METHOD AND SYSTEM FOR ASSESSMENT OF ACUTE ISCHEMIC STROKE

BACKGROUND

[0001]
Embodiments of the present specification relate generally to magnetic resonance imaging (MRI), and more particularly to a method and a system for assessment of acute ischemic strokes.

[0002]
Stroke has been a leading cause of death and disability in recent times, with more than 15 million people suffering from strokes each year globally. Typically, a stroke may be hemorrhagic or ischemic. A hemorrhagic stroke occurs when a blood vessel ruptures, thus flooding a portion of the brain with blood. A majority of strokes, however, are ischemic strokes that occur when a blood vessel is blocked, for example, due to a clot. The blocked blood vessel causes oxygen deprivation in cerebral tissues, which if left untreated for more than a few hours leads to necrosis of the cerebral tissues
.
[0003]
Ischemic strokes may be treated using thrombolytic agents that are designed to dissolve an obstructive clot and restore blood flow to hypoperfused and/or depolarized areas of the cerebral tissues. Rapid restoration of blood flow may potentially salvage portions of the affected cerebral tissues that have not yet been irreversibly damaged. Such portions of the cerebral tissues are commonly referred to as "ischemic penumbra," while portions of the cerebral tissues that have been irreversibly damaged due to oxygen deprivation are referred to as "core ischemic zones."

[0004]
Generally, cerebral tissues cease to function if deprived of oxygen for more than sixty to ninety seconds. However, after approximately three hours, the tissues may suffer irreversible injury possibly leading to a cerebral infarct or death of the tissues. Accordingly, only a narrow window of time after the onset of the stroke is available to a medical practitioner for administration of a thrombolytic agent for salvaging the ischemic penumbra.

[0005]
A decision to administer the thrombolytic agent, however, is not automatic because reperfusion of severely hypoperfused areas can result in hemorrhage and associated complications. Accordingly, the thrombolytic agent is administered only if an estimation of the ischemic penumbra in the patient is large enough to justify pharmacological treatment and a corresponding risk of hemorrhage. In case of a patient with insignificant salvageable cerebral tissue, however, administering the thrombolytic agent may unnecessarily aggravate a risk to patient health and mortality. Accurate and rapid assessment of risk to determine viability of thrombolytic administration in the early minutes of treating the patient with stroke symptoms, therefore, is key to a successful patient recovery.

[0006]
Generally, assessment of the risk associated with thrombolytic administration entails estimation of a size of the ischemic penumbra and/or core ischemic zones. The size of the ischemic core may be estimated using diagnostic images such as diffusion weighted MRI (DWI) images. Particularly, in an acute ischemic stroke, one or more parameters determined from the DWI images may be used to distinguish between normal and abnormal tissues in the cerebrum. By way of example, a receiver operating channel characteristics (ROC)-based classifier may be applied to the DWI images to determine an optimal Apparent Diffusion Coefficient (ADC) threshold. Further, regions of the DWI image having ADC values that are lower than the ADC threshold may be classified as abnormal tissues.

[0007]
However, accuracy of ADC values depends on accuracy of a b-factor or q-factor used during diffusion-weighted magnetic resonance imaging. Generally, the b-factor and/or q-factor are representative of diffusion sensitivity of an MRI sequence and may aid in ascertaining influence of diffusion gradients on the DWI images. The b-factor and/or q-factor used for DWI, however, may vary due to system-specific factors such as gradient non-linearity and concomitant field effects. These system-specific factors may themselves differ between different vendors and MRI systems, thus, resulting in inconsistent b-factor and/or q-factor values. In addition, ADC values may also exhibit anisotropy, for example, due to patient motion during imaging. Such anisotropy and/or inconsistency in the b-factor or q-factor values, in turn, may result in an erroneous decision regarding suitability of the thrombolysis treatment for a stroke patient.

BRIEF DESCRIPTION

[0008]
In accordance with aspects of the present specification, a method for magnetic resonance imaging is presented. The method includes receiving magnetic resonance imaging data corresponding to a target volume of a subject, wherein the magnetic resonance imaging data includes diffusion weighted imaging data. The method further includes generating at least one apparent diffusion coefficient map from the diffusion weighted imaging data. Additionally, the method includes generating one or more masks corresponding to one or more regions of interest in the target volume. The method also includes applying the one or more masks to the diffusion weighted imaging data and the at least one apparent diffusion coefficient map to determine a subset of diffusion weighted imaging data and a subset of the at least one apparent diffusion coefficient map corresponding to the one or more regions of interest. Further, the method includes normalizing the subset of the diffusion weighted imaging data and/or the subset of the at least one apparent diffusion coefficient map using a determined diffusion weighted imaging intensity and/or a determined value corresponding to the at least one apparent diffusion coefficient map. Moreover, the method includes determining an optimal classifier for classifying the one or more regions of interest with a desired specificity and/or a desired sensitivity using the normalized subset of the diffusion weighted imaging data and the normalized subset of the at least one apparent diffusion coefficient map. Furthermore, the method includes automatically segmenting the one or more regions of interest using the optimal classifier to distinguish between normal and abnormal tissues in the one or more regions of interest with the desired specificity and/or the desired sensitivity.

[0009]
In accordance with another aspect of the present specification, a magnetic resonance imaging system is disclosed. The system includes a scanner configured to scan a target volume of a subject to acquire magnetic resonance imaging data. The system further includes a processing subsystem operatively coupled to the scanner and configured to generate diffusion weighted imaging data and at least one apparent diffusion coefficient map using the magnetic resonance imaging data. Additionally, the processing subsystem is configured to generate one or more masks corresponding to one or more regions of interest in the target volume. The processing subsystem is also configured to apply the one or more masks to the diffusion weighted imaging data and the at least one apparent diffusion coefficient map to determine a subset of diffusion weighted imaging data and a subset of the at least one apparent diffusion coefficient map corresponding to the one or more regions of interest. Further, the processing subsystem is configured to normalize the subset of the diffusion weighted imaging data and/or the subset of the at least one apparent diffusion coefficient map with a determined diffusion weighted imaging intensity and/or a determined value corresponding to the at least one apparent diffusion coefficient map. Moreover, the processing subsystem is configured to determine an optimal classifier for classifying the one or more regions of interest with a desired specificity and/or a desired sensitivity using the normalized subsets of the diffusion weighted imaging data and the at least one apparent diffusion coefficient map. Additionally, the processing subsystem is configured to automatically segment the one or more regions of interest using the optimal classifier to distinguish between normal and abnormal tissues in the one or more regions of interest with the desired specificity and/or the desired sensitivity.

DRAWINGS

[0010]
These and other features, aspects, and advantages of embodiments of the present invention will become better understood when the following detailed description is read with reference to the accompanying drawings in which like characters represent like parts throughout the drawings, wherein:

[0011]
FIG. 1 is a schematic representation of an exemplary MRI system, in accordance with aspects of the present specification;

[0012]
FIG. 2 is a flowchart depicting an exemplary method for magnetic resonance (MR) imaging, in accordance with aspects of the present specification;

[0013]
FIGs. 3A and 3B illustrate a flowchart depicting an exemplary method for determining an optimal classifier described with reference to the method of FIG. 2, in accordance with aspects of the present specification;

[0014]
FIG. 4 is a graphical representation depicting an example of a cumulative histogram representative of normal and abnormal tissues in a plurality of subjects and generated using the method of FIGs. 3A and 3B, in accordance with aspects of the present specification;

[0015]
FIG. 5 is a graphical representation depicting exemplary ROC curves corresponding to one or more classifiers determined using the method of FIGs. 3 A and 3B, in accordance with aspects of the present specification;

[0016]
FIG. 6 is a graphical representation depicting exemplary values of sensitivity determined at desired values of specificity using the method of FIGs. 3 A and 3B, in accordance with aspects of the present specification; and

[0017]
FIG. 7 illustrates exemplary images depicting a comparison between an exemplary diffusion lesion segmentation in a DWI image performed using the method of FIG. 2 with a corresponding manual diffusion lesion segmentation, in accordance with aspects of the present specification.

DETAILED DESCRIPTION

[0018]
The following description presents exemplary systems and methods for automatic segmentation of a diffusion lesion for accurately assessing an extent of an acute ischemic stroke. Typically, diagnosis and treatment of the ischemic stroke entails identification of healthy and diffusion lesion regions using ADC-based classification. However, as previously noted, system-specific factors such as gradient non-linearity and concomitant field effects may result in errors in the ADC measurements, thereby adversely affecting any diagnosis.

[0019]
Accordingly, embodiments described hereinafter disclose a method and a system for determining an optimal classifier that may allow for enhanced segmentation of a diffusion lesion in a DWI image. Particularly, the optimal classifier may be determined via ROC analysis of both ADC and DWI data. The ROC analysis may aid in identifying the optimal classifier that allows for an accurate distinction between normal and abnormal tissues in a region of interest (ROI) of a patient. The accurate distinction between normal and abnormal tissues, in turn, may allow a clinician such as a radiologist to identify a type and extent of the stroke, thereby aiding in determining an appropriate treatment for the patient.

[0020]
Although exemplary embodiments of the present systems and methods are described with reference to ischemic stroke assessment, it will be appreciated that use of embodiments of the present systems and methods in various other imaging applications is also contemplated. For example, embodiments of the present systems and methods may find use in allowing for robust and reproducible segmentation of diffusion and/or perfusion lesions. Particularly, embodiments of the present systems and methods may allow for robust

segmentation of lesions caused by different pathologies, for example tumors, and occurring in different anatomical regions such as abdomen, breast, liver, kidney, and/or brain. An exemplary environment that is suitable for practising various implementations of the present system is discussed in the following sections with reference to FIG. 1.

[0021]
FIG. 1 illustrates an MR] system 100 configured for MR imaging. In certain embodiments, the system 100 may be configured to aid in the assessment of an acute ischemic stroke. More particularly, the system 100 may be configured to automatically segment regions of interest (ROIs) in brain tissues following an ischemic stroke. To that end, in one embodiment, the MRI system 100 may include a scanner 102, a system controller 104, and an operator interface 106. The scanner 102 may further include a patient bore 108 into which a table 110 may be positioned for disposing a patient 112 in a desired position for scanning.

[0022]
Further, in certain embodiments, the scanner 102 may also include a series of associated coils for imaging the patient 112. In one embodiment, for example, the scanner 102 may include a primary magnet coil 114 energized via a power supply 116 for generating a primary magnetic field generally aligned with the patient bore 108. The scanner 102 may further include a series of gradient coils 118, 120 and 122 grouped in a coil assembly for generating accurately controlled magnetic fields, the strength of which may vary over a designated field of view (FOV) of the scanner 102.

[0023]
Additionally, in one embodiment, the scanner 102 may include a radiofrequency (RF) coil 124 configured to generate RF pulses for exciting a gyromagnetic material, typically bound in tissues of the patient 112. In certain embodiments, the RF coil 124 may also serve as a receiving coil. Accordingly, the RF coil 124 may be operationally coupled to transmit-receive circuitry 126 in passive and/or active modes for receiving emissions from the gyromagnetic material and for applying RF excitation pulses, respectively. Alternatively, the IVIK.I system IUU may inciuae various coniigurauons or receiving cons ainerent from the RF coil 124.

[0024]
In certain embodiments, the system controller 104 may be configured to control operation of the MR coils 118, 120, 122, and 124 for generating a desired magnetic field and RF pulses. To that end, in one embodiment, the system controller 104 may include a pulse sequence generator 128, timing circuitry 130, and a processing subsystem 132. In one embodiment, the system controller 104 may be configured to use the pulse sequence generator 128, the timing circuitry 130, and/or the processing subsystem 132 to generate and/or control imaging gradient waveforms and RF pulse sequences employed during a medical procedure.

[0025]
Further, the system controller 104 may also include amplification circuitry 134 and interface circuitry 136 configured to control and/or interface between the pulse sequence generator 128 and the coils of scanner 102. For example, the amplification circuitry 134 and/or the interface circuitry 136 may be configured to drive the RF coil 124 and amplify corresponding response signals for further processing. The amplified response signals, in turn, may be transmitted to the processing subsystem 132 for determining information for use in image reconstruction.

[0026]
The processing subsystem 132, for example, may include one or more application-specific processors, graphical processing units (GPUs), digital signal processors (DSPs), microcomputers, microcontrollers, Application Specific Integrated Circuits (ASICs) and/or Field Programmable Gate Arrays (FPGAs). In certain embodiments, the processing subsystem 132 may be configured to process the response signals emitted by excited patient nuclei in response to the RF pulses. Specifically, the processing subsystem 132 may be configured to demodulate, filter, and/or digitize the response signals for determining the image reconstruction information. Additionally, the processing subsystem 132 may be configured to transmit digitized MRI data to an image processing unit 138 to allow reconstruction of desired images of a target volume of interest (VOI) in the patient 112.

[0027]
Particularly, in one embodiment, the image processing unit 138 may be configured to use the digitized MRI data to generate DWI images and/or corresponding ADC maps. Further, the image processing unit 138 and/or the processing subsystem 132 may also be configured to identify a ROI corresponding to a probable diffusion lesion in the DWI images. Particularly, the image processing unit 138 may be configured to identify regions of perfusion change, namely the ischemic penumbra, the core ischemic zones, and/or the probable diffusion lesion in the DWI images with greater accuracy. In certain embodiments, the regions of perfusion change may be determined based on measurements obtained using DWI-based and ADC-based segmentation of a ROI. The ROI, for example, may correspond to a probable diffusion lesion.

[0028]
Accuracy of the segmentation of the diffusion lesion, however, may depend upon an accuracy of DWI data acquisition by the MRI system 100. In certain scenarios, the DWI data acquisition may be corrupted by patient motion and/or system imperfections. DWI trace maps, DWI images, and/or ADC maps generated from DWI data acquired during such a corrupt DWI acquisition, thus, may exhibit anisotropy, thereby resulting in false positives during the segmentation of the diffusion lesion.

[0029]
Conventional segmentation methods fail to address anisotropy in the DWI data due to corruption of the DWI acquisition, and thus, continue to use the corrupted DWI data for segmentation of the diffusion lesion and subsequent analysis of lesion characteristics. For example, histogram distributions of ADC and DWI values corresponding to normal and abnormal tissues are known to overlap, thus resulting in over-estimation or under-estimation of the lesion volumes. Embodiments of the system 100, however, allow for mitigation of the anisotropy in the DWI data during segmentation and/or the subsequent analysis of the diffusion lesion characteristics.

[0030]
Particularly, in one embodiment, the processing subsystem 132 may be configured to generate ROI masks for the DWI image to mitigate anisotropy in the DWI data by preventing false positive lesion markings. The ROI masks may also be used to normalize the DWI data. The normalized DWI data along with the normalized values of the ADC map may be used to generate a histogram that may be representative of a distribution of voxels corresponding to normal and/or abnormal tissues in the ROI.

[0031]
Accordingly, in one embodiment, the processing subsystem 132 may be configured to generate the ROI masks using a priori information. The a priori information, for example, may include an anatomical atlas, model-based mapping, previously generated images of the patient under investigation and/or other patients. Moreover, the processing subsystem 132 may be configured to normalize a subset of the DWI data and/or the ADC map corresponding to the ROI, for example, the probable diffusion lesion using a determined DWI intensity. In one embodiment, the processing subsystem 132 may be configured to normalize the DWI data by dividing DWI intensity at each voxel with a desired DWI intensity value. The desired DWI intensity value, for example, may correspond to 98th percentile of the DWI intensities in the ROI. The normalization aids in performing sensitivity and specificity analysis on data generated from combining information corresponding to a collection of patient data.

[0032]
Further, in accordance with exemplary aspects of the system 100, the processing subsystem 132 may be configured to determine an optimal classifier using the normalized subsets of the DWI data and the ADC map for classifying the probable diffusion lesion. Particularly, the optimal classifier may be used to aid in determining a classification threshold that may allow for an accurate distinction between normal and abnormal tissues in the probable diffusion lesion. Use of the optimal classifier, thus, may allow for identification of true diffusion lesion regions. Lesion markings, thus obtained, on the DWI image using the optimal classifier may further be refined. In one embodiment, for example, morphological processing such as a morphological "close" operation that uses a disk shaped structuring element of a determined radius may be applied to the probable diffusion lesion region to remove regions having negligible volumes. Removal of regions of negligible volume may result in efficient lesion segmentation with a desired sensitivity and/or specificity.

[0033]
As used herein, the term "sensitivity" may be used to refer to a proportion of diffusion lesion regions that are correctly identified as the lesion. Sensitivity, thus, corresponds to a true positive rate. Further, as used herein, the term "specificity" may be used to refer to a proportion of healthy tissue regions that are correctly identified as the normal tissues. Specificity, thus, corresponds to a true negative rate. A measure of specificity may also aid in determining a false positive rate (1-specificity), that is, in identification of tissues that may be incorrectly identified as the diffusion lesion. An exemplary method for determining the optimal classifier for accurately classifying normal and abnormal tissues in an ROI with the desired specificity and/or sensitivity will be described in greater detail with reference to FIGs. 2, 3A and 3B.

[0034]
In one embodiment, the desired specificity and/or the desired sensitivity may be pre-programmed into the MRI system 100 and/or may be defined by the medical practitioner 140 via the operator interface 106. To that end, the operator interface 106 may include one or more input devices 142 that are operationally connected to the MRI system 100 via a communications link 144, such as a backplane or Internet. The input devices 142, for example, may include a keyboard, a mouse, a trackball, a joystick, a touch-activated screen, a light wand, a control panel, and/or an audio input device such as a microphone associated with corresponding speech recognition circuitry. The input devices 142 may also allow the medical practitioner 140 to request for image-derived information such as diffusion lesion characteristics for evaluating stroke parameters corresponding to the patient 112.

[0035]
Moreover, in one embodiment, the image processing unit 138 may be configured to provide the medical practitioner 140 with the requested image-derived information in real-time through one or more output devices 146. For example, the image processing unit 138 may be configured to provide the medical practitioner 140 with a measure of a volume of the identified diffusion lesion via the output devices 146. To that end, the output devices 146, for example, may include a display 148, a printer 150, and/or an audio output device 152. In one embodiment, the display 148 may be integrated into wearable eyeglasses, or may be ceiling or cart mounted to allow the medical practitioner 140 to observe the DWI images and other medically relevant information during imaging.

[0036]
Accordingly, in certain embodiments, the image processing unit 138 may be configured to transmit the requested information to the medical practitioner 140 as a visual report on the display 148 and/or the printer 150. Additionally, the image processing unit 138 may be configured to convey the requested information to the medical practitioner 140 audibly through the audio output device 152.

[0037]
Alternatively, the image processing unit 138 may be configured to store the requested information in a storage repository 154. In one embodiment, the storage repository 154 may include devices such as a hard disk drive, a floppy disk drive, a compact disk-read/write (CD-R/W) drive, a Digital Versatile Disc (DVD) drive, a flash drive, and/or a solid-state storage device. The storage repository 154 may also be configured to store the acquired MRI data, reconstructed DWI images, ADC maps, and/or diffusion lesion characteristics for use in diagnosis and/or treatment of the patient 112.

[0038]
In one embodiment, for example, the image processing unit 138 may be configured to use the stored diffusion lesion characteristics to identify the ischemic penumbra and core ischemic regions in the brain tissues. The medical practitioner 140 may rely on this information for determining whether to administer the thrombolytic agent to the patient 112. Additionally, the medical practitioner 140 may use this information for evaluating an effect of the thrombolytic agent in real-time and further for determining whether to terminate or continue therapy based on the evaluated effect. In case of a longitudinal and/or serial scan performed on the patient 112, the information may be used to determine a status of patient recovery to evaluate effectiveness of therapy and/or to aid in determining appropriate treatment for rehabilitation and/or early discharge of the patient 112.

[0039]
Embodiments of the present system 100, thus, allow for automatic lesion segmentation in DWI images using an optimal classifier. The optimal classifier may be determined using a combination of ADC and DWI data. Particularly, use of the combination of the ADC-based and DWI-based based data may mitigate certain erroneous effects of system-specific b-factor and/or q-factor variability, thus allowing for greater accuracy and consistency in assessing stroke parameters. Accurate assessment of the stroke parameters may aid in determining an extent of an infarct in the brain tissues, which in turn, may aid in determining appropriate treatment for the patient 112.

[0040]
An exemplary method for improving the accuracy of segmentation of a diffusion lesion based on an optimal classifier determined using combination of the ADC and DWI based data will be described in greater detail with reference to FIG. 2.

[0041]
FIG. 2 illustrates a flow chart 200 depicting an exemplary MR imaging method for differentiating between normal and abnormal tissues with a desired sensitivity and/or specificity. Embodiments of the exemplary method may be described in a general context of computer executable instructions on a computing system or a processor. Generally, computer executable instructions may include routines, programs, objects, components, data structures, procedures, modules, functions, and the like that perform particular functions or implement particular abstract data types.

[0042]
Embodiments of the exemplary method may also be practised in a distributed computing environment where optimization functions are performed by remote processing devices that are linked through a wired and/or wireless communication network. In the distributed computing environment, the computer executable instructions may be located in both local and remote computer storage media, including memory storage devices.

[0043]
Further, in FIG. 2, the exemplary method is illustrated as a collection of blocks in a logical flow chart, which represents operations that may be implemented in hardware, software, or combinations thereof. The various operations are depicted in the blocks to illustrate the functions that are performed, for example, during the steps of receiving MRI data, normalizing subsets of DWI data and/or ADC map, determining an optimal classifier, and automatic segmentation phases of the exemplary method. In the context of software, the blocks represent computer instructions that, when executed by one or more processing subsystems, perform the recited operations.

[0044]
The order in which the exemplary method is described is not intended to be construed as a limitation, and any number of the described blocks may be combined in any order to implement the exemplary method disclosed herein, or an equivalent alternative method. Additionally, certain blocks may be deleted from the exemplary method or augmented by additional blocks with added functionality without departing from the spirit and scope of the subject matter described herein. For discussion purposes, the exemplary method will be described with reference to the elements of FIG. 1.

[0045]
Conventionally, during lesion segmentation, histogram distributions of ADC or DWI image data may be used for differentiating between normal and abnormal tissues in an ROI. These histograms, however, are known to overlap, thus resulting in an over-estimation or under-estimation of the diffusion lesion characteristics. As previously noted, successful recovery of a patient suffering from an acute ischemic stroke may depend upon an accurate and rapid
quantification of the diffusion lesion characteristics such as lesion volume computed from the DWI images. An inaccurate determination of the diffusion lesion regions, and in turn, the diffusion lesion volumes may adversely affect a decision regarding administration of a thrombolytic agent.

[0046]
Accordingly, embodiments of the present method describe a robust and reproducible method for MR imaging that allows for an efficient assessment of an acute ischemic stroke. Particularly, embodiments of the present method describe automatically identifying and segmenting diffusion lesion regions in the DWI images using an optimal classifier. The segmented diffusion lesion regions may be used, for example, to estimate lesion volume for assessing a need for administration of the thrombolytic agent. To that end, in one embodiment, a patient, such as the patient 112 of FIG. 1, is suitably positioned on an examination table associated with an MRI system, such as the MRI system 100 of FIG. 1. Particularly, the patient may be positioned such that a desired portion of head of the patient is positioned within a field of view (FOV) of the MRI system. Subsequently, the patient may be scanned to acquire MRI data corresponding to the desired portion in presence and/or absence of a contrast agent.

[0047]
Further, one or more DWI images corresponding to a target volume of the patient may be generated using the acquired MRI data. In one embodiment, the DWI images may be generated such that an intensity of each image element or voxel in the DWI images may reflect an accurate estimate of a rate of water diffusion at that location. Generally, mobility of water is driven by thermal agitation and is highly dependent on water's cellular environment. Accordingly, the DWI images may be indicative of changes to a pathological state of the brain tissues corresponding to the target volume. Particularly, the DWI images may allow for a more efficient capture of early changes after an ischemic stroke as opposed to more traditional MRI measurements such as spin-lattice relaxation time (Tl) and spin-spin relaxation time (T2) rates.

[0048]
An exemplary embodiment of the present method begins at step 202, where MRI data corresponding to a target volume of a subject may be received. The MRI data, in one embodiment, may be received from a processing subsystem such as the processing subsystem 132 of FIG. 1 and/or a memory device such as the storage repository 154 of FIG. 1. In one example, the received MRI data may include DWI data.

[0049]
Further, at step 203, an ADC map may be generated from the received DWI data. Typically, a diffusion lesion region may include a range of tissue types with varying levels of ischemia and cellular damage owing to differences in a stroke mechanism, an anatomical location of the diffusion lesion, supplying vascular apparatus and/or a time to onset of an ischemic stroke. As previously noted, conventional segmentation techniques that use only ADC-based information may not account for regional heterogeneity of the diffusion lesion region, thus failing to accurately differentiate between healthy and lesion regions for use in diagnosis and treatment planning. In contrast to such conventional techniques that employ only ADC-based information, embodiments of the present method may be configured to determine diffusion lesion information using a combined ADC and DWI-based diffusion lesion segmentation.

[0050]
Accordingly, at step 204, one or more masks corresponding to one or more ROIs in the target volume may be generated. In one embodiment, the masks may correspond to different ROIs in the brain. By way of example, the masks may include a cerebrum mask, a cerebellum mask, and/or a brain ventricle mask. The masks may allow exclusion of false positive lesion markings from regions determined to be unrelated to an assessment of a stroke.

[0051]
Furthermore, at step 206, the one or more masks may be applied to the DWI data and the ADC map to determine a subset of the DWI data and a subset of the ADC map corresponding to the one or more ROIs. In one embodiment, the generated masks may be used to divide the DWI data and the ADC map into subsets of image data that correspond to different regions of the brain such as the cerebrum, the cerebellum, and a brain ventricle. Such a division of the DWI data and/or the ADC map allows for a focused segmentation of the diffusion lesion in each region. Moreover, use of the masks may allow removal of unrelated regions, such as the brain ventricles, where there is no probability of occurrence of the diffusion lesion.

[0052]
Additionally, at step 208, the subset of the DWI data and/or the subset of the at least one ADC map may be normalized based on a determined diffusion weighted imaging intensity and/or a determined ADC value respectively. For example, in one embodiment, the DWI data may be normalized by dividing the DWI intensity at each voxel with an intensity value corresponding to the 98th percentile of the DWI intensities in the ROI. The normalization process aids in performing sensitivity and specificity analysis on the data generated from combining the information from a complete collection of patient data.

[0053]
Moreover, at step 210, an optimal classifier may be determined using the normalized subsets of the DWI data and/or the ADC map. The optimal classifier may be employed for classifying the one or more regions of interest with a desired specificity and/or sensitivity. As previously noted, the desired specificity and/or the desired sensitivity may be pre-programmed into the MRI system and/or may be defined by the medical practitioner.

[0054]
In one embodiment, histogram distributions of normalized subsets of the DWI data and/or the ADC maps corresponding to a plurality of patients may be generated. Further, a cumulative histogram may be generated from the histogram distributions corresponding to the plurality of the patients. In addition, an ROC analysis may be performed on the cumulative histogram. An optimal classifier may be determined based on the ROC analysis. An exemplary method for determining the optimal classifier will be described in greater detail with reference to FIGs. 3A and 3B.

[0055]
In one embodiment, the optimal classifier may be used to determine a classification threshold that may allow for an accurate differentiation between normal and abnormal tissues with the desired specificity and/or sensitivity. Furthermore, at step 212, the one or more ROIs may be automatically segmented using the optimal classifier. Automatically segmenting the one or more ROIs using the optimal classifier may aid in distinguishing between normal and abnormal tissues with the desired specificity and/or sensitivity. In conventional segmentation techniques, there is a trade-off between the specificity and sensitivity such that only one of the specificity and sensitivity may have a high value. Embodiments of the present method, however, allow for use of a robust classification threshold that aids in improving the performance of the segmentation such that a higher sensitivity may be simultaneously achieved with comparatively higher specificity.

[0056]
Thus, embodiments of the method described herein may aid in identifying a greater proportion of true diffusion lesion regions from all probable diffusion lesion regions as compared to conventional segmentation techniques. Similarly, embodiments of the present specification may aid in identifying a greater proportion of the true healthy regions from all assumed healthy regions using the optimal classifier as compared to conventional segmentation techniques.

[0057]
FIGs. 3A and 3B illustrates a flow chart 300 depicting an exemplary MR imaging method for determining an optimal identifier that aids in differentiating between normal and abnormal tissues with a desired sensitivity and/or specificity. In FIG. 3A, steps 302-308 are performed in a manner that is substantially similar to steps 202-208 of FIG. 2 barring one difference. The difference between the implementations of steps 202-208 and steps 302-308 is that steps 202-208 are performed for DWI and ADC data corresponding to a single subject. Steps 302-308, however, are performed for DWI and ADC data corresponding to a plurality of subjects.

[0058]
Accordingly, at step 302, MRI data corresponding to a target volume of a plurality of subjects may be received. The received MRI data may include DWI data. Further, at step 303, at least one ADC map corresponding to each subject may be generated from the corresponding DWI data. Additionally, at step 304, one or more masks corresponding to one or more ROIs in the target volume of each of the plurality of subjects may be generated.
[0059] Further, at step 306, the one or more masks may be applied to the corresponding DWI data and the ADC map to determine a subset of DWI data and a subset of the ADC map corresponding to the one or more ROIs associated with each of the plurality of subjects. Moreover, at step 308, the subset of the DWI data and/or the subset of the ADC map corresponding to each of the plurality of subjects may be normalized using a determined diffusion weighted imaging intensity and/or a determined ADC value within the ROI of the subject.

[0060]
Referring now to FIG. 3B, at step 310, a plurality of histograms may be generated using the normalized subsets of the DWI data and/or the at least one ADC map corresponding to each of the plurality of subjects. Each histogram may be generated by plotting the normalized subset of the ADC map on a vertical axis and the normalized subset of the DWI data corresponding to each of the plurality of subjects on a horizontal axis. For example, in one embodiment, a two-dimensional (2D) histogram may be generated for each subject. The histogram may be representative of a distribution of voxels corresponding to healthy and diseased or abnormal tissues in the ROI.

[0061]
The plurality of histograms may then be combined to generate a cumulative histogram, as depicted by step 312. The cumulative histogram may aid in a binary classification of the voxels into normal and abnormal tissues. Accordingly, at step 314, an ROC analysis may be iteratively performed on the cumulative histogram with one or more classifiers. In certain embodiments, the one or more classifiers may include a plurality of linear classifiers. For example,in one embodiment, the linear classifier defined in equation (1) may be used to distinguish between normal and abnormal tissues.
ADC n for a = 90° (2)
where tan a corresponds to a slope and n corresponds to an intercept of the linear classifiers.

[0074]
Moreover, in the graphical representation 500, the vertical axis corresponds to a true positive rate or sensitivity, whereas the horizontal axis corresponds to a false positive rate (1-specificity). Further, reference numeral 502 corresponds to an ROC curve generated using an ADC-only classification (a = 0°), whereas reference numeral 504 corresponds to an ROC curve generated using normalized DWI-only classification (a = 90°). Additionally, reference numeral 506 is representative of a plurality of ROC curves generated using a combination of an ADC-based and DWI-based classification. Selection of an optimal classifier based on the analysis of the plurality of the ROC curves will be described in greater detail with reference to FIG. 6.

[0075]
Further, FIG. 6 illustrates a graphical representation 600 depicting exemplary values of sensitivity determined at desired values of specificity using the method of FIGs. 3A and 3B. Particularly, the graphical representation 600 depicts a variation in sensitivity achieved for a plurality of desired values of specificity using the method of FIGs. 3A and 3B when using a linear classifier. For example, a curve 602 corresponds to candidate linear classifiers that may provide a desired specificity of about 98 percent. Reference numeral 604 is representative of highest sensitivity value corresponding to the curve 602. The angle a corresponding to the highest sensitivity value 604 may be representative of a particular classifier in the candidate linear classifiers that provides the highest sensitivity for the desired specificity of about 98 percent. Thus, in case of the curve 602, the linear classifier represented by a = 50° corresponding to the highest sensitivity value 604 on the curve provides the highest sensitivity and may be selected as the optimal classifier.

[0076]
FIG. 7 depicts a diagrammatical illustration 700 of exemplary images 702 and 704 that are representative of an exemplary diffusion lesion segmentation in a DWI image performed using the method of FIG. 2 and a corresponding manual diffusion lesion segmentation, respectively. In particular, image 702 depicts diffusion lesion segmentation that may be achieved automatically using the method of FIGs. 2, 3A and 3B. More specifically, the diffusion lesion segmentation may be achieved by employing an optimal linear classifier determined using the ROC analysis of both ADC and normalized DWI data.

[0077]
Further, the image 704 depicts a segmentation of diffusion lesion regions 706 that are manually marked by a radiologist on the image 704. In the image 704, even though certain additional regions represented using reference numeral 708 have lower ADC values, the radiologist accurately identifies the regions 708 as normal tissues since these correspond to anisotropic regions of the white matter. As evident from the depictions of the image 702, the automated segmentation identifies lesion regions 710 that agree closely with the lesion region 706 manually marked by the radiologist. The close agreement demonstrates the robustness of the present method and system as compared to ADC-only based classification.

[0078]
Embodiments of the present specification, thus, provide systems and methods for enhanced segmentation of diffusion lesion regions following acute ischemic strokes. Particularly, embodiments described herein employ a combination of ADC and DWI data to mitigate erroneous effects of system-specific b-factor and/or q-factor variability, thus allowing for greater accuracy and consistency in assessing stroke parameters. Moreover, the embodiments of the present system and method also allow for the identification of robust classification criteria using the ROC analysis of the ADC and DWI data. The robust classification criteria may allow for accurate differentiation between the normal and abnormal tissues with a desired specificity and/or sensitivity. Accurate differentiation between the normal and abnormal tissues aids in determining an extent of diffusion lesion in the brain tissue, which in turn, may assist in determining an appropriate treatment for the patient.

[0079]
It may be noted that the foregoing examples, demonstrations, and process steps that may be performed by certain components of the present systems, for example, by the system controller 104, the processing subsystem 132, and the image processing unit 138 may be implemented by suitable code on a processor-based system, such as a general-purpose or a special-purpose computer. It may also be noted that different implementations of the present system may perform some or all of the steps described herein in different orders or substantially concurrently.

[0080]
Additionally, the functions may be implemented in a variety of programming languages, including but not limited to Ruby, Hypertext Pre-processor (PHP), Perl, Delphi, Python, C, C++, or Java. Such code may be stored or adapted for storage on one or more tangible, machine-readable media, such as on data repository chips, local or remote hard disks, optical disks (that is, CDs or DVDs), solid-state drives, or other media, which may be accessed by the processor-based system to execute the stored code.
[0081] Although specific features of various embodiments of the present system may be shown in and/or described with respect to some drawings and not in others, this is for convenience only. It is to be understood that the described features, structures, and/or characteristics may be combined and/or used interchangeably in any suitable manner in the various embodiments, for example, to construct additional assemblies and techniques for use in MRI.

[0082]
While only certain features of the present method and system have been illustrated and described herein, many modifications and changes will occur to those skilled in the art. It is, therefore, to be understood that the appended claims are intended to cover all such modifications and changes as fall within the true spirit of the claimed invention.

We claim:

1. A method for magnetic resonance imaging, comprising:

Receiving magnetic resonance imaging data corresponding to a target volume of a subject, wherein the magnetic resonance imaging data comprises diffusion weighted imaging data;
Generating at least one apparent diffusion coefficient map from the diffusion weighted imaging data;
Generating one or more masks corresponding to one or more regions of interest in the target volume;
Applying the one or more masks to the diffusion weighted imaging data and the at least one apparent diffusion coefficient map to determine a subset of the diffusion weighted imaging data and a subset of the at least one apparent diffusion coefficient map corresponding to the one or more regions of interest;
Normalizing the subset of the diffusion weighted imaging data, the subset of the at least one apparent diffusion coefficient map, or a combination thereof, using a determined diffusion weighted imaging intensity, a determined value corresponding to the at least one apparent diffusion coefficient map, or a combination thereof;
Determining an optimal classifier for classifying the one or more regions of interest with a desired specificity, a desired sensitivity, or a combination thereof, using the normalized subset of the diffusion weighted imaging data and the normalized subset of the at least one apparent diffusion coefficient map; and
Automatically segmenting the one or more regions of interest using the optimal classifier to distinguish between normal and abnormal tissues in the one or more regions of interest with the desired specificity, the desired sensitivity, or a combination thereof.

2. The method of claim 1, further comprising generating one or more diffusion weighted magnetic resonance images corresponding to the target volume of the subject, wherein the diffusion weighted imaging data is determined from the one or more diffusion weighted magnetic resonance images.

3. The method of claim 1, wherein the determined diffusion weighted imaging intensity corresponds to a ninety-eighth percentile of a diffusion weighted imaging intensity within the target volume of the subject.

4. The method of claim 1, wherein determining the optimal classifier comprises:

Receiving magnetic resonance imaging data associated with a target volume corresponding to a plurality of subjects, wherein the magnetic resonance imaging data comprises diffusion weighted imaging data;
Generating at least one apparent diffusion coefficient map corresponding to each of the plurality of subjects from the diffusion weighted imaging data;
Generating one or more masks corresponding to one or more regions of interest in the target volume corresponding to each of the plurality of subjects;
applying the one or more masks to the diffusion weighted imaging data and the at least one apparent diffusion coefficient map corresponding to each of the plurality of subjects to determine a corresponding subset of diffusion weighted imaging data and a corresponding subset of the at least one apparent diffusion coefficient map associated with the one or more regions of interest;
Normalizing the subset of the diffusion weighted imaging data, the subset of the at least one apparent diffusion coefficient map, or a combination thereof, corresponding to each of the plurality of subjects using a determined diffusion weighted imaging intensity, a determined value corresponding to the at least one apparent diffusion coefficient map, or a combination thereof;
Generating a plurality of histograms using the normalized subset of the diffusion weighted imaging data and the normalized subset of the atleast one apparent diffusion coefficient map corresponding to each of the plurality of subjects;
Combining the plurality of histograms to generate a cumulative histogram;
Iteratively analyzing receiver operating channel characteristics corresponding to the cumulative histogram using one or more classifiers; and
Selecting the optimal classifier based on the analysis of the receiver operating channel characteristics corresponding to the one or more classifiers.

4. The method of claim 4, wherein the optimal classifier corresponds to a linear classifier.

5. The method of claim 4, wherein the optimal classifier corresponds to a non-linear classifier.

7. The method of claim 4, wherein determining the optimal classifier comprises:

Generating a receiver operating channel characteristics curve corresponding to each of the one or more classifiers;
determining a slope and an intercept of the receiver operating channel characteristics curve corresponding to each of the one or more classifiers to determine a corresponding specificity and sensitivity; and
selecting a classifier having the desired specificity, the desired sensitivity, or a combination thereof, as the optimal classifier based on the determined slope and the intercept of the receiver operating channel characteristics curve corresponding to each of the one or more classifiers.

8. The method of claim 4, wherein determining the optimal classifier comprises:

Determining an area under the receiver operating channel characteristics curve for each of the one or more classifiers; and
Selecting a classifier having a largest area under the corresponding receiver operating channel characteristics curve from the one or more classifiers as the optimal classifier.

9. The method of claim 1, further comprising:

Determining a volume of the abnormal tissues in the one or more regions of interest; and
Determining a pathological condition of the subject based on the determined volume of the abnormal tissues.

10. A magnetic resonance imaging system, comprising:

A scanner configured to scan a target volume of a subject to acquire magnetic resonance imaging data;
A processing subsystem operatively coupled to the scanner and configured to:
generate diffusion weighted imaging data and at least one apparent diffusion coefficient map using the magnetic resonance imaging data;
Generate one or more masks corresponding to one or more regions of interest in the target volume;
Apply the one or more masks to the diffusion weighted imaging data and the at least one apparent diffusion coefficient map to determine a subset of diffusion weighted imaging data and a subset of the at least one apparent diffusion coefficient map corresponding to the one or more regions of interest;
Normalize the subset of the diffusion weighted imaging data, the subset of the at least one apparent diffusion coefficient map, or a combination thereof, with a determined diffusion weighted imaging intensity, a determined value corresponding to the at least one apparent diffusion coefficient map, or a combination thereof;
Determine an optimal classifier for classifying the one or more regions of interest with a desired specificity, a desired sensitivity, or a combination thereof, using the normalized subsets of the diffusion weighted imaging data and the at least one apparent diffusion coefficient map; and
Automatically segment the one or more regions of interest using the optimal classifier to distinguish between normal and abnormal tissues in the one or more regions of interest with the desired specificity, the desired sensitivity, or a combination thereof.