The present invention relates to method for processing a stroke of a brain volume of a subject who has suffered a stroke, to obtain a predictive value indicative of a prognosis for the patient.
A stroke is the rapid loss of brain function due to a disturbance in the blood supply to the brain of subject. It can be due to ischemia (lack of blood flow) caused by a blockage, or a hemorrhage (bleeding within the skull). Ischemic strokes produce cerebral infarctions, in which a region of the brain (an “infarct”) dies due to local lack of oxygen. Ischemic and hemorrhagic phenomena are often mixed, for example because capillaries can be ruptured during reperfusion of an ischemic lesion (that is, when blood flow to an ischemic lesion recommences), triggering a hemorrhage.
Stokes, which have been named the second leading cause of death worldwide, often require rapid diagnosis and appropriate treatment which depends upon the type of stroke, to minimize the rise of permanent neurological damage. Computed tomography (CT) provides the first-line of diagnosis for the evaluation of acute strokes in an emergency situation.
In general terms the present invention proposes an automatic technique for stroke identification, localization, quantification and prediction, having the main steps of receiving a CT scan (and optionally other data), pre-processing it to extract the portion corresponding to a brain volume of a subject and to obtain landmarks, identifying whether a hemorrhage is present in the brain volume and if so obtaining data characterizing the hemorrhage; otherwise identifying whether an infarct is present and if so obtaining data characterizing it; analyzing the results using a brain atlas; and using the results of the analysis, obtaining at least one predictive value characterizing a prediction about the subject.
The invention may be expressed as a computerized method, or alternatively as a computer having a processor and a memory device storing software for implementation by the processor to carry out the method. Alternatively, it may be expressed as a computer program product, such as a tangible recording medium, carrying the software.
The predictive value obtained by the method may be employed in further steps of treating the patient. For example, the method enables a quantified prediction of several parameters, such as outcomes (expressed in neurological scales including the Modified Rankin Scale (mRS) or Barthel Index), survival or hospitalization stay. These parameters may be used to select a treatment for a patient, since they will have impact on treatment and healthcare cost (such as hospitalization stay). For instance, patients with a parameter value indicative of a bad predicted outcome could be treated more radically which is usually associated with a higher risk. Patients with a parameter value indicative of a good outcome prediction could take less risky treatment.
The invention may be performed by the computer system automatically (that is, without human involvement except typically to initiate the method) or semi-automatically (that is, with limited human involvement or supervision at one or more of the steps). Automatic implementation is preferable, so that the invention may be performed quickly and even when an expert operator is not available.
An embodiment of the invention will now be described, for the sake of example only, with reference to the following figures, in which:
Referring to
In a first step 1 of the method, data is received. The data is an un-enhanced CT scan of (all or part of) the brain volume of a subject. The scan is acquired as a stack of 2D slices. From the image processing standpoint, it is a 3D volume. Some or all of the processing may be performed on the 3-D volume together, rather than slice by slice. For instance, the performance of some of the sub-steps of the method may involve the calculation of the histograms for the whole brain volume (or for each hemisphere separately). Some processing however may be slice-by-slice, or using 2-dimensional sections of the data which are not the same as the acquired slices, or using 3-dimensional subsets of the 3D brain volume. Specific processing may be in any orientation (axial, coronal, sagittal) or any volume/region of interest.
There may be a preliminary step of identifying slices containing strokes, but this is more naturally performed during steps 3 or 4 since slice processing is part of localization of stroke in 3D.
Further data may be received additionally, such as perfusion (CTP) and/or angiography (CTA) scans of the brain volume.
In the second step 2 of the method, the data is pre-processed. The sub-steps of step 2 are illustrated in
The third sub-step 23 is to segment the ventricles within the image, for example using the techniques of [5] and [17].
The results of step 2 are written into the report, and also used in step 3.
Firstly, in step 3 an attempt is made in sub-step 31 to identify the hemorrhage using a method such as the one in [1]. If no hemorrhage can be identified (“N” in
There is then a localization analysis in which the location of the hemorrhage is characterized, thereby determining the type of hemorrhage: an intracranial hemorrhage (ICH), an intraventricular hemorrhage (IVH) or a subarachnoid/subdural hemorrhage (SAH/SDH). Specifically, in sub-step 34 it is determined whether the hemorrhage is outside the brain in the surrounding volume, meaning that the extracted hemorrhage is neighboring the brain extracted in sub-step 21, or neighboring the mid-sagittal plane (MSP) obtained in sub-step 22, and if so in sub-step 35 it is classified as SAH/SDH. Whether the hemorrhage is “neighboring” can be determined by testing whether one or more criteria are met. Some possible criteria for determining if the hemorrhage neighbors the MSP include: closeness in pixels (e.g. the edge of the hemorrhage closest to the MSP is within a certain number of pixels of the MSP); whether the hemorrhage is within a given radius, or minimum distance (Euclidian or Hausdorff); or whether the hemorrhage overlaps with the MSP. The MSP intersects the third ventricle of the ventricular system, so the overlapping of the hemorrhage with the MSP is only for the cortical areas (i.e., anterior brain and posterior brain), and not within the ventricular system. Another possible criterion (though difficult to implement) is the distance of the centroid of the hemorrhage to the MSP.
In sub-step 36 it is determined if the hemorrhage is in the ventricular system extracted in sub-step 23, and if so in sub-step 37 it is classified as IVH. If both sub-steps 34 and 36 give negative results, then in sub-step 38 the hemorrhage is classified as ICH. The order of steps 34 and 36 may be different in other embodiments of the invention, but it is more natural to perform step 34 first since it is likely to be faster.
The step 4 of infarct processing is illustrated in
If CTP and CTA scans were received in step 1, these two may be processed, for instance as in [8]. The penumbra can be segmented and quantified by measuring its volume. The mismatch may be calculated. The CTA scan may be analyzed, for instance by employing the method of [8] and visualizing it in three-dimensions. Again, the result is collected for use in the report. In addition, the localization of vascular occlusion may be performed, by employing for instance, method [9], and the result incorporated in the report.
Turning to
Again, the result of step 5 is collected and used for the report. Already by this stage, enough information has been collected for the report to be a useful tool for diagnosis and treatment, even if the subsequent predictive step 6 is omitted. Note that step 5 is typically performed not in a common space suitable for any individual, but in the image space (patient's space).
The prediction step 6 is performed based on two principles: quantification of hemorrhagic transformation (that is, the amount of the brain which has been transformed by the hemorrhage), and statistical on the basis of a probabilistic stroke atlas. The first principle motivates a sub-step 61 of obtaining the ratio of the hemorrhage volume to the ischemic lesion volume, using the results of sub-steps 33 and 43. This ratio can be used as a predictor. If the ratio is bigger than a given ratio (such as a predetermined value, for instance 30%), it indicates a poor outcome.
The second principle motivates a sub-step 62 of using a probabilistic stroke atlas, for example as discussed in [16], [18], to derive at least one predictive value representing a prediction of concerning the subject. The prediction may be in terms of survival and outcome. For instance, the predictive value may be a prediction of how the subject will score on the Barthel Index at a given time in the future, such as after 30, 60, 180, 360 days. The Barthel index is a scale used to measure the performance of basic activities of daily living. A rating on the Barthel Index is calculated by rating the ability of a patient to perform ten variables describing the activities of daily living and mobility, such that a higher number is associated with a greater likelihood of being able to live at home with a degree of independence following discharge from hospital. Alternatively, the predictive value may be prediction how the subject will score on the modified Rankin scale (mRS) at a given time in the future, such as 7, 30, 90, 180 and 360 days. The modified Rankin scale is a commonly used scale for measuring the degree of disability of dependence in the daily activities of people who have suffered a stroke.
A probabilistic stroke atlas is one of the types of probabilistic atlas proposed in [16] and evaluated in [18]. It is generated by obtaining brain images for many patients who suffered from a stroke earlier. [16] explains the probabilistic stroke atlas particularly in terms of ischemic infarcts, but the method is useful also for hemorrhagic strokes. Each of the images contains damage caused by the stroke (e.g. an ischemic lesion) at a plurality of locations, and each is associated with the value of one or more parameters (Pn) characterizing the corresponding patient (e.g. a Barthel Index score or a modified Rankin Scale, for example as measured a certain number of days after the image was captured). To construct this atlas, the brain images and neurological parameters are transformed into a common space defined based on a brain atlas. The probabilistic stroke atlas is then generated in two portions. A first segment (PSA_S) indicates, for each point of the common space, the number of the brain images for which one of the locations was at the corresponding point. The second segment (PSA_Pn) exists for each of the parameters, and for each point of the common space, a value indicative of the value of the parameter for those patients for whom one of the locations was at the corresponding point.
In step 62, the probabilistic stroke atlas is mapped onto the subject-specific data, e.g. in the same way as the anatomical atlas is mapped in step 5. If it is determined for a given subject that he or she has stroke damage at a given set of locations, then the probabilistic stroke atlas can be used to produce an estimate of one of the parameters—which, for certain of the parameters means a prediction relating to the patient (for example, if the parameter is a Barthel index score or a modified Rankin scale score for a certain time after the image is captured). This may done for example by extracting, from the portion of the second segment of the probabilistic atlas corresponding to the parameter to be estimated, the values at the locations at which the subject has stroke damage. The distribution of the extracted values gives an estimate for the parameter for the subject, and a measure of the uncertainty in that estimate. The probabilistic stroke atlas can be superimposed onto the patient's scan by using the same methods as those for atlas-scan mapping.
Note that steps 61 and 62 are independent. Step 61, based on clinical observation, is heuristic and qualitative to predict outcome. Step 62 based on the probabilistic stroke atlas is quantitative and can predict not only outcome (in terms of the stroke scales), but also other parameters, such as survival.
As noted above, the localization analysis (sub-steps 34 to 38) classifies the hemorrhage (i.e. ICH, IVH, or SAH/SDH). The embodiment uses this information for the report, but step 6 does not take this information into account. However, in other embodiments of the invention, the hemorrhage classification can be used in step 6 for producing the predictive values and/or in selecting a treatment, e.g. by using a statistical analysis to find correlations between the hemorrhage type and the respective success rates of possible treatments.
The step 62 of using the probabilistic stroke atlas in step 62 can be performed for the whole brain or any of its part defined by the individualized atlases calculated in step 5. For example, over one or more of the blood supply territories obtained in step 5.
The CTP and CTA scans are used to make decision about thrombosysis, which is associated with certain risk. Step 6 facilitates this risk assessment. In principle it would also be possible to generate a probabilistic stroke atlas based on CTP/CTA, though that is not implemented in the present embodiment.
Note that the treatment given to a given subject, and its outcome prediction, depends upon a given situation: a hemorrhage only, ischemia only, or ischemia with hemorrhagic transformation. The invention identifies all three situations.
If both of steps 31 and 42 are negative, then no stroke is detected in the method and steps 5 and 6 are skipped.
Number | Date | Country | Kind |
---|---|---|---|
201100862-0 | Feb 2011 | SG | national |
Filing Document | Filing Date | Country | Kind | 371c Date |
---|---|---|---|---|
PCT/SG2012/000027 | 1/30/2012 | WO | 00 | 7/2/2013 |