WO2012020211A1

WO2012020211A1 - Method and system for parameter determination from medical imagery

Info

Publication number: WO2012020211A1
Application number: PCT/GB2011/001155
Authority: WO
Inventors: Ajay Shah; Andrea Protti
Original assignee: King's College London
Priority date: 2010-08-12
Filing date: 2011-07-29
Publication date: 2012-02-16
Also published as: GB201013580D0

Abstract

Embodiments of the present invention provide an improved method to allow for volumes and other parameters relating to internal features and artefacts to be determined from medical imagery such as MRI images. In particular, within one embodiment of the invention a cluster-based segmentation algorithm is used that can allow voxels representing an internal feature which it is desired to measure to be segmented within multiple MRI images in parallel, and for measurements based on the segmentation thus obtained to be automatically generated. For example, in one embodiment left ventricle volume throughout the cardiac cycle can be determined by processing in parallel multiple MRI slices across multiple frames captured throughout the cycle, whereby cardiac measures such as ejection fraction, stroke volume, EDV, ESV and cardiac output may then be found quickly and easily. By determining the segmentation threshold for a single slice in advance of then applying the threshold across the whole image space, it can be certain that a threshold is used that takes into account the particular image data characteristics, such as contrast between and blood and muscle, and image brightness.

Description

Method and System for Parameter Determination from Medical Imagery

Technical Field The present invention relates to a method and system that is capable of determining the parameters of internal features of an object from medical imagery thereof. For example, one embodiment of the invention relates to a method and system that is able to determine various cardiological measures from MRI imagery. Background to the Invention

Computer software analysis has become essential in many research fields. Medical research, in particular, relies on computer programs to study the large image datasets provided by new medical equipment, such as Magnetic Resonance Imaging (MRI), in order to facilitate and improve diagnosis or determine viable treatments. Hence, the progression towards automated or semi-automated investigation tools plays a central role in enhancing accuracy and in the reduction of analysis time.

Cardiac studies are nowadays at the cutting-edge of medical research where images play a key role in diagnosing diseases or in evaluating treatment outcomes. Moreover, MRI is one of the state-of-the-art techniques most used in clinical and preclinical cardiac studies. However, the analysis of MRI datasets is far from being fully automated. In addition, current segmentation techniques (analysis of regional function or properties) do not work robustly when image artefacts affect picture quality, involving a rather time-consuming manual intervention.

MRI datasets are arranged in frames, and slices. In particular, a "frame" is a set of image slices of an object obtained at a particular point in time, whereas a particular "slice" within a frame is a particular spatial section through the object being imaged at the time the frame was captured. Figures 1 and 2 illustrate the concept of frames and slices in more detail. In Figure 1 , an object 1 at time tl is imaged by taking multiple spatial slices 11 therethrough. Each slice, 1 1a, l ib, 1 1c, Hi, therefore represents a cross-section through the object at the slice position. Figure 2 shows object 1 at time t2, where object 1 in this case has changed in size. Image slices 12 are taken through object 1 , to provide individual slices 12a, 12b, 12i, each representing a spatial cross-section through the object. Importantly, each slice in fact represents the MR response across a finite volume, which means that each pixel making up a slice image in fact represents a three-dimensional volume in the object; hence in MRI terms pixels are often referred to as "voxels" in that they represent respective volumes in the objects being imaged.

Prior Art

As noted above, it has been known to employ software tools to try and process cardiac images, and three such tools are known. The first program is called CMRTools, and its plug- in LV tools, described at http://www.cmrtools.com/cmrTools/index.php ?m=9. CMRtools can be run on PC or Linux-based workstations and is the most widely used analysis package for clinical MR images. Although very robust, it presents several pitfalls: 1) CMRTools can only be employed on DICOM files produced by dedicated clinical MR scanners such as Philips;

2) computational time. Semi-automated border detection is implemented, but a considerable amount of time is spent to manually define or improve segmentation for an accurate estimation of EF, EDV or ESV;

3) license costs. Several research groups may not be able to buy a license, which is very expensive (several thousands of pounds).

The second computer software is called Segment, which is available to download from http://segment.heiberg.se/. It runs on PCs only, it uploads DICOM files of a specific format, therefore the header information usually has to comply with the software specifications. It relies on a semi-automated segmentation and when compared to CMRTools, for the calculation of EF and volumes, it is undoubtedly faster. However, non-expert users might find it difficult to work with due to the numerous analysis options available where only few of them are crucial for segmentation.

The last software is OsiriX, available to download from http://www.osiriximaging.com resources/ . OsiriX runs on Mac workstations, is highly interactive, includes a good database management and is easy to use. However, detection of borders must be handled manually and, although a plug-in is well-suited for EF estimation, the corresponding analysis time is too long to be applied to a large number of images.

In addition, US 6,438,403 describes an imaging system that enables cardiac functioning within a particular cardiac chamber to be imaged. The system acquires imaging data that includes intensity values for four-dimensional voxels within a region of interest (ROI). A seed voxel is identified, and neighbor voxels to the seed voxel are also identified. The intensity values for each neighbor voxel are compared to a threshold to determine whether the voxel corresponds to blood or muscle tissue. For each neighbor voxel corresponding to blood, its neighbor voxels are identified and compared to the threshold, and this process is repeated until a pre-established spatial boundary is encountered or the number of new neighbor voxels indicates that processing is migrating into an adjacent cardiac chamber. At that point, the identified blood voxels are counted into bins of cardiac phases, cardiac images for each phase are reconstructed, and parameters such as ejection fraction are calculated.

The threshold value used in US 6,438,403 may be selected manually by an operator, or may be selected to be an average of the intensity of blood and muscle intensity values. However the threshold is selected, it is selected in advance of any image segmentation being performed. In particular, the selected threshold is applied to perform segmentation across the whole region of interest, in four dimensions, irrespective of whether the selected threshold actually achieves accurate segmentation. That is, the threshold is applied from a seed voxel with position (x, y, z, t) to eight neighboring voxels at positions (x ± 1 , y ± 1, z ± 1 , t ± 1) to segment the whole data set, without first determining whether the threshold is correct. Where the threshold is not correct, then inaccurate segmentation of blood from muscle will be obtained across the image space.

Summary of the Invention

Embodiments of the present invention provide an improved method to allow for volumes and other parameters relating to internal features and artefacts to be determined from medical imagery such as MRI images. In particular, within one embodiment of the invention a cluster- based segmentation algorithm is used that can allow voxels representing an internal feature which it is desired to measure to be segmented within multiple MRI images in parallel, and for measurements based on the segmentation thus obtained to be automatically generated. For example, in one embodiment left ventricle volume throughout the cardiac cycle can be determined by processing in parallel multiple MRI slices across multiple frames captured throughout the cycle, whereby cardiac measures such as ejection fraction, stroke volume, end-diastolic volume, end-systolic volume, and cardiac output may then be found quickly and easily. In particular, a segmentation threshold is first found that gives accurate segmentation of cardiac chambers in a single MRI slice image. This threshold value is then used to perform cluster based segmentation of other slices of the same frame, as well as other slices in different frames, to obtain segmentation of a cardiac chamber across the whole cardiac volume and cycle. By determining the segmentation threshold for a single slice in advance of then applying the threshold across the whole image space, it can be certain that a threshold is used that takes into account the particular image data characteristics, such as contrast between and blood and muscle, and image brightness.

In view of the above, from a first aspect there is provided a method of determining one or more parameters of features of an imaged object from image data thereof, the method comprising: receiving an image data set comprising one or more image frames, each frame having one or more image slices of known spatial resolution, each slice representing a spatial slice through the imaged object of known thickness; determining a start pixel position in a first image slice containing image data relating to a feature at least one parameter of which is to be determined, the pixel position being located within the part of the image slice relating to the feature; forming a cluster of segmented pixels extending from the determined start pixel position by comparing pixel values to a threshold condition and including pixels in the cluster if the threshold condition is met, wherein the threshold condition is selected such that the cluster extends to the boundaries of the feature the parameters of which are to be determined; counting the number of segmented pixels in the cluster; calculating at least one parameter of the feature in dependence on the number of segmented pixels in the cluster and the known spatial resolution and slice thickness; and outputting the calculated parameter to a user.

In one embodiment the cluster formation comprises a) displaying the segmented pixels on the image slice to a user; b) receiving an input from the user to alter the threshold condition, c) re-forming the cluster of segmented pixels from the start pixel position using the altered threshold condition; and d) repeating a), b) and c) as necessary on command of the user until the threshold condition is obtained such that the cluster extends to the boundaries of the feature the parameters of which are to be determined. In one embodiment the threshold condition is that the pixel values are greater or less than a single threshold value. In one embodiment the threshold condition is that the pixel values are between an upper threshold value and a lower threshold value.

In one embodiment the method further comprises, for other image slices in the same frame as the first image slice, forming respective clusters of segmented pixels in each slice, in each case extending from a pixel in a corresponding position in the slice as the determined start pixel position in the first image slice, the respective clusters being formed by comparing pixel values to the threshold condition determined for the first image slice and including pixels in the cluster if the threshold condition is met; counting the number of segmented pixels in the respective clusters; calculating the at least one parameter of the feature for the frame in dependence on the number of segmented pixels in all the respective clusters and the known spatial resolution and slice thickness; and outputting the calculated parameter for the frame to a user.

In one embodiment the method further comprises, for multiple other slices in multiple other frames, forming respective clusters of segmented pixels in each slice, in each case extending from a pixel in a corresponding position in the slice as the determined start pixel position in the first image slice, the respective clusters being formed by comparing pixel values to the threshold condition determined for the first image slice and including pixels in the cluster if the threshold condition is met; the method further comprising, for a respective frame counting the number of segmented pixels in the respective clusters in the slices of the frame; calculating the at least one parameter of the feature for the frame in dependence on the number of segmented pixels in all the respective clusters and the known spatial resolution and slice thickness; and outputting the calculated parameter for the frame to a user. Within the above embodiment the method preferably further determines the maximum parameter value and the minimum parameter value, and the corresponding frames. Moreover, the method may further comprise calculating one or more metrics values in dependence on the maximum and minimum parameter values. In one preferred embodiment the image data set is a MRI image data set, and preferably the imaged object is a heart, and the feature to be parameterised is the left or right ventricle, wherein the parameter determined is the volume of the left or right ventricle. Within this embodiment the maximum parameter value may be the end diastolic volume and the minimum parameter value may be the end systolic volume, wherein the metrics calculated are one or more from the group comprising: ejection fraction, end-diastolic volume, end-systolic volume, stroke volume and cardiac output.

From another aspect there is further provided a computer program or suite of computer programs so arranged such that when executed by a computer it/they cause the computer to operate in accordance with the method of the first aspect. A further aspect also provides a computer readable storage medium storing the computer program or at least one of the suite of computer programs. Another aspect of the invention presents a system for determining one or more parameters of features of an imaged object from image data thereof, the system comprising: at least one processor; and at least one memory including computer program code; the at least one memory and the computer program code configured to, with the at least one processor, cause the apparatus to perform at least the following: receive an image data set comprising one or more image frames, each frame having one or more image slices of known spatial resolution, each slice representing a spatial slice through the imaged object of known thickness; determine a start pixel position in a first image slice containing image data relating to a feature at least one parameter of which is to be determined, the pixel position being located within the part of the image slice relating to the feature; form a cluster of segmented pixels extending from the determined start pixel position by comparing pixel values to a threshold condition and including pixels in the cluster if the threshold condition is met, wherein the threshold condition is selected such that the cluster extends to the boundaries of the feature the parameters of which are to be determined; count the number of segmented pixels in the cluster; calculate at least one parameter of the feature in dependence on the number of segmented pixels in the cluster and the known spatial resolution and slice thickness; and output the calculated parameter to a user.

Another aspect of the invention provides a method of determining one or more myocardial parameters from an MRI data set containing cardiac MRI imagery arranged in frames and slices, the method comprising: displaying an MRI image slice to a user; from a pixel position in the image slice identified as being within a cardiac feature such as the left or right ventricle, using a clustering algorithm to segment those pixels in the image within the cardiac feature and determining a clustering threshold that achieves the segmentation; from the same pixel position in other MRI image slices of the same or other frames, using the clustering algorithm to segment those pixels in the image slices within the cardiac feature using the determined clustering threshold; counting the segmented pixels in the respective clusters in the respective slices whereby to determine frame cardiac feature volume parameters; and calculating the one or more myocardial parameters from the determined frame volume parameters.

Within the above, the myocardial parameters may include one or more selected from the group comprising ejection fraction end-diastolic volume, end-systolic volume, stroke volume and cardiac output.

A further aspect of the invention also provides a system for determining one or more myocardial parameters from an MRI data set containing cardiac MRI imagery arranged in frames and slices, the system comprising: at least one processor; and at least one memory including computer program code; the at least one memory and the computer program code configured to, with the at least one processor, cause the apparatus to perform at least the following: display an MRI image slice to a user; from a pixel position in the image slice identified as being within a cardiac feature such as the left or right ventricle, use a clustering algorithm to segment those pixels in the image within the cardiac feature e and determine a clustering threshold that achieves the segmentation; from the same pixel position in other MRI image slices of the same or other frames, use the clustering algorithm to segment those pixels in the image slices within the cardiac feature using the determined clustering threshold; count the segmented pixels in the respective clusters in the respective slices whereby to determine frame cardiac feature volume parameters; and calculate the one or more myocardial parameters from the determined frame volume parameters.

Within the above, the myocardial parameters may include one or more selected from the group comprising ejection fraction, LVEDV, LVESV, stroke volume and cardiac output.

Brief Description of the Drawings Further features and advantages of the present invention will become apparent from the following description of embodiments thereof, presented by way of example only, and by reference to the accompanying drawings, wherein like reference numerals refer to like parts, and wherein: -

Figure 1 is a diagram illustrating how images are arranged in frames and slices in an MRI data set; Figure 2 is also a diagram illustrating how images are arranged in frames and slices in an MRI data set;

Figure 3 is a block diagram of a system according to a first embodiment of the present invention;

Figure 4 is a flow diagram illustrating the operation of the system according to the first embodiment of the invention;

Figure 5 is a flow diagram illustrating the operation of part of the system of the first embodiment;

Figure 6 is a flow diagram illustrating the operation of part of the system of the first embodiment; Figure 7 is an example MRI slice illustrating how a first clustering segmentation method is used in the first embodiment;

Figure 8 is an MRI slice illustrating the operation of the first clustering segmentation method; Figure 9 is an MRI slice illustrating the operation of a second segmentation method used in the first embodiment;

Figure 10 is an MRI slice illustrating the operation of the second segmentation method; Figure 1 1 is a flow diagram illustrating the operation of part of the system of the first embodiment;

Figure 12 is a flow diagram illustrating the operation of part of the system of the first embodiment;

Figure 13 is a screen shot comprising a plurality of MRI slices illustrating how segmentation can be applied in parallel to multiple slices; Figure 14 is a screen shot showing multiple MRI slices and how segmentation can be applied in parallel thereto;

Figure 15 is a screenshot showing multiple MRI slices from the same frame; Figure 16 is a screenshot showing multiple MRI slices and how the second method of clustering may be applied in parallel thereto;

Figure 17, 18, and 19 are a series of screenshots showing a single MRI slice illustrating how a third segmentation method may be applied in the first embodiment;

Figure 20 is a flow diagram illustrating how various cardiac related metrics may be calculated from the segmentations that are obtained.

Figure 21 is a screen shot illustrating how the first and third methods of clustering may be applied to the same MRI slice;

Figure 22 is a screenshot illustrating the various cardiac metrics that may be calculated from the segmentations; and Figure 23 is a flow diagram illustrating the operation of part of the system of the first embodiment.

Description of Embodiments Overview

Embodiments of the invention are predominantly software based, and built with the objective of being able to upload any DICOM file and to effectively compute functional cardiac parameters. In particular, some embodiments of the invention substitute fully-automated border detection algorithms with the integration of semi-automated and optimised manual segmentation operations.

Preferred embodiments of the invention are able to upload DICOM files from various MRI manufacturers, bypassing header information. Any file is preferably associated with the corresponding frame-slice. This correspondence can optionally be changed by the user whenever appropriate by specifying the total number of slices (the number of frames is the number of dynamic images per slice). Once the data have been uploaded, a slice is typically displayed but some embodiments of the invention permit the visualisation of 2x2, 3x3 or 4x4 (or more) images simultaneously; these can pertain to one slice (and in this case the rendering mode is called herein "multi-frame") or to one frame ( referred to herein as "multi-slice). In preferred embodiments the location of the current working frame and slice is visualized in real-time. These working modalities and the ones presented below are preferably managed through a mouse-based menu and/or keyboard buttons; for example, image groups can be changed with the keyboard arrows.

Embodiments of the invention aim to overcome some of the problems that other segmentation software present, by exploiting threshold based techniques through which clusters are generated to agglomerate pixels characterised by similar intensities. Any thresholds used are preferably displayed in real time and can be adjusted, for example using mouse or keyboard inputs. For instance, in a cardiac imaging embodiment, if a cluster is created within the left ventricle (LV) blood pool, embodiments of the invention are able to detect the endocardial wall as a border using clustering from within the pixels representing the LV blood pool. In contrast prior art software will usually adopt an algorithm to identify the borders directly, which is prone to artefacts.

A particular aspect of some embodiments of the invention, which enhances user's interactivity, is the ability to perform feature editing and clustering by means of three different modalities, i.e. single-image, multi-image and full-dataset. In particular, the algorithms related to the last two modes have been optimized to make execution and analysis interactive even whilst operating with several tens of high temporally/spatially resolved images concurrently. The first modality allows the work on a single image where all the actions will only applied to the current image. The multi-image modality permits the operation on multiple images simultaneously. If the current dataset is visualized in multi- frame mode, in some embodiments the user can intervene on all frames corresponding to the slice of choice. Similarly, in embodiments providing a multi-slice operational setting, the user can operate on all slices of the frame of choice. Lastly, in embodiments providing the aforementioned full-dataset option, any action taken on one image is applied to every image. For accuracy purposes in any of these modalities, a cluster can be locally improved whenever it becomes inadequate during the aforementioned semi-automated process, by adding or deleting single pixels manually, and some embodiments provide a manual editing mode to achieve such functionality. More particularly, a so-called "whole volume" mode can be implemented using the full data set, wherein multiple slices in multiple frames can have clustering performed on them at the same time, effectively in parallel, from the view point of the user. In such a whole volume mode a start pixel selected by user as a start pixel for a clustering process in one slice is used as the start pixel in all the slices processed. Likewise, a clustering threshold used in the first slice is automatically applied to the other slices processed. In this way, segmentation of corresponding features at substantially the same spatial location in the multiple slices can be performed automatically all at once. This also allows for whole volumes of features to be found quickly and automatically, as will be described in more detail later. Within embodiments of the invention, as will be described in more detail, clusters are created using threshold-based techniques, and in some cases within a user-selected intensity window. A cluster is typically generated by pointing at a start pixel to be the beginning of the cluster, clicking the left mouse button, and then modifying the cluster threshold by moving the mouse up and down. Cluster thresholds are displayed in real time on the display. When clusters are to be produced on multiple images concurrently, rapid and accurate left or right ventricle endocardial border detection becomes a very simple, trivial task. For accuracy purposes a cluster can be locally improved whenever it becomes inadequate during the aforementioned semi-automated process; this can be accomplished by adding or deleting single pixels or groups of pixels, depending on the cursor diameter, which can be adjusted as desired. Some embodiments of the invention allow the user to place clusters in multiple parts of an image whenever this is requested. Such action might be very useful for instance in the simultaneous segmentation of the left and right ventricles. The clusters preferably appear in different colours depending on the current editing options. Color-wise updating and 'priority' rules have been introduced and designed so as to optimise user needs. Specifically, the development process and capabilities of embodiments of the invention have been tuned through the intense and continuous collaboration between an expert programmer and an active user.

A first type of clustering segmentation used in embodiments is created over image pixels which present a signal intensity higher than the applied threshold; whenever the cluster encounters areas of lower intensity, its propagation comes to an end. In cardiac imagery this method is ideally applied in cavities such as the LV blood pool where blood presents higher signal intensity compared to the myocardium. Pixels that are includes in such a cluster are coloured with a predetermined colour, such as red.

A second type of clustering segmentation used in embodiments is created within an area limited by a used-defined line and it can expand within a threshold window whose upper and lower limits can be modified as desired. The generated clusters will include pixels within a signal intensity range. Clusters segmented using such a process are given a second predetermined colour, such as orange.

A third clustering segmentation method that may be employed in embodiments of the invention is characterised by a third colour (e.g. blue) and is threshold free; hence a blue cluster can be created anywhere in the image. A blue cluster however must be generated within a limited area drawn manually by single blue pixels.

Embodiments of the invention are preferably inclusive of a wide option menu. The main options preferably allow the user to rotate images, to change visual parameters such as opacity and contrast and perform X-ray rendering of a time instance interactively. It should also be possible in some embodiments to exclude unwanted images whenever they do not have to take part in the segmentation process. However, the same command can be used to re-include images previously ticked off. In preferred embodiments LV functional information and all area-related measurements, for volumetric calculation, are written to file for offline analysis by the activation of a save command. Once activated, the ejection fraction (EF) value will be displayed on the image window. In addition, areas and volumes of different clusters, therefore different colours, are also preferably displayed in real-time, Also the end-diastolic and end-systolic image frame are displayed in real time. This is important in cardiac images because, once such frames have been identified, all the others can be excluded leaving the user to optimize clusters only in these two frames.

First Embodiment

A first, preferred, embodiment will now be described with respect to Figures 3 to 22. The first embodiment is particularly although not exclusively adapted for processing MRJ data sets containing cardiac imagery, and in particular the first embodiment provides for the automatic calculation of various cardiac functional and volumetric parameters, such as stroke volume (SV), ejection fraction (EF), and cardiac output (CO).

Figure 3 is a block diagram illustrating an arrangement of a system according to the first embodiment. More particularly, as noted previously, embodiments of the present invention are predominantly software implemented, and are designed to run on general purpose desktop or laptop computers. Therefore, according to the first embodiment, a computing apparatus 30 is provided having a central processing unit 306, and random access memory (RAM) 304 into which data, program instructions, and the like can be stored and accessed by the CPU. The apparatus 30 is provided with a display screen 32, and input peripherals in the form of a keyboard 34, and mouse 36. Keyboard 34, and mouse 36 communicate with the apparatus 30 via a peripheral input interface 308. Similarly, a display controller 302 is provided to control display 30, so as to cause it to display images under the control of CPU 306. Image data sets, such as MRI data sets 38, can be input into the apparatus and stored via image data input 310. In this respect, apparatus 30 comprises a computer readable storage medium 312, such as a hard disk drive, writable CD or DVD drive, zip drive, solid state drive or the like, upon which image data 322 corresponding to the MRI data sets input can be stored. Computer readable storage medium 312 also stores various programs, which when executed by the CPU 306 cause the apparatus 30 to operate in accordance with the first embodiment. In particular, a control interface program 318 is provided, which when executed by the CPU 306 provides overall control of the computing apparatus, and in particular provides a graphical interface on the display 32, and accepts user inputs using the keyboard 34 and mouse 36 by the peripheral interface 308. The control interface program 318 also calls, when necessary, other programs to perform specific processing actions when required. In particular, single threshold clustering segmentation program 314 is provided which is able to operate on image data indicated by the control interface program 318, so as to perform a clustering segmentation operation using a single threshold. Similarly, double threshold clustering segmentation program 316 is also provided, which, under control of the control interface program 318, operates on image data passed thereto so as to apply a clustering segmentation using a double threshold clustering operation. The operations of the single threshold clustering segmentation program 314 and the double threshold clustering segmentation program 316 will be described in more detail later. Additionally provided is manual segmentation program 324, which, when called by the control interface program 318, allows a user to manually edit image data passed to it, so as to segment or de-segment pixels.

Additionally, within the particular embodiment, which is adapted to automatically produce myocardium functional and volumetric parameters, a volume calculator program 320 is provided, which can use the segmentation information provided from the clustering programs 314 and 316, so as to calculate various cardiac related functional and volumetric parameters.

The detailed operation of the computing apparatus 30 will now be described with respect to Figure 4.

Firstly, the user launches the control interface program 318. The control interface program 318 is loaded into RAM 304, and is executed by the CPU 306. The control interface program controls the CPU to cause an input box to be displayed, wherein the user must indicate which MRI data set of the image data 322 stored on the computer readable medium 312 is to be processed. In addition, parameters of the data set must also be included, including the slice thickness (in millimeters), spatial resolution of the image (in millimeters), measured heart rate during the MRI image capture (in beats per minute), and the number of slices per frame. In addition, a file path wherein a text result file containing the parameters to be found should be indicated. Having indicated this to the control interface program 318, at block 4.2 the control interface program 318 loads the indicated MRI data set comprising multiple frames, and multiple slices per frame, having the spatial resolution, and slice thickness indicated. The spatial resolution and slice thickness is required in order to know the volume that each pixel (voxel) in an image slice represents. As noted previously, the MRI data sets loaded are typically DICOM data sets.

A user of the system will typically want to know one or more of several parameters from the image data, such as, for example, end diastolic volume (EDV) and end systolic volume (ESV) and hence stroke volume, as well as ejection fraction, and cardiac output. In addition, myocardium left ventricle wall volume may also be desired. In order to calculate these various parameters the following processing is undertaken.

At block 4.4, a first image is selected to start working on, and this is displayed in a window on display 32. Various image processing operations, such as brightening or contrast processing operations may be undertaken, so as to brighten the images, and improve the image contrast. At block 4.6, having undertaken this optional image pre-processing to improve brightness and contrast, a first image is selected to be displayed. For example, the user may select an image slice taken from around the middle of the stack of slices of a frame, with the slice being taken from a frame generally in the middle of the cardiac cycle.

Depending on what the user is desiring to find, a particular cluster mode is then selected at block 4.8. As will be described, this cluster mode may be a single threshold cluster mode performed by the single threshold clustering segmentation program 314, or a double threshold cluster mode, performed by the double threshold clustering segmentation program 316. Alternatively, a manual cluster mode may be selected, so as to manually segment features in the displayed image.

For the purposes of the present description, let us assume first that at block 4.8 the user first selects a single threshold cluster based segmentation to be performed. In preferred embodiments of the invention this would typically be used to perform left (or right) ventricle blood pool segmentation. Thus, at block 4.10, the user will have indicated, using the control interface program 318, that he wishes to perform a single threshold cluster based segmentation operation. By so doing the control interface program expects the user to point to a start pixel using the mouse 36, and using the indicated pixel a single threshold cluster based segmentation is performed from the indicated start pixel, as will be described.

At block 4.12, therefore, the user indicates a start pixel, which is to be the pixel from which clustering is to commence, based on an initial threshold. Single threshold segmentation is then performed at block 4.14, and shown in detail in Figures 5, 7 and 8.

More particularly, with respect to Figures 5 and 7, Figure 7 shows a screenshot illustrating a single M I slice. In order to perform single threshold clustering, at block 5.2 an initial clustering threshold is set, and this can be seen in the upper left corner of Figure 7, noted as the "red threshold". This is a Scalar value, relating to pixel intensity. In order to perform the single threshold clustering, as mentioned the user points with the mouse pointer at a start pixel in the image from which clustering should proceed. In this respect, in order to segment the left ventricle blood pool, for example, a user should select a pixel generally in the centre of the left ventricle, such as at position 72 in Figure 7. Once selected, at block 5.4 a cluster of connected pixels to the selected pixel each of which have an intensity higher than the set threshold is formed. That is, from the start pixel indicated by the user, the intensity of adjacent pixels to the start pixel is examined, and if the intensity of those pixels is greater than the threshold presently set then those pixels are included in the cluster. The cluster operation then iterates for each of the new pixels included in the cluster, so as to look at the adjacent pixels to those pixels, and examine the intensities of those adjacent pixels to see if the intensities are greater than the set threshold. This clustering procedure continues to iterate so as to examine adjacent pixels to every pixel included in the cluster, whereby pixels continue to be added to the cluster when their intensity is greater than the set threshold. This results in the cluster propagating from the start pixel across the image until it reaches boundary pixels where adjacent pixels are less than the set threshold. Figure 7 illustrates a cluster around start pixel 72 which has propagated so as to encompass a large proportion of the volume of the left ventricle, but not all of the left ventricle volume. In the context of Figure 5, image 70 would be produced as a result of block 5.6, wherein the clustered pixels are displayed as coloured pixels overlaid on the image displayed on the screen.

As shown in Figure 7, the cluster based segmentation using the threshold shown in Figure 7 has not completely segmented the left ventricle blood pool. Therefore, recognising this, the user would use a mouse input to change the threshold, as shown at block 5.8. As discussed previously, this could be, for example, by holding the left mouse button down and moving the mouse up or down so as to change the threshold value used for segmentation. As the threshold is changed, the clustering algorithm iterates so as to re-cluster from the start pixel based on the changed threshold. Figure 8 illustrates an example of this, wherein the same image as Figure 7 is shown, but here the threshold has been reduced (the "red threshold" shown in the upper left hand corner of the screenshot has been reduced from 1230 to 765). With this lower threshold the clustering algorithm should propagate further, and as shown in Figure 8 in fact propagates throughout the left ventricle to the boundary of the left ventricle wall. In this respect, as mentioned, as the threshold is changed the clustering algorithm iterates as shown in Figure 5, such that the display is updated as the threshold is changed. By so doing, the user can change the threshold until an accurate segmentation of the left ventricle is displayed. At that point, the segmentation is complete.

Determining the threshold that achieves accurate segmentation of a cardiac chamber on a single image slice as described above provides advantages in that a threshold can be determined that is specific to the particular image data set, and which is able to accurately segment blood from muscle in that particular data set. In this respect, depending on MRJ scanner settings and the patient being imaged, image data sets with different image properties, such as contrast and brightness, can be obtained. That is, across different image data sets a different voxel value may represent blood or muscle, such that there is no universal segmentation threshold that may reliably be used across multiple data sets. In embodiments of the present invention this problem is addressed by performing a user based segmentation on a single image slice in a data set, the user varying the threshold and observing the resulting cluster-based segmentation that is obtained, until a threshold value is reached that provides accurate segmentation of cardiac chambers for the particular data set. Because the image data in the other slices and frames of the data set were collected at the same time with the same MRI scanner settings and from the same patient, the threshold determined by the user for the single frame can then be reliably used in the other slices and frames of the same data set, and should produce accurate blood/muscle segmentation across all slices/frames of the data set.

Thus far, as shown in Figure 8 the left ventricle blood pool has been segmented in a single slice of a single frame. Because the control interface program knows the spatial resolution and slice thickness, it is able to attribute a particular volume to each pixel. Therefore, by counting the number of segmented pixels in the cluster, and multiplying by the known pixel volume, a total volume for the segmented cluster can be found for the slice, and this is shown at the bottom left hand corner of Figure 8 as the "v red" figure. At this point because only a single slice has had the cluster based segmentation performed thereon, a second volume measure, being "v_tot" is also shown, which corresponds to the v_red figure. However, once other slices have been segmented, as will be described, then the v_tot figure will in fact equal the total interior volume of the left ventricle for the present MRI frame.

Returning to Figure 4, having performed the cluster based segmentation on a first slice, it is then possible to carry over the cluster based segmentation to: 1) the other slices in the same frame; 2) the corresponding slice in other frames; and then 3) multiple slices in mulitple frames, as shown at blocks 4.24, 4.26, and 4.28. This feature is described in more detail with respect to Figures 1 1 to 15, and 23. In particular, in order to apply the segmentation to other slices in the same frame, the user selects the multi-slice option, at block 1 1.2. This causes the various slices of the present frame being processed to be displayed at block 1 1.4. The cluster based segmentation is then automatically carried over onto the other slices using an iterative process that processes each slice image in turn and commences the clustering process in the same manner as described, at the corresponding pixel location in each slice image as the selected start pixel in the first slice that was processed. The same segmentation threshold (the "red threshold") is used as was eventually set in respect of the first slice image, and a result is obtained as shown in Figure 15 which displays example slices 1 to 12 of example frame 1, with the left ventricle blood pool completely segmented in each slice. The success of this procedure relies on the fact that the slices of any frame are spatially aligned with each other such that, for example, a pixel which is located at the centre of the left ventricle in one slice of a frame, should be also at about the centre of the left ventricle in other slices of the frame, and hence can be the start pixel for the clustering agglomeration of pixels using the single- threshold clustering algorithm. In addition, and as mentioned previously, the same threshold can also be used, as the threshold has effectively been "normalised" for the present data set by being specifically set, based on a single slice, to segment blood from muscle given the image properties such as contrast and brightness of the voxels in the data set.

In addition to performing the single threshold clustering on each slice of the same frame, it is also possible to apply the single threshold clustering that is applied in one slice in one frame over to the same slice in different frames, and this is shown in Figures 12 to 14. In particular, the user selects the multi-frame option at block 12.2, and this results in the corresponding slices of other frames in the data set being displayed at block 12.4. A processing iteration is then commenced such that each corresponding slice image is processed in turn, applying the clustering segmentation based on the threshold from the first image, at the corresponding pixel location in each slice to the selected start pixel location in the first slice processed. By then applying the same segmentation thresholds using the same clustering algorithm the clustering segmentation obtained in the first slice processed is applied to the corresponding slices in other frames. This is shown, for example, in Figures 13 and 14, wherein Figure 13 shows respective slices 6 of example frames 1 to 9, and Figure 14 shows respective slices 8 of example frames 1 to 9.

In addition to the above, it is also possible to undertake a "whole volume" approach that permits multiple slices across the volume in multiple frames to be processed at the same time, using the segmentation settings i.e. start pixel location and clustering threshold found for the first slice processed. Figure 23 illustrates this operation in more detail. Here, at block 23.2 a whole volume processing is selected. This starts a number of processing loops at blocks 23.4 and 23.6 to process each relevant slice in each relevant frame relating to the feature being characterised. That is, for each slice image of each frame, clustering is performed based on the same thresholds as previously, using the corresponding start pixel location in each slice image as the location selected by the user in the first image. The final outcome of such processing (performed in the context of Figure 4 in block 4.28) is that in each slice of each frame the left ventricle blood pool (for example, or any other three dimensional feature such as the right ventricle) can be segmented, which leads to the ability to automatically determine the volume of the left ventricle, and how that volume changes throughout the cardiac cycle. In particular, this function provides the ability to obtain the required parameters of the feature quickly, using, from the user's point of view, parallel clustering on all the relevant slices in the all the relevant frames. Further details as to how the functional and volumetric parameters in the cardiac embodiment can be calculated from the cluster based segmentations obtained will be described later.

Returning to Figure 4, thus far we have concentrated on how a single threshold clustering can be performed, for example to determine the left ventricle volume. However, embodiments of the present invention provide two other types of segmentation, and in particular using a double threshold, and also using manual segmentation. The double threshold cluster based segmentation will now be described.

In Figure 4, assume that at block 4.8 the user has selected that a double threshold cluster based segmentation be performed, at block 4.16. The first operation necessary is for the user to manually define a pixel boundary, at block 4.18. Screenshot 90 in Figure 9 shows how a manually defined pixel boundary 92 can be drawn on a slice, for example, using the mouse. This boundary defines the limit of any segmentation that is performed, in that only pixels within this boundary are considered.

Next, at block 4.20 the user indicates a start pixel, for example by pointing at the desired start pixel with a mouse. From this start pixel the intensity of pixels surrounding the start pixel are looked at, to determine whether they fall within upper and lower thresholds. As will be apparent in Figures 9 and 10, in the top left corner two thresholds referred to as "min orange threshold" and "max orange threshold" are shown, which are the two thresholds applied in the double threshold cluster based segmentation algorithm. The clustering proceeds by looking at adjacent pixels to the start pixel and determining whether the intensities of the pixels are within the two thresholds. If so, then each examined pixel which meets this criterion is included in the cluster. The clustering procedure then iterates to look at all surrounding pixels to pixels included in the cluster, to see if their intensities are within the ranges set. In this way, the cluster propagates from pixel to pixel with adjacent pixels being included within the cluster if their intensity is within the set ranges.

As with the single threshold clustering, a mouse input can be used to alter the maximum and minimum thresholds, and as the thresholds are changed the cluster obtained alters, and the displayed cluster is displayed on the screen. Therefore, the user can optimise the cluster obtained using mouse movements. For example, as shown in Figure 10 a cluster 102 corresponding to the myocardium wall can be formed within boundary 92. As with the single threshold clustering, the double threshold clustering can also be applied to multiple slices in the same frame and to corresponding slices in different frames, and then to multiple slices in multiple frames, using the same principles as described previously in respect of the single threshold clustering. That is, in other slices a corresponding start pixel at the same location in the slice as the selected start pixel in the initially processed slice is used as the start point for the clustering, and then the same thresholds as were determined in respect of the first slice are then applied. As noted, this operation can be applied to slices in the same frame as the first slice processed, as well as to corresponding slices in other frames. Once corresponding slices in other frames have been processed, it is then possible, within each frame, to extend the segmentation to each slice in each frame, using the same principles. As noted, this operation relies on the fact that in an MRI data set the slices are spatially aligned one on top of each other within a frame, and whilst frames are inherently time aligned generally they will also be spatially aligned provided the object being imaged does not move too much during the imaging period.

As with the single threshold segmentation, in the double threshold segmentation the same advantages regarding finding the two thresholds that give the desired segmentation prior to then applying the thresholds to other slices/frames are obtained as when using single threshold segmentation, that is that the thresholds are normalised on the single slice to the image characteristics, such as contrast and brightness of the images, and then set to give the desired segmentation of blood from muscle. The thresholds thus found can then be reliably applied to other slices in the same data set, and the same accurate segmentation should be obtained. A third type of segmentation that can be performed on the image is manual segmentation, shown at block 4.34 and 4.36 of Figure 4. Figures 17, 18 and 19 illustrate how manual segmentation may be performed by manually drawing on, using the mouse, borders between different image features. For example, mass calculation can be achieved by drawing the endocardial border 172, and the epicardial border 182 (see Figures 17 and 18). By then filling in the space between these two borders a blue cluster 192 can be obtained, in this case, corresponding to the myocardial wall. By counting the number of segmented pixels in the blue cluster an estimate of the volume and hence mass of the myocardial wall can be obtained. The present embodiment therefore provides multiple methods of being able to segment features particularly within cardiac MRI images in a semi automated fashion. However, one of the particular strengths of the present embodiment is the ability, particularly in relation to the single threshold based segmentation, to apply the known spatial alignment present in MRI slices and frames, as well as the same background image characteristics in each slice, to automatically apply the segmentation achieved on a single slice to other slices both in the same frame, and other frames in an automated manner. This allows segmentation of, for example, the left ventricle blood pool, to be automatically achieved across an MRI data set captured across the whole cardiac cycle. By so doing it then also becomes possible to automatically calculate myocardium functional and volumetric parameters, and to display these in real time on the segmented images. Figure 20 illustrates an example process as to how these parameters can be obtained.

The procedure of Figure 20, which will be performed by the volume calculator program 320 under the control of the control interface program 318, assumes that the single threshold clustering segmentation program 314 has been applied so as to segment the left ventricle (for example - other features such as the right ventricle may also be found) in each relevant slice of each frame. By each relevant slice we mean a slice which contains pixels representing part of the interior volume of the left ventricle. The segmentation on each slice may have been applied using any of the modes of operation described previously. Assuming that this segmentation is correct (and bearing in mind that all segmentations can be displayed to the user in the multi frame, multi slice, or whole volume modes, and manually cleaned up by selected and deselecting pixels), it becomes possible to calculate myocardial parameters such as the end diastolic volume (EDV), the end systolic volume (ESV), the stroke volume, the ejection fraction, and the cardiac output, by counting the number of segmented pixels in each slice in each frame, to determine the volume of the left ventricle for each frame. By then finding the maximum and minimum volumes the end diastolic volume (EDV) and end systolic volume (ESV) can be found respectively. From these figures the stroke volume, ejection fraction, and cardiac output can be found.

Turning to Figure 20, and assuming the above noted single threshold segmentations of the left ventricle, a processing loop is started at block 20.2 to process each frame in turn. A frame volume total variable frame _vol tot [x] is then initialised to zero at block 20.4, and then at block 20.6 a second processing loop is started to process each slice in each frame. For each slice in each frame the number of pixels segmented as being in the left ventricle (i.e. subjected to the single threshold segmentation) is counted. Because each pixel in fact represents a volume i.e. is a voxel, the volume of the left ventricle in the present slice being processed can be found, and this volume is added to the frame volume total variable frame _yol ot [x], at block 20.10. Thereafter the next slice in the present frame is processed, and the number of segmented voxels in the next slice counted, and added to the frame volume total frame oljotfxj, at blocks 20.8 and 20.10. This procedure repeats for every slice in the present frame being processed, with the result that at the end of this processing loop i.e. once every relevant slice in the present frame has been processed, the frame volume total variable frame vol Jot fx] should equal the volume of the left ventricle for the present frame. This variable frame vol ot [x] is then saved for the present frame at block 20.14, and then the next frame is selected at block 20.16, whereupon processing returns to block 20.4, and a frame volume total variable frame ol Jot [x] initialised for the next frame. The result of the processing loops of blocks 20.2 to 20.16 are that for each frame the number of pixels, and hence the volume of the left ventricle in each slice in each frame are summed, such that a frame >ol jot [x] value representing the total volume of the left ventricle in that particular frame is found for each frame. Once the total volumes for each frame have been found, at block 20.18 these values can be sorted, and at block 20.20 the maximum and minimum total frame volume values found. In this respect, the maximum frame volume corresponds to the end diastolic volume (EDV) and the minimum volume corresponds to the end systolic volume (ESV) of a cardiac cycle. Having found the end diastolic volume and end systolic volume (ESV) it then becomes possible to calculate the stroke volume, the ejection fraction, and the cardiac output, as follows: -

The equations related to the results shown in the text file are the following: Stroke volume (SV)

SV = End Diastolic Volume (EDV) - End Systolic Volume (ESV) 1) expressed in μί. Where EDV corresponds to 'Vmax' (or end-diastolic volume) and ESV corresponds to 'Vmin' (or end-systolic volume).

Ejection fraction (EF)

EF = SV/EDV * 100 2) expressed in %. Cardiac output (CO)

CO = SV * heart rate 3) expressed in mL/min. Here 'heart rate' was introduced as part of the information entered by the user to launch the control interface program.

Once these parameters have been calculated, they can then be displayed on the display screen 32. In addition, the various frame vol Jot [x] values as well as the calculated parameters are saved to a text file, as shown in Figure 22. In addition, once these parameters have been calculated, they can be displayed in real time with the image slices, as shown in Figure 21, wherein the ejection fraction (EF) is shown in the upper left hand corner of the screen, and the total volume of the left ventricle for the present frame is shown in the bottom left hand corner of the screen.

The present embodiment thus provides numerous advantages over the known prior art systems. In particular:

1) First of all, it can upload any kind of DICOM files bypassing header information. This gives a clear advantage in analysis accuracy whenever utilizing data from different scanners and significantly increases its applicability.

2) Secondly, the present embodiment is entirely designed to achieve a high level of interactivity between the software and the user. This is achieved by using the mouse to engage most of the commands. It is therefore easy and intuitive to use.

3) Thirdly, the present embodiment does not rely on border detection algorithms. The advantage of this relates to the freedom in choosing an optimal cluster threshold. Areas of interest usually present homogeneous signal intensity and only at the interface with other tissues or materials do grey areas affect the image. The present embodiment of the invention provides the opportunity to include or to exclude such partial volume areas depending on user experience, without relaying on rigid automatic border detection, which may lead to an incorrect segmentation. The disadvantage might relate to the absence of a fully automated segmentation achieved in some research labs although it presents strong limitations for instance when outflow artefacts affect LV blood pool.

4) Fourthly, commercially available software do not include as many image visualization and cluster creation options as the present embodiment provides. Single or multiple image visualization combined with multi-frame or multi-slice or whole volume clustering, provide the user with several options and allow to choose the best setting. 5) Fifthly, we have validated the preferred embodiment against manual segmentation and found it to be as accurate as Segment and CMRTools but to provide much faster image analysis. In contrast, CMRTools involves high manual segmentation skills making the software relatively slow, while Segment is certainly more interactive but does not reach the processing speed accomplished with the present embodiment. The present embodiment provides a similar interactivity to that of Segment, but manual intervention is faster due to the simplicity of adding and deleting single or multiple pixels whenever needed.

6) Finally, the automated ability to calculate cardiac functional and volumetric parameters extremely quickly, by effectively being able to apply automatically a cluster based segmentation achieved by a user in a semi -automated manner in one slice to other slices, provides one of the most significant advantages of the invention, as these parameters are often required by the physician in deciding on diagnosis, prognosis, and treatment. Embodiments of the present invention provide these parameters accurately and quickly with very little user input required.

Whilst the above described embodiment of the invention is directed at processing cardiac imagery obtained via MRI, other embodiments of the invention may use other medical imaging technologies, such as computerised tomography (CT), positron emission tomography (PET) or ultrasound. Moreover, other embodiments of the invention may be applied in other fields than cardiology, and embodiments may be employed whenever regions of interest have to be captured and analysed. For instance, embodiments of the invention can differentiate and quantify soft tissue areas characterised by different contrast to noise ratio; in this context, the gadolinium enhancement MRI is a representative example. For example, a myocardium infarcted area, following contrast enhancement by means of gadolinium compound segregation, can be differentiated from blood and viable myocardium. Embodiments of the invention can be adopted to locate clusters on the enhanced infarcted tissue, on viable myocardium and/or on blood thereby effectively providing tissue characterization and the estimation of the left ventricle infarcted area. Embodiments of the invention may also be applied to right ventricle segmentation in the same manner has been applied to the left ventricle. Also to fMRI studies, where enhancements in brain areas determine the amount of activation. Also, in fat-composition studies where MRI images reveal the amount of fat from areas of similar signal intensity throughout the body. Another example embodiment concerns experimentation where volumes have to be compared, e.g. within inflammation experiments where the amount of swelling is not easy to estimate whilst comparing images before and after tissue enlargement. Finally, as mentioned embodiments of the invention can be extended to imaging techniques other than MRI. For instance, in PET the amount of signal detected has to be often correlated to the covered area.

In the above described embodiment the user is relied upon to indicate the "start" pixel for the cluster-based segmentation operation. However, in other embodiments automated image processing routines may be used to try and identify the location of a suitable start pixel. For example, using a priori knowledge of the features which are to be located and characterised, it is possible to use a series of image processing operations to identify a suitable start pixel location or locations. In one such example, where a pixel inside an image of the left ventricle is to be located as the start pixel, firstly a priori knowledge is used of the whole image, and in particular that the left ventricle is typically found in the lower left hand quarter of the image. This restricts the area of the image in which the start pixel can be located. Next, the relevant area of the image may be subject to an edge detection algorithm, in order to detect edges in the image. This would result in the generally circular wall of the left ventricle being highlighted. After the edge detection, a template based feature detection can be used to detect the curved edge of the ventricle wall, for example using a template representing the same expected curvature of the wall, and then correlating the template across the edge detected image to detect the location of the wall. A start pixel may then be selected within the detected region.

Various further modifications may be made to provide additional embodiments. Example such modifications include:

• Image saved: All the information about the current segmentation may be saved, for example as a Dicom file. Once the file is saved and loaded again in the program, the user may modify, if he likes, the segmentation. The saved filed appears in a folder named with consequential numbers.

· Diastolic and systolic visualization: Once the diastolic and systolic slices have been identified, a command may be used to visualize the slices side by side in the image window. Once accessing this function, only these two slices (and their related frames) can be modified and all the studied parameters (functional and volumetric) will be calculated regarding these two slices. If, after modifying one of the slices, the user wants to visualize again the entire data set, any modifications made on the systolic and diastolic slices are included in the segmentation and new diastolic and systolic values are calculated. It might happen that while in the diastolic-systolic visualization, the user modifies the clustering of one of the two slices making it, for instance, too big to be the systolic. In this case, in the diastolic-systolic view, the parameter will still be calculated using data from these two slices, but once the user goes back to the initial data view, diastole and systole slices can appear to be on a different slice.

• Mass calculation: The factor 1.055 is multiplied by the cluster blue area of slice one (all the frames). However, it would be preferable, as a default, for the mass calculated from the blue cluster area to be associated with the first slice (all the frames), but providing an option where the user chooses the slice (from all the frames) from which to calculate the mass, and also leaving the user to choose from which color (blue or orange).

• Spline and Zoom: In some embodiments the user interface provides the ability to draw lines over the images to help in analysis thereof (spline) and/or to permit zooming in of the images, for example to more accurately determine blood muscle boundaries.

Further modifications, either by way of addition, deletion, or substitution, may be made to the above described embodiments to provide further embodiments, any and all of which are intended to be encompassed by the appended claims.

Claims

1. A method of determining one or more myocardial parameters from an MRI data set containing cardiac MRI imagery arranged in frames and slices, the method comprising: displaying a first MRI image slice to a user; from a pixel position in the first image slice identified as being within a cardiac feature, using a clustering algorithm to segment those pixels in the first image within the cardiac feature and determining a clustering threshold that achieves the segmentation; from the same pixel position in other MRI image slices of the same or other frames, using the clustering algorithm to segment those pixels in the other image slices within the cardiac feature using the determined clustering threshold; counting the segmented pixels in the respective clusters in the respective slices whereby to determine frame cardiac feature volume parameters; and calculating the one or more myocardial parameters from the determined frame volume parameters.

2. The method of claim 1 , wherein the myocardial parameters include one or more selected from the group comprising: stroke volume, end-diastolic volume (EDV), end- systolic volume (ESV), ejection fraction, and cardiac output.

3. The method of claims 1 or 2, wherein the cardiac feature is one of the left ventricle or right ventricle.

4. The method of any of the preceding claims, wherein the pixel segmentation in the first image further comprises a) displaying the first image slice to a user; b) receiving an input from the user to indicate the pixel position identified as being within the cardiac feature; and c) forming a cluster of pixels from the indicated pixel position of neighbouring pixels that meet an initial clustering threshold.

5. The method of any of the preceding claims, wherein the clustering threshold determination further comprises: a) displaying segmented pixels on the first image slice to a user; b) receiving an input from the user to alter the threshold condition, c) re-forming the cluster of segmented pixels from the start pixel position using the altered clustering threshold; and d) repeating a), b) and c) as necessary on command of the user until the clustering threshold is obtained such that the cluster extends to the boundaries of the cardiac feature the parameters of which are to be determined.

6. A method according to any of the preceding claims, wherein the clustering threshold is a single threshold value which the clustered pixel values are greater or less than..

7. A method according to any of the preceding claims, wherein the clustering threshold comprises an upper threshold value and a lower threshold value which the clustered pixel values are between

8. A system for determining one or more myocardial parameters from an MRI data set containing cardiac MRI imagery arranged in frames and slices, the system comprising:

at least one processor; and

at least one memory including computer program code

the at least one memory and the computer program code configured to, with the at least one processor, cause the apparatus to perform at least the following:

display an MRI image slice to a user; from a pixel position in the image slice identified as being within a cardiac feature, use a clustering algorithm to segment those pixels in the image within the cardiac feature and determine a clustering threshold that achieves the segmentation; from the same pixel position in other MRI image slices of the same or other frames, use the clustering algorithm to segment those pixels in the image slices within the cardiac feature using the determined clustering threshold; count the segmented pixels in the respective clusters in the respective slices whereby to determine frame cardiac feature volume parameters; and calculate the one or more myocardial parameters from the determined frame volume parameters.

9. The system of claim 8 wherein the myocardial parameters include one or more selected from the group comprising: ejection fraction, stroke volume, end-diastolic volume (EDV), end-systolic volume (ESV), and cardiac output.

10. The system of claims 8 or 9, wherein the cardiac feature is one of the left ventricle or right ventricle.

1 1. The system of any of the claims 8 to 10, wherein the pixel segmentation in the first image further comprises a) displaying the first image slice to a user; b) receiving an input from the user to indicate the pixel position identified as being within the cardiac feature; and c) forming a cluster of pixels from the indicated pixel position of neighbouring pixels that meet an initial clustering threshold.

12. The system of any of claims 8 to 1 1 , wherein the clustering threshold determination further comprises: a) displaying segmented pixels on the first image slice to a user; b) receiving an input from the user to alter the threshold condition, c) re-forming the cluster of segmented pixels from the start pixel position using the altered clustering threshold; and d) repeating a), b) and c) as necessary on command of the user until the clustering threshold is obtained such that the cluster extends to the boundaries of the cardiac feature the parameters of which are to be determined.

13. A system according to any of claims 8 or 12, wherein the clustering threshold is a single threshold value which the clustered pixel values are greater or less than..

14. A system according to any of claims 8 or 13, wherein the clustering threshold comprises an upper threshold value and a lower threshold value which the clustered pixel values are between

15. A method of determining one or more parameters of features of an imaged object from image data thereof, the method comprising:

receiving an image data set comprising one or more image frames, each frame having one or more image slices of known spatial resolution, each slice representing a spatial slice through the imaged object of known thickness;

determining a start pixel position in a first image slice containing image data relating to a feature at least one parameter of which is to be determined, the pixel position being located within the part of the image slice relating to the feature;

forming a cluster of segmented pixels extending from the determined start pixel position by comparing pixel values to a threshold condition and including pixels in the cluster if the threshold condition is met, wherein the threshold condition is selected such that the cluster extends to the boundaries of the feature the parameters of which are to be determined; for other slices in the data set, forming respective clusters of segmented pixels in each slice, in each case extending from a pixel in a corresponding position in the slice as the determined start pixel position in the first image slice, the respective clusters being formed by comparing pixel values to the threshold condition determined for the first image slice and including pixels in the cluster if the threshold condition is met;

counting the number of segmented pixels in the respective clusters;

calculating the at least one parameter of the feature in dependence on the number of segmented pixels in the respective clusters and the known spatial resolution and slice thickness; and

outputting the calculated parameter to a user.

16. A method according to claim 15, wherein the cluster formation in the first slice comprises a) displaying the segmented pixels on the image slice to a user; b) receiving an input from the user to alter the threshold condition, c) re-forming the cluster of segmented pixels from the start pixel position using the altered threshold condition; and d) repeating a), b) and c) as necessary on command of the user until the threshold condition is obtained such that the cluster extends to the boundaries of the feature the parameters of which are to be determined.

17. A method according to claims 15 or 16, wherein the threshold condition is that the pixel values are greater or less than a single threshold value.

18. A method according to claims 15 or 16, wherein the threshold condition is that the pixel values are between an upper threshold value and a lower threshold value.

19. A method according to any of claims 15 to 18, wherein the other slices are in the same frame as the first slice, and the calculated parameter is a parameter of the feature fro the frame.

20. A method according to any of claims 15 to 19, wherein the other slices are multiple slices in multiple frames, the parameter of the feature being calculated for multiple frames.

21. A method according to claim 20, and further comprising determining the maximum parameter value and the minimum parameter value, and the corresponding frames.

22. A method according to claim 21, and further comprising calculating one or more metrics values in dependence on the maximum and minimum parameter values.

23. A method according to any of claims 15 to 22, wherein the image data set is a MRI image data set.

24. A method according to any of claims 15 to 23, wherein the imaged object is a heart, and the feature to be parameterised is one of the left or right ventricle, wherein the parameter determined is the volume of the left or right ventricle.

25. A method according to claim 24 when dependent on claim 22, wherein the maximum parameter value is the end diastolic volume and the minimum parameter value is the end systolic volume, wherein the metrics calculated are one or more from the group comprising: ejection fraction, stroke volume, end-diastolic volume (EDV), end-systolic volume (ESV) and cardiac output.

26. A computer program or suite of computer programs so arranged such that when executed by a computer it/they cause the computer to operate in accordance with the method of any of claims 1 to 7 and 15 to 25.

27. A computer readable storage medium storing the computer program or at least one of the suite of computer programs according to claim 26.

28. A system for determining one or more parameters of features of an imaged object from image data thereof, the system comprising:

at least one processor; and

at least one memory including computer program code

receive an image data set comprising one or more image frames, each frame having one or more image slices of known spatial resolution, each slice representing a spatial slice through the imaged object of known thickness;

determine a start pixel position in a first image slice containing image data relating to a feature at least one parameter of which is to be determined, the pixel position being located within the part of the image slice relating to the feature; form a cluster of segmented pixels extending from the determined start pixel position by comparing pixel values to a threshold condition and including pixels in the cluster if the threshold condition is met, wherein the threshold condition is selected such that the cluster extends to the boundaries of the feature the parameters of which are to be determined;

for other slices in the data set, form respective clusters of segmented pixels in each slice, in each case extending from a pixel in a corresponding position in the slice as the determined start pixel position in the first image slice, the respective clusters being formed by comparing pixel values to the threshold condition determined for the first image slice and including pixels in the cluster if the threshold condition is met;

count the number of segmented pixels in the respective clusters;

calculate the at least one parameter of the feature in dependence on the number of segmented pixels in the respective clusters and the known spatial resolution and slice thickness; and

output the calculated parameter to a user.