GB2468589A

GB2468589A - Identifying a Region of Interest in a Series ofMedical Images

Info

Publication number: GB2468589A
Application number: GB1004040A
Authority: GB
Inventors: Nicholas Delanie Hirst Dowson; Timor Kadir; Thomas George Wright
Original assignee: Siemens Medical Solutions USA Inc
Current assignee: Siemens Medical Solutions USA Inc
Priority date: 2009-03-12
Filing date: 2010-03-11
Publication date: 2010-09-15
Anticipated expiration: 2030-03-11
Also published as: GB0904274D0; GB2468589B; GB201004040D0

Abstract

A list of voxels of the series of images of the same subject at different time points is sorted according to a first variable, e.g. intensity, using for example an Iterative Connected Component (ICC) algorithm. A user selection of an initial voxel (510) in one image of the series of images is then registered. A region of interest (502) is then selected, the region being a plurality of voxels from the sorted list, the plurality comprising voxels from at least two images of the series, wherein each voxel of the plurality is either spatially or temporally adjacent in the series of images to another voxel of the plurality, and wherein voxels are selected for the region of interest according to a property in relation to the user-selected initial voxel.

Description

METHODS OF IDENTIFYING A REGION OF INTEREST IN A SERIES OF

MEDICAL IMAGES

This invention is directed to methods and apparatus for identifying a region of interest in a series of medical images, where the images are of the same subject at different time points.

The definition of regions or volumes of interest (ROUVOI) is a typical precursor to quantitative analysis of medical images, such as nuclear medicine emission images (for example, PET or SPECT). Such regions may be defined around areas of high intensity which correspond to high tracer uptake (hotspots). For example, in FDG-PET images for oncology, such areas may be indicative of the presence of a tumour. Oncology physicians frequently annotate lesions in PET scans for the purpose of making a diagnosis, or for use in radiotherapy. The mean or maximum tracer uptake can aid a reader in determining the likelihood of cancer. In longitudinal studies, considering the change in intensity or uptake on corresponding VOIs from images at different temporal stages may be used to determine whether a tumour has regressed.

The ROlNOl delineation step is generally a user interactive process. In PET, it is common to define such regions using a manuaIy adjusted threshold either defined on an absolute scale or with reference to a local maximum in intensity, or some other reference region.

As noted in the applicant's co-pending UK patent application no. 0914637A, incorporated herein by reference, the issue in such threshold based segmentations is the adjustment of the threshold. It should be adjusted such that the object of interest is included in the VOl but such that the background is not.

In some cases this adjustment is made difficult by the presence of other high uptake structures or features of the image adjacent to the region of interest. For example a lung tumour may be close to the heart left ventricle, a site of typical high uptake in FDG-PET. Alternatively, there may be several tumours in close proximity to one another and the user may wish to delineate each separately.

2008P 17260GB 01 The difficulty is more pronounced in 3D than in 2D since the user must check each slice over which the VOl is defined, since connectivity between voxels included in the object of interest may be present across voxels not in the current displayed slices. This can be slow and laborious.

Previously considered methods of hotspot identification or lesion annotation indude the following: Local Threshold & Connected Component Method: a containing region, surrounding a particular hotspot, is selected and a threshold is chosen to select those voxels within the region corresponding to the lesion. The threshold is used to apply a Boolean inclusion criterion to each voxel within the region and the locations of the included voxels are stored in an annotated region. Optionally, a final step to label islands of included voxels individually may be applied, all or some of which may be accepted by the user as valid annotated regions. The problems with this approach are: * the speed of the algorithm depends on the size of the containing region, and can be too slow for real-time feedback for larger regions * if the original containing region excluded part of the annotated region (for example on a different slice of the volume) it will need to be redrawn and the process repeated * multiple user interactions (e.g. mouse clicks) are required (for example, to create the containing region and update the threshold value) Watershed method: a point is selected and expanded until an intensity threshold is reached, defining an annotated region. The threshold is used as a termination criterion of, for example, a watershed type algorithm. Although this method has a 2008P17260GB01 simpler user interface, it is also problematic since very permissive thresholds could potentiaUy select most of the voxels within the image making the algorithm slow. The slowness occurs because the algorithm's speed is dependent on the number of voxels included in the final ROl. Two separate actions are required of the user: selecting the initial seed-point, then updating the threshold until the annotated region is acceptable (until it has segmented a lesion correctly).

Global Threshold & Connected Component Method: a global threshold may be applied to the image, where only voxels above the threshold are included. The sub-region surrounding a lesion is then selected. This algorithm can be slow and requires several inputs from the user. This is essentially the same as the Local Threshold and Connected Component Method, except the containing region is the entire image.

Manually segmenting out a region in the image: this can be very time-consuming if there are many lesions to be annotated or if the lesion covers multiple slices.

Automatic systems for selecting VOls exist but these typically generate spurious regions which must be rejected. Those remaining typically require individual manual adjustment as well.

Many of these methods may be launched from a determined bounding region in the image data. To prevent the inclusion of extraneous structures that are within close proximity to an ROl/VOl, a user can manually define a sufficiently small bounding region that takes the form of a box or an ellipse. The threshold operation is defined only within this outer region. This works in some cases but is difficult and sometimes impossible in others. For example, in some cases this may not be possible, as shown for a 2D example in Figure 1, where a box is used to define the bounding region.

Figure 1 is a diagram showing a cropped image taken from a single PET slice at a particular time point. In the image there are two maxima (102, 104) indicated 2008P17260GB01 by the two crosses. The region of interest (101) to the user s indicated by the solid freeform curve (106) in the image, and the bounding region is indicated by the dashed rectangle (108). An additional region (103) that is included by the box is indicated by the dotted freeform curve (110).

Particular problems arise in annotating corresponding regions of interest across frames in longitudinal scans, dynamic scans and gated scans.

As an example, consider the creation of ROls corresponding to tumours, which are examined for changes over time. These ROls are useful for staging cancer, eg., in longitudinal scans tumour changes can indicate response to treatment, while in dynamic scans a particular pattern of uptake may indicate an excessive vasculature typical of tumours. In gated scans changes in the ROl could simply indicate movement at various respiratory phases.

However before any such analysis is required, the ROI needs to be annotated in each frame.

Until this point, the ROls have been individuay annotated in each frame of the PET scan. This could be achieved by contouring the border of the ROI by hand, by using simple thresholding, or by using thresholding foowed by a connected component analysis based on a seed-point to keep only voxels connected to the seed point; however, all these methods rely on the user repeating the computation on each individual frame, a process that is relatively laborious. An example of a slice from a gated PET study of a breast cancer patient is shown in Figure 2. In this example the ROI would need to have been explicitly annotated three times, once for each time frame.

Figure 2 is a diagram showing an example of a region of interest (202) in a longitudinal study, which has shrunk (in Time period 2) and then grown again (Time 3). The rows respectively show the PET intensity (204), the region of 2008P17260GB01 interest in each frame (206) and the overlay of the region of interest and PET intensity (208). The columns show each frame from Time 1 to Time 3, respectively One method, which has been considered, is to propagate a seed-point between images and use connected component analysis to expand the seed point so that aU voxels are above a threshold. The problem with this method is that the seed point may be below the defined threshold in one of the frames and not link to the correct region of interest, as shown in Figure 3.

Figure 3 is a diagram showing an example of an ROl (302) that shrinks over 3 frames, Time 1 to Time 3. Only one 2D slice (306) of the 3D image is shown in each case. If the seed-point (204) is propagated between the frames and connected component analysis is applied, the third ROl (302c) will not be found because the seed point no longer lies within the ROI.

The present invention aims to address these problems and provide improvements upon the known devices and methods.

Aspects and embodiments of the invention are set out in the accompanying claims.

In general terms, one embodiment of a first aspect of the invention can provide a method of identifying a region of interest in a series of medical images, wherein the images of the series are of the same subject at different time points, the method comprising: obtaining a ist of voxels of the series of images sorted according to a first variable; registering a user selection of an initial voxel in one image of the series of images; and selecting as a region of interest a plurality of voxels from the sorted ist, said pluraity comprising voxels from at least two images of the series, wherein each voxel of the plurality is adjacent in the series of images to another voxel of the plurality, and wherein voxels are selected for 2008P17260GB01 the region of interest according to a property in relation to the user-selected initial voxeL This aUows selection of a region of interest spanning any of the four dimensions (x, y, z and time) in the image series, from a single user selected voxeL Suitably, a voxel adjacent to another voxel is either spatiay or tern poray adjacent to that other voxeL Preferably, the method comprises: pre-processing the series of images to generate the list of voxels sorted according to the first variable; and foUowing generation of the list, registering the user selection of the initial voxel.

This provides a simple means for fast selection of the region of interest.

More preferably, the step of selecting according to a property comprises: determining a value of the first variable for the user-selected initial voxel; and selecting as the region of interest a plurahty of voxels from the list having values for the first variable equal to or higher than that for the user-selected initial voxel.

In an embodiment, the step of pre-processing the series of images comprises pre-processing to generate a plurality of sets of voxels, the voxels of each set sorted according to a first variable, and arranging the sets of voxels in a hierarchy as a function of the first variable.

Suitably, the method further comprises: following the generation of the sets, registering the user selection of the initial voxel; determining an initial one of the sets which includes the initial voxel; and selecting as the region of interest: i. a subset of the initial set of voxels containing those voxels having values for the first variable equal to or higher than that for the user-selected initial voxel; and ii.

any other sets having a minimum voxel value for the first variable equal to or 2008P17260GB01 higher than that for the user-selected initial voxel, and having at least one voxel adjacent to a voxel of the subset.

Preferably, the puraty of sets of voxels is generated by a connected-component &gorithm.

In another embodiment, the step of obtaining the st of voxels of the series of images sorted according to the first variable comprises: determining a set of candidate regions of interest in the series of images; and determining a hierarchy among the set of candidate regions according to the first variable.

Suitably, the step of se'ecting the p'urality of voxels from the sorted list comprises: selecting at least one region from the hierarchy according to a property of said at least one region in relation to the user-s&ected initial voxel.

Preferably, the at least one region selected includes the region c'osest to the user-selected voxel.

More preferably, each candidate region is associated with a loca' maximum value for the first variable, and the at least one region selected is at least one region associated with the closest local maximum value to the user-selected voxel. Still more preferably, the at least one region selected inc'udes the user-selected voxel.

Suitably, the step of se'ecting further comprises selecting a set of regions, the method further comprising allowing a user to select at least one region from the set.

One embodiment, further comprises: determining a subset of candidate regions of interest from the set of candidate regions; determining a hierarchy among the 2008P17260GB01 subset of candidate regions; and selecting at least one region from the subset hierarchy.

In another embodiment, the hierarchy of candidate regions is generated by a connected-component algorithm.

Suitably, the first variable is an intensity value.

One embodiment of a second aspect of the invention can provide apparatus for identifying a region of interest in a series of medical images, wherein the images of the series are of the same subject at different time points, the images captured by a medical imaging apparatus, the apparatus comprising: a processor adapted to: obtain a list of voxels of the series of images sorted according to a first variable; register a user selection of an initial voxel in one image of the series of images; and select as a region of interest a plurality of voxels from the sorted list, said plurality comprising voxels from at least two images of the series, wherein each voxel of the plurality is adjacent in the series of images to another voxel of the plurality, and wherein voxels are selected for the region of interest according to a property in relation to the user-selected initial voxel; and a display device adapted to display the selected region of interest in the series of images.

An embodiment of another aspect of the invention can provide a media device storing computer program code adapted, when loaded into or run on a computer, to cause the computer to become apparatus, or to carry out a method, according to the aspects described above.

The above aspects and embodiments may be combined to provide further aspects and embodiments of the invention.

The invention will now be described by way of example with reference to the accompanying drawings, in which: 2008P17260GB01 Figure 1 is a diagram showing a cropped image taken from single PET slice; Figures 2 and 3 are diagrams illustrating difficulties with identifying a region of interest; Figure 4 is a diagram illustrating steps of a method according to an embodiment of the invention; Figure 5 is a diagram il'ustrating a method of identifying a region of interest according to an embodiment of the invention; and Figure 6 is a diagram il'ustrating an apparatus according to an embodiment of the invention.

When the following terms are used herein, the accompanying definitions can be applied: PET -Positron Emission Tomography, a method for imaging a subject in 3D using an ingested radio-active substance. Typically the image shows biological function.

ROl -Region of Interest VOl -Volume (Region) of Interest ICC -Iterative Connected Component a'gorithm CT -Computer (Aided) Tomography, a method for imaging a subject in 3D using X-Rays. Typically the image provides anatomical information.

Threshold -a particular intensity va'ue within an image, often above or below which all pixels or voxels are accepted for a process or &gorithm.

2008P17260GB01 Deneate -select a boundary within which a voxes are distinguished from the surrounding voxels.

Embodiments of the invention essentiaUy aUow selection of a region of interest spanning any of the four dimensions in the image series, from a single user selected voxel, by selecting as the region of interest a plurality of voxels, adjacent in any of the four dimensions, from a pre-obtained ist or hierarchy (from, for example, an ICC algorithm -see below) which are linked by some property to the user-selected initial voxel.

For example, this property may be intensity, which may be represented in the form of a threshold. Figure 4 sets out features of the method which may be involved in a first embodiment.

In one embodiment, the ist or hierarchy is obtained by pre-processing the image.

In certain embodiments of the invention, this pre-processing is done using an iterative connected component (ICC) algorithm. An algorithm on which this ICC agorithm is in part based was introduced by Matas (Matas et a, Robust Wide Baseline Stereo from Maximally Stable Extremal Regions, Proc. Of British Machine Vision Conference, 2002). Variants have been implemented by Hong (Hong et a, Combining Topological and Geometric Features of Mammograms to Detect Masses, Proc. Of Medical Image Understanding and Analysis, London, Sep. 2004.) amongst others.

In embodiments of the invention, the pre-processing operates by sorting the intensities in the image into descending order. The sorted list of intensities is traversed; the first voxel is labelled as the first loca maximum, and the second, if not a neighbour of the first voxel, is labelled as a second local maximum. Thus those locations that currently have no neighbours form new labels (each label is associated with one local maximum) and those adjacent to labelled voxels take the (adjacent) label with the highest maximum.

2008P17260GB01 Merges are recorded; if a voxel s connected by neighbours to both (for example) the first and second maxima, it is labeed as such. The output of the algorithm is: a label image, an intensity sorted ist of voxels and a list of merges and a list of starting points.

Further description of this type of algorithm is given in the co-pending UK patent application no. 0914637.4 It should be noted that the methods described are not restricted to the ICC algorithm, although with current technology this appears the most generay applicable. For example, a watershed-type algorithm could be used for the same purpose, given sufficient advances in computer technology giving speed improvements sufficient for real-time performance (or restriction of the problem, for example by limiting the maximum size of the region segmented).

Therefore, in contrast to the method shown in Figure 3, rather than using the same seed point or initial voxel for each time point, this embodiment aHows selection of the seed in one of the series of images, and the algorithm then checks for connectivity (here using an ICC algorithm) in four dimensions, therefore not only within that image, but across the temporal image series. Any object (e.g. lesion) which is therefore an unusual shape, or perhaps misses the seed point in other images in the series (as in Figure 3) can still be found robustly using this method, and using only a single initial voxel. The search from the seed point can be a typical intensity/threshold search.

One specific embodiment of the proposed invention comprises four steps.

1. Firstly, each frame in the study is co-registered -this improves voxel correspondences 2. Secondly, the co-registered frames are considered together as a 4D image for input into the rest of the algorithm.

2008P17260GB01 3. Thftdly, an especiay modified incremental connected components algorithm is applied to the 4D image, checking for four-way connectivity for each voxel (three ways in space, one in time, to the neighbouring frames).

4. Finally, the region corresponding to a user selected voxel or seed-point (and any other parameters such as threshold) is extracted from the output of the modified ICC algorithm.

A standard connected component algorithm or flood fill algorithm could also be used, also with 4D connectivity, but these algorithms are significantly more expensive to use as discussed above. The ICC algorithm is different in that it constructs a lookup structure describing the image in a hierarchical manner, which allows regions to be generated more rapidly than when using existing other algorithms.

This method has advantages over a of the prior solutions described. It is significantly less laborious than annotating the ROI individually in each frame, and secondly, unlike seed-point propagation, it does not rely on the ROI being connected only at the seed point, but considers connectivity at every voxel within the region. This makes ROI delineation more robust to shrinkage or movement of the ROI between frames.

The images may need to be normalised to each other so that their intensities correspond and the same threshold may be used across all the images.

Alternatively, the threshold may be varied according to which frame (i.e. which time-slice) is considered, which is less practical as it would require changes to the intensity sorting algorithm.

EXAMPLE 1

For example, take 3 PET lung oncology scans (504, 506, 508) that have been taken over the course of chemotherapy, Figure 5. It is required that the 3 scans 2008P17260GB01 are compared to assess the change in tumour volume over the period of the treatment to assess whether the tumour has grown, shrunk or stayed the same.

The three 3D PET datasets are rigidly registered (translation and rotation) using the software available on the scanner workstation so that the position of the tumour of interest is weU aligned across the 3 timepoints (504, 506, 508).

Figure 5 is a diagram showing this first example of a region of interest in a longitudinal study, which has shrunk and then grown again. The rows respectively show the PET intensity, the region of interest in each frame and the overlay of the region of interest and PET intensity (514). The columns show each frame from time 1 to time 3 (504, 506, 508), respectively.

A seed point (510) is selected in the first scan (504) to identify the position of the tumour (512) and the ICC algorithm is implemented as described above in order to propagate the region of interest (502) in 4D through the 3 datasets. The region of interest can then be viewed on the 3 scans and the volumes of the tumour at each timepoint can be compared (514-ROI only).

EXAMPLE 2

In a second example a respiration gated scan is taken, such that each 3D dataset corresponds to a different point in the respiration cycle. There are typically eight gates. The requirement is to watch how the tumour moves over the course of the respiration cycle so that its positions can be determined prior to treatment, for example radiation therapy.

The eight 3D gates of the PET data are rigidly registered (translation and rotation) using the software available on the scanner workstation so that the position of the tumour of interest is well aligned across the gates.

A seed point (initial voxel) is selected in a middle gate (mid-way between full inspiration and full expiration) to identify the position of the tumour and the ICC 2008P17260GB01 a'gorithm is implemented as described above in order to propagate the region of interest in 4D through the gates. The region of interest can then be viewed on each gate and its motion and change in position used for further treatment p'anning.

In other embodiments, other four dimensional connected component algorithms are used (rather than an ICC algorithm). In one embodiment, aU voxes above a threshold are masked-in. Masked-in regions where the vox&s are connected are grouped together, and the region which includes the click point is returned.

This algorithm differs from lower dimensiona' algorithms in that additional adjacency checks are made to account for the additional dimensions (similar to the ICC algorithm). The complexity of the algorithm is O((Nd-s-1)N), where N is the number of pixels in the image and Nd is number of image dimensions. The additional dimensionality significantly raises the cost of the algorithm. Moreover the algorithm should be re-run every time a new component is selected.

A 4D flood-fill could also be applied, where a voxel is iterativ&y grown outwards until no further voxels can be added without going below an intensity threshold.

Again, adjacency is also checked in the added (fourth) dimension. The complexity of this algorithm is O(NdNr), where Nr is the number of pixels in the region. The additiona' dimensionality also makes the flood fill algorithm expensive to use. Likewise the algorithm should be re-run whenever a new component is selected.

On the other hand, the pre-computation for ICC is O(N log(N)) i.e. it is only dependent on the number of pixels, as the initial sorting of the pixel intensities is the most expensive part of the ICC algorithm. The algorithm can be generalized to higher dimensions with little increase in computation cost. Increasing the dimensionality while keeping N constant, i.e. by running ICC in four dimensions (with 4D adjacency checking described in Step 3 in Section 4) on all frames together rather than running ICC in three dimensions individually on each frame 2008P17260GB01 has virtuaUy no effect on algorithm performance. AdditionaUy, after pre-computation the algorithm selecting the region is essentially a lookup, O(Nr), and is significantly faster than re-running standard connected components.

Of course, the embodiments described herein can be used to select regions associated with local minima as opposed to maxima, i.e. cold-spots. In addition, the embodiments can be applied to other types of images such as gradient images and distance transforms.

Referring to Figure 6, the above embodiments of the invention may be conveniently realized as a computer system suitably programmed with instructions for carrying out the steps of the methods according to the invention.

For example, a central processing unit 604 is able to receive data representative of medical scans via a port 605 which could be a reader for portable data storage media (eg. CD-ROM); a direct link with apparatus such as a medical scanner (not shown) or a connection to a network.

Software applications loaded on memory 606 are executed to process the image data in random access memory 607.

A Man -Machine interface 608 typically includes a keyboard/mouse/screen combination (which allows user input such as initiation of applications) and a screen on which the results of executing the applications are displayed.

It will be appreciated by those skilled in the art that the invention has been described by way of example only, and that a variety of alternative approaches may be adopted without departing from the scope of the invention, as defined by the appended claims.

2008P17260GB01

Claims

CLAIMS1. A method of identifying a region of interest in a series of medical images, wherein the images of the series are of the same subject at different time points, the method comprising: obtaining a list of voxels of the series of images sorted according to a first variable; registering a user selection of an initial voxel in one image of the series of images; and selecting as a region of interest a plurality of voxels from the sorted list, said plurality comprising voxels from at least two images of the series, wherein each voxel of the plurality is adjacent in the series of images to another voxel of the pluraUty, and wherein voxels are selected for the region of interest according to a property in relation to the user-selected initial voxel.
2. A method according to any preceding claim, wherein a voxel adjacent to another voxel is either spatially or temporally adjacent to that other voxel.
3. A method according to Claim 1 or Claim 2, further comprising: pre-processing the series of images to generate the list of voxels sorted according to the first variable; and following generation of the list, registering the user selection of the initial voxel.
4. A method according to any preceding claim, wherein the step of selecting according to a property comprises: determining a value of the first variable for the user-selected initial voxel; and 2008P17260GB01 selecting as the region of interest a pkirality of voxels from the list having values for the first variable equal to or higher than that for the user-selected initial voxel.
5. A method according to Claim 3 or Claim 4, wherein the step of pre-processing the series of images comprises pre-processing to generate a plurality of sets of voxels, the vox&s of each set sorted according to a first variable, and arranging the sets of voxels in a hierarchy as a function of the first variable.
6. A method according to Claim 5, further comprising: following the generation of the sets, registering the user selection of the initial voxel; determining an initial one of the sets which incfudes the initial voxel; and selecting as the region of interest: i. a subset of the initial set of voxels containing those voxels having values for the first variab'e equal to or higher than that for the user-selected initial voxel; and ii. any other sets having a minimum voxel value for the first variable equal to or higher than that for the user-selected initial voxe, and having at least one voxel adjacent to a voxel of the subset.
7. A method according to Claim 5 or Claim 6, wherein the plurality of sets of voxels is generated by a connected-component algorithm.
8. A method according to any of the Claims 1 to 3, wherein the step of obtaining the list of voxels of the series of images sorted according to the first variable comprises: 2008P17260GB01 determining a set of candidate regions of interest in the series of images; and determining a hierarchy among the set of candidate regions according to the first variable.
9. A method according to Claim 8, wherein the step of selecting the plurality of voxels from the sorted list comprises: selecting at least one region from the hierarchy according to a property of said at least one region in relation to the user-selected initial voxel.
10. A method according to Claim 9, wherein the at least one region selected includes the region closest to the user-selected voxel.
11. A method according to any one of the Claims 8 to 10, wherein each candidate region is associated with a ocal maximum value for the first variable, and wherein the at least one region selected is at least one region associated with the closest local maximum value to the user-selected voxel.
12. A method according to Claim 11, wherein the at least one region selected includes the user-selected voxel.
13. A method according to any one of the Claims 8 to 12, wherein the step of seecting further comprises selecting a set of regions, the method further comprising allowing a user to select at least one region from the set.
14. A method according to any one of the Claims 8 to 13, further comprising: 2008P17260GB01 determining a subset of candidate regions of interest from the set of candidate regions; determining a hierarchy among the subset of candidate regions; and selecting at least one region from the subset hierarchy.
15. A method according to any one of the Claims 8 to 14, wherein the hierarchy of candidate regions is generated by a connected-component algorithm.
16. A method according to any preceding claim, wherein the first variable is an intensity value.
17. Apparatus for identifying a region of interest in a series of medical images, wherein the images of the series are of the same subject at different time points, the images captured by a medical imaging apparatus, the apparatus comprising: a processor adapted to: obtain a list of voxels of the series of images sorted according to a first variable; register a user selection of an initial voxel in one image of the series of images; and select as a region of interest a plurality of voxels from the sorted list, said plurality comprising voxels from at least two images of the series, wherein each voxel of the plurality is adjacent in the series of images to another voxel of the plurality, and wherein voxels are selected for the region of interest according to a property in relation to the user-selected initial voxel; and a display device adapted to display the selected region of interest in the series of images.
18. A media device storing computer program code adapted, when loaded into or run on a computer, to cause the computer to become apparatus, or to carry out a method, according to any preceding claim.2008P17260GB01