimes by a previous surgery or an inadequate field of view. We have developed an automated method for contour extraction that connects discontinuities using a multiresolution pyramid. Steps of the method are: (1) Contour points are found by an edge-tracking algorithm; (2) A multiresolution pyramid of contour points is constructed; (3) Contour points of reduced images are found; (4) From the continuous contour found at the lowest resolution, contour points at a higher resolution are found; (5) Step 4 is repeated until contour points at the highest resolution (original image) are found. The method runs fast and has been successful in extracting contours from MRI and CT images. We illustrate the method and its performance using MRI and CT images of the human brain.     

An Internet atlas of mouse development

A prototype two- and three-dimensional color atlas of mouse development is described. The prototype has been developed using two embryos, a 13.5 d normal mouse embryo and a PATCH mutant embryo of the same age. Serial sections of the embryos, with an external registration marker system, introduced into the paraffin embedding process, were prepared by standard histological methods. For the 2D atlas, color images were digitized from 100 consecutive sections of the normal embryo. For the 3D atlas, 300 gray scale images digitized from the mutant embryo were conformally warped and reconstructed into a 3D volume dataset. The external fiducial system facilitated the three-dimensional reconstruction by providing accurate registration of consecutive images and also allowed for precise spatial calibration and the correction for warping artifacts. The atlases, with their associated anatomical knowledge base, will be integrated into a multimedia on-line information resource via the Internet's World Wide Web (WWW) using an enhanced (patent pending, Eòlas Technologies) version of the Mosaic WWW browser program from the National Center for Supercomputer Applications. These programs will provide research biologists with a set of advanced tools to analyze normal and abnormal development.     

Metachronous bilateral Wilms'' tumor: the importance of time interval to the development of a second tumor

Radiosurgery, a bladeless brain surgery without opening skull, requires higher imaging accuracy as compared to microsurgery. Accordingly, we must refine the ways we use the MR scanner and interpret the obtained images. A well tuned and regularly calibrated MR scanner provides excellent images, which allow us to define detailed intracranial structures without distortion. This enables us to obtain a reliable imaging diagnosis despite the fact that pathologic diagnosis is not available in many radiosurgical patients. Due to its three-dimensional imaging and excellence in tissue and spatial contrast, MR is important in exploring new radiosurgical indications. Large arteriovenous malformations (AVM), dural AVM of cavernous sinus and trigeminal neuralgia are some of the successful examples. By using MR, longitudinal investigation of radiosurgical effects becomes feasible. For tumors, the longitudinal studies are important in optimization of dose selection. For AVM, MR is reliable in verification of radiosurgical result. The reliability is comparable to conventional x-ray angiography. Application of MR enhances the essence of noninvasiveness of radiosurgery.     

Feasibility study of magnetic resonance imaging-guided intranasal flexible microendoscopy

Interventional magnetic resonance imaging (MRI) offers potential advantages over conventional interventional modalities such as X-ray fluoroscopy, ultrasonography, and computed tomography (CT). In particular, it does not use ionizing radiation, can provide high-quality images, and allows acquisition of oblique sections. We have carried out a feasibility study on the use of interventional MRI to track a flexible microendoscope in the paranasal sinuses. In this cadaver study, high-speed MRI was used to track a passive marker attached to the end of the endoscope. Automatic image registration algorithms were used to transfer the coordinates of the endoscope tip into the preoperative MRI and CT images, enabling us to display the position of the endoscope in reformatted orthogonal views or in a rendered view of the preoperative images. The endoscope video images were digitized and could be displayed alongside an approximately aligned, rendered preoperative image. Intraoperative display was provided in the scanner room by means of an liquid crystal display (LCD) projector. We estimate the accuracy of the endoscope tracking to be approximately 2 mm.     

Risk analysis of linear accelerator radiosurgery

PURPOSE: To evaluate the toxicity of stereotactic single-dose irradiation and to compare the own results with already existing risk prediction models. METHODS AND MATERIALS: Computed tomography (CT) or magnetic-resonance (MR) images, and clinical data of 133 consecutive patients treated with linear accelerator radiosurgery were analyzed retrospectively. Using the Cox proportional hazards model the relevance of treatment parameters and dose-volume relationships on the occurrence of radiation-induced tissue changes (edema, localized blood-brain barrier breakdown) were assessed. RESULTS: Sixty-two intraparenchymal lesions (arteriovenous malformation (AVM): 56 patients, meningioma: 6 patients) and 73 skull base tumors were selected for analysis. The median follow-up was 28.1 months (range: 9.0-58.9 months). Radiation-induced tissue changes (32 out of 135, 23.7%) were documented on CT or MR images 3.6-58.7 months after radiosurgery (median time: 17.8 months). The actuarial risk at 2 years for the development of neuroradiological changes was 25.8% for all evaluated patients, 38.4% for intraparenchymal lesions, and 14.6% for skull base tumors. The coefficient: total volume recieving a minimum dose of 10 Gy (VTREAT10) reached statistical significance in a Cox proportional hazards model calculated for all patients, intraparenchymal lesions, and AVMs. In skull base tumors, the volume of normal brain tissue covered by the 10 Gy isodose line (VBRAIN10) was the only significant variable. CONCLUSIONS: These results demonstrate the particular vulnerability of normal brain tissue to single dose irradiation. Optimal conformation of the therapeutic isodose line to the 3D configuration of the target volume may help to reduce side effects.     

基于监督学习的前列腺MR/TRUS图像分割和配准方法

前列腺核磁超声图像配准融合有助于实现前列腺肿瘤的靶向穿刺。传统的配准方法主要是针对手动分割的前列腺核磁（Magnetic resonance, MR）和经直肠超声（Trans-rectal ultrasound, TRUS）图像上对应的生理特征点作为参考点,进行刚体或非刚体配准。针对超声图像因成像质量低导致手动分割配准效率低下的问题,提出一种基于监督学习的前列腺MR/TRUS图像自动分割方法,与术前核磁图像进行非刚体配准。首先,针对图像分割任务训练前列腺超声图像的活动表观模型（Active appearance model, AAM）,并基于随机森林建立边界驱动的数学模型,实现超声图像自动分割。接着,提取术前分割的核磁图像与自动分割的超声图像建立轮廓的形状特征矢量,进行特征匹配与图像配准。实验结果表明,本文方法能准确实现前列腺超声图像自动分割与配准融合,9组配准结果的戴斯相似性系数（Dice similarity coefficient, DSC）均大于0.98,同时尿道口处特征点的平均定位精度达1.64 mm,相比传统方法具有更高的配准精度。      

Registration of MR and SPECT without using external fiducial markers

The aim of our work is to present, test and validate an automated registration method used for matching brain SPECT scans with corresponding MR scans. The method was applied on a data set consisting of ten brain IDEX SPECT scans and ten T1- and T2-weighted MR scans of the same subjects. Of two subjects a CT scan was also made. (Semi-) automated algorithms were used to extract the brain from the MR, CT and SPECT images. Next, a surface registration technique called chamfer matching was used to match the segmented brains. A perturbation study was performed to determine the sensitivity of the matching results to the choice of the starting values. Furthermore, the SPECT segmentation threshold was varied to study its effect on the resulting parameters and a comparison between the use of MR T1- and T2-weighted images was made. Finally, the two sets of CT scans were used to estimate the accuracy by matching MR to CT and comparing the MR-SPECT match to the SPECT-CT match. The perturbation study showed that for initial perturbations up to 6 cm the algorithm fails in less than 4% of the cases. A variation of the SPECT segmentation threshold over a realistic range (25%) caused an average variation in the optimal match of 0.28 cm vector length. When T2 is used instead of T1 the stability of the algorithm is comparable but the results are less realistic due the large deformations. Finally, a comparison of the direct SPECT-MR match and the indirect match with CT as intermediate yields a discrepancy of 0.4 cm vector length. We conclude that the accuracy of our automatic matching algorithm for SPECT and MR, in which no external markers were used, is comparable to the accuracies reported in the literature for non-automatic methods or methods based on external markers. The proposed method is efficient and insensitive to small variations in SPECT segmentation.     

Reactive oxygen species act at both TGF-beta-dependent and -independent steps during induction of apoptosis of transformed cells by normal cells

PURPOSE: In clinical practice, physicians are constantly comparing multiple images taken at various times during the patient's treatment course. One goal of such a comparison is to accurately define the gross tumor volume (GTV). The introduction of three-dimensional treatment planning has greatly enhanced the ability to define the GTV, but there are times when the GTV is not visible on the treatment-planning computed tomography (CT) scan. We have modified our treatment-planning software to allow for interactive display of multiple, registered images that enhance the physician's ability to accurately determine the GTV. METHODS AND MATERIALS: Images are registered using interactive tools developed at the University of North Carolina at Chapel Hill (UNC). Automated methods are also available. Images registered with the treatment-planning CT scan are digitized from film. After a physician has approved the registration, the registered images are made available to the treatment-planning software. Structures and volumes of interest are contoured on all images. In the beam's eye view, wire loop representations of these structures can be visualized from all image types simultaneously. Each registered image can be seamlessly viewed during the treatment-planning process, and all contours from all image types can be seen on any registered image. A beam may, therefore, be designed based on any contour. RESULTS: Nineteen patients have been planned and treated using multimodality imaging from November 1993 through August 1994. All registered images were digitized from film, and many were from outside institutions. Brain has been the most common site (12), but the techniques of registration and image display have also been used for the thorax (4), abdomen (2), and extremity (1). The registered image has been an magnetic resonance (MR) scan in 15 cases and a diagnostic CT scan in 5 cases. In one case, sequential MRs, one before treatment and another after 30 Gy, were used to plan patient's initial fields and boost, respectively. Case illustrations are shown. CONCLUSIONS: We have successfully integrated multimodality imaging into our treatment-planning system, and its routine use is increasing. Multimodality imaging holds out the promise of improving treatment planning accuracy and, thus, takes maximum advantage of three dimensional treatment planning systems.