The present invention relates to a system, having method and apparatus aspects, for analysing brain scan images of a patient suspected of having a stroke, particularly a system which is of use in the emergency department (ED) of a hospital.
The speed and efficiency of patient evaluation are critical factors in an emergency department. The key questions to be answered are whether 1) a patient has suffered a stroke or not, and if so whether 2) the stoke is ischemic or hemorrhagic.
The present invention aims to provide a new and useful methods and systems to determine from a scan whether a patient is suffering from a stroke.
In general terms the invention proposes that a plurality of brain atlases are co-registered, and mapped to a scan of a brain, and then the mapping is used to determine the presence of a stroke.
Preferred embodiments of the invention provide an accurate way of identifying strokes from patient brain scan data (such as CT or MRI data), and one which is particularly suitable for use in an emergency situation such as the emergency department of a hospital.
The invention may be expressed as a method of processing brain scans, as an apparatus for doing so, or as a computer program product storing software operable by a computer system to perform the method.
An embodiment of the invention will now be described, for sake of example only, with reference to the accompanying drawings, in which:
Referring to
The embodiment employs three input data sets which are respective brain atlases:
Any existing brain atlases can be used for this purpose, for instance the atlases developed in our lab [1][2][3][4]. The atlases are fully segmented and labeled, and their 3D version constructed.
In a first step 4 of the embodiment, all the atlases 1, 2, 3 are mutually co-registered using any existing techniques, for instance the FTT approach [5]. In an alternative form of the method, the co-registration process is performed in advance, and data representing the three atlases and their co-registration is input to the computer system, for example on the computer program product mentioned above.
The other input to the method is scan data 5 obtained by one or more scans of a particular patient.
In a step 6, the skull and scalp as well as any other extra-cerebral objects are removed from the scan. For this purpose, any existing method can be used, for instance that disclosed in [8] [10].
In a step 7, the midsagittal plane is extracted from the scan. For this purpose, any existing method can be used, for instance that disclosed in [6] [7]. In an alternative form of the embodiment, which improves accuracy, particularly, for the brains with a curved interhemispheric fissure, step 7 is replaced with a step of obtaining the midsagittal lines for each slice as calculated, for example in [11].
In a step 8, the mutually co-registered atlases obtained in step 4 are mapped into the scan. For this purpose any existing method can be used, for instance the FTT [5] or the statistical-based approach [9]. In addition, warping against ventricles can be used, particularly, for elder people with prominent vascular dilation. For ventricular extraction, the method described in [12] can be used. For atlas warping against the ventricles, the method described in [13] and [14] can be employed.
The mapping between the segmented and labeled brain atlases delineate regions of interest in the scan data. Any set of regions of interests with anatomical structures, vessels, and/or blood supply territories can be identified and used for analysis in the following steps of the embodiment.
In a step 9, a series of tests 91, 92, 93 applied to the output of step 9, in order to make reach a view in relation to the test. The decision tree is shown in
When performing tests 91, 92, 93, all regions of interest identified using the atlases, or any subset of them, are compared. The comparison can be done by comparing corresponding identified regions of interest in the left and right hemispheres individually (i.e. one to one) or for any group of regions of the same patient, and/or by comparing the identified regions of interest to data obtained from normal patients.
The first test 91 is to determine whether the image contains asymmetry. The comparison can employ statistics of various kinds, in particular, the mean values and standard deviations (e.g. by obtaining values for these for each hemisphere, and declaring asymmetry if they differ by more than a predetermined threshold), as well as other standard statistical tests available in SPSS eg. [15]. More advanced techniques to capture asymmetry can also be applied including [16] [17] [18].
Statistical testing can be combined with image processing techniques to eliminate certain unwanted features from the image. In particular, low and high intensity thresholds can be set manually eliminating certain image regions (i.e. the ones outside the range between these thresholds), so that ventricles and/or the skull can be removed, optionally the images may be smoothed initially by performing median or anisotropic smoothing, and then the statistical tests can applied to the intensities within the defined range [19] [20] [21] [22] [23] [24] [25][26] [27]. In principle the smoothing step can be useful even if the thresholds are not employed.
The scan is considered normal if all the corresponding regions tested produce no significant difference. If any region varies significantly from that in the contralateral hemisphere (or normal), the scan is considered abnormal.
The second test 92 is to determine whether the asymmetry is due to a stroke, or instead to some other factor. There are several situations mimicking the stroke, and additional acquisitions and human intervention may be necessary to distinguish stroke from no stroke pathology [28], [29].
Sub-step 93 is discrimination between ischemic and hemorrhagic scans. This can be done based on intensity distribution [28] and [29]. In CT scans, hyperdensity signals hemorrahage, while hypointensity indicates ischemia. On T2 MR scans, this relationship is the reverse. Hounsfield Units (HU) can further be used for discrimination; for instance, HU range of 60-100 corresponds to blood.
In a final step 10, the results of step 9 are output.
24. Gerig G, Kubler O, Kikinis R, Jolesz F A: Nonlinear anisotropic filtering of MRI data. IEEE Trans Med Imaging 1992, 11: 221-232.
25. Seramani S, Zhou J, Chan K L, Malmurugan N, Nagappan A: Denoising of MR Images using non linear anisotropic diffusion filtering as a preprocessing step. International Journal of BioSciences and Technology, IJBST 2008, 1(1):17-21.
Filing Document | Filing Date | Country | Kind | 371c Date |
---|---|---|---|---|
PCT/SG08/00303 | 8/18/2008 | WO | 00 | 7/1/2010 |
Number | Date | Country | |
---|---|---|---|
60935936 | Sep 2007 | US |