Claims:1. A method, comprising:
receiving dynamic contrast image data corresponding to a subject from an imaging modality, wherein the dynamic contrast image data comprises a plurality of voxels representative of signal intensity changes over time;
performing a data conditioning operation on the dynamic contrast image data to generate physiologically relevant image data;
determining a plurality of source signature signals corresponding to the physiologically relevant image data using a non-negative source separation technique;
determining a plurality of source images corresponding to the dynamic contrast image data based on the plurality of source signature signals; and
generating a diagnostic information based on at least one of the plurality of source images and the plurality of source signature signals, wherein the diagnostic information is used for detecting a health condition of the subject.
2. The method of claim 1, further comprising fitting a pharmacokinetic model for at least one of the plurality of source signature signals.
3. The method of claim 1, wherein receiving comprises acquiring four dimensional dynamic contrast enhanced-magnetic resonance imaging (DCE-MRI) data.
4. The method of claim 1, wherein the data conditioning operation comprises identifying a noise voxel among the plurality of voxels.
5. The method of claim 1, wherein the data conditioning operation comprises removing motion artifacts from the dynamic contrast image data.
6. The method of claim 1, wherein the data conditioning operation comprises determining physiologically relevant voxels among the plurality of voxels based on one or more parameters of an arterial input function (AIF).
7. The method of claim 6, wherein determining the physiologically relevant voxels comprises using one of a supervised machine learning technique, an un-supervised machine learning technique, and a deep learning technique.
8. The method of claim 1, wherein the data conditioning operation comprises excluding voxels corresponding to randomly fluctuating components based on a feature parameter determined in a frequency domain of the dynamic contrast image data.
9. The method of claim 1, wherein determining the plurality of source signature signals comprises performing source separation using a convex analysis of a mixture of non-negative sources (CAMNS) technique.
10. The method of claim 1, wherein generating the diagnostic information comprises characterizing a source signature signal as one of an AIF type, a leakage type, and a washout type based on phase information of Fourier transform of the source signature signal.
11. The method of claim 1, wherein generating the diagnostic information comprises generating a diagnostic image indicative of a tumor in an organ of interest of the subject.
12. The method of claim 1, wherein generating the diagnostic information comprises generating a source signature signal indicative of a health anomaly condition of the subject.
13. The method of claim 1, wherein generating the diagnostic information comprises determining a volume transfer coefficient indicative of vascular permeability corresponding to a source signature signal.
14. The method of claim 1, wherein generating the diagnostic information comprises determining a dominant image based on the plurality of source images and the plurality of source signature signals.
15. A system, comprising:
a data acquisition unit configured to acquire dynamic contrast image data corresponding to a subject from an imaging modality, wherein the dynamic contrast image data comprises a plurality of voxels representative of signal intensity changes over time;
a data conditioning unit communicatively coupled to the data acquisition unit and configured to perform a data conditioning operation on the dynamic contrast image data to generate physiologically relevant image data;
a source separator unit communicatively coupled to the data conditioning unit and the data acquisition unit and configured to:
determine a plurality of source signature signals corresponding to the physiologically relevant image data using a non-negative source separation technique;
determine a plurality of source images corresponding to the dynamic contrast image data based on the plurality of source signature signals;
a diagnosis unit communicatively coupled to the source separator unit and configured to generate a diagnostic information based on at least one of the plurality of source images and the plurality of source signature signals, wherein the diagnostic information is used for detecting a health condition of the subject.
16. The system of claim 15, wherein the data acquisition unit is further configured to acquire four dimensional DCE-MRI data.
17. The system of claim 15, wherein the data conditioning unit is further configured to identify a noise voxel among the plurality of voxels.
18. The system of claim 15, wherein the data conditioning unit is further configured to remove motion artifacts from the dynamic contrast image data.
19. The system of claim 15, wherein the data conditioning unit is further configured to determine physiologically relevant voxels among the plurality of voxels based on one or more parameters of an arterial input function (AIF).
20. The system of claim 19, wherein the data conditioning unit is further configured to identify the physiologically relevant voxels using one of a supervised machine learning technique, an un-supervised machine learning technique and a deep learning technique.
21. The system of claim 15, wherein the data conditioning unit is further configured to exclude voxels corresponding to randomly fluctuating components based on a feature parameter determined in a frequency domain of the dynamic contrast image data.
22. The system of claim 15, wherein the source separator unit is further configured to fit a pharmacokinetic model for at least one of the plurality of source signature signals.
23. The system of claim 15, wherein the source separator unit is further configured to perform source separation using a CAMNS technique.
24. The system of claim 15, wherein the diagnosis unit is further configured to characterize at least one of the plurality of source signature signals as one of an AIF type, a leakage type and a washout type based on phase information of the source signature signal.
25. The system of claim 15, wherein the diagnosis unit is further configured to generate a diagnostic image indicative of a tumor in an organ of interest of the subject.
26. The system of claim 15, wherein the diagnosis unit is further configured to determine a health anomaly condition in the subject based on at least one of the plurality of source signature signals.
27. The system of claim 15, wherein the diagnosis unit is further configured to determine a volume transfer coefficient indicative of vascular permeability corresponding to a source signature signal of the plurality of source signature signals.
28. The system of claim 15, wherein the diagnosis unit is further configured to determine a dominant image based on the plurality of source images and the plurality of source signature signals.
, Description:BACKGROUND
[0001] A system and a method are disclosed for analysis of dynamic data. More specifically, the disclosed techniques are related to decomposition of observed dynamic data into components corresponding to a plurality of image sources.
[0002] Medical imaging used for diagnosing and treatment of physical conditions, encompasses different non-invasive techniques to image and visualize internal structures and/or characterize functional behavior of organs and tissues within a patient. Some of the currently available imaging modalities include ultrasound systems, computed tomography (CT) systems, X-ray systems, positron emission tomography (PET) systems, single photon emission computed tomography (SPECT) systems, and magnetic resonance (MR) imaging systems. Imaging techniques such as dynamic contrast enhanced-magnetic resonance imaging (DCE-MRI), dynamic susceptibility contrast MRI (DSC-MRI), or dynamic contrast enhanced CT (DCE-CT) enable study of contrast uptake and delivery of dose. Such studies lead to understanding of perfusion characteristics and cell structures, which may be indicative of tumor properties or tissue pathology.
[0003] Dynamic data set acquired from the imaging modalities include two dimensional (2D), 2.5D (2D + time), 3D, and/or 4D (3D + time) acquisitions provide data having components corresponding to patient motion and related artifacts. Knowledge of patient motion is desirable as corrective processing may be employed using this knowledge of patient motion to enhance the quality of images and retrieve quality diagnostic information. Conventional techniques for detection and correction of patient motion include use of feature-based methods, and registration methods. Further, use of contrast agents while imaging may adversely affect the detection of motion as uptake of the contrast agent may confound visual perception of motion.
[0004] Dynamic data analysis may be conveniently performed by identifying underlying sources generating the observed data. Techniques such as independent component analysis (ICA) and non-negative matrix factorization (NMF) are employed to identify the underlying sources. While the ICA technique can result in negative sources, the NMF may not be reliable as the NMF technique may not provide a reproducible solution.
BRIEF DESCRIPTION
[0005] In accordance with one aspect of the present technique, a method includes receiving dynamic contrast image data corresponding to a subject from an imaging modality. The dynamic contrast image data comprises a plurality of voxels representative of signal intensity changes over time. The method further includes performing a data conditioning operation on the dynamic contrast image data to generate physiologically relevant image data. The method also includes determining a plurality of source signature signals corresponding to the physiologically relevant image data using a non-negative source separation technique. Further, the method includes determining a plurality of source images corresponding to the dynamic contrast image data based on the plurality of source signature signals. The method also includes generating a diagnostic information based on at least one of the plurality of source images and the plurality of source signature signals. The diagnostic information is used for detecting a health condition of the subject.
[0006] In accordance with one aspect of the present technique, a system includes a data acquisition unit configured to acquire dynamic contrast image data corresponding to a subject from an imaging modality. The dynamic contrast image data comprises a plurality of voxels representative of signal intensity changes over time. The system further includes a data conditioning unit communicatively coupled to the data acquisition unit and configured to perform a data conditioning operation on the dynamic contrast image data to generate physiologically relevant image data. The system also includes a source separator unit communicatively coupled to the data conditioning unit and the data acquisition unit and configured to determine a plurality of source signature signals corresponding to the physiologically relevant image data using a non-negative source separation technique. The source separator unit is further configured to determine a plurality of source images corresponding to the dynamic contrast image data based on the plurality of source signature signals. The system also includes a diagnosis unit communicatively coupled to the source separator unit and configured to generate a diagnostic information based on at least one of the plurality of source images and the plurality of source signature signals. The diagnostic information is used for detecting a health condition of the subject.
DRAWINGS
[0007] These and other features and aspects of the present technology 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:
[0008] FIG. 1 is a schematic representation of a system for analyzing dynamic data in accordance with an exemplary embodiment;
[0009] FIG. 2 is a schematic representation of a model for the observed images in accordance with an exemplary embodiment;
[0010] FIG. 3 is a schematic representation of an exemplary signal flow of non-negative source separation technique in accordance with an exemplary embodiment;
[0011] FIGs. 4A-4B are graphical representations illustrating effect of removal of rogue voxels on a plurality of source signature signals in accordance with an exemplary embodiment;
[0012] FIGs. 5A-5C are graphs illustrating categorization of a source signature signal in accordance with an exemplary embodiment;
[0013] FIGs. 6A-6B are images illustrating effectiveness of non-negative source separation technique for detecting a tumor in accordance with an exemplary embodiment;
[0014] FIGs. 7A-7B are graphical representations illustrating effectiveness of a non-negative source separation technique to distinguish healthy cells from non-healthy cells in accordance with an exemplary embodiment;
[0015] FIG. 8A is a graphical representation illustrating CAMNS based filtering in accordance with an exemplary embodiment;
[0016] FIGs. 8B-8C illustrate CAMNS based filtering technique in accordance with an exemplary embodiment;
[0017] FIG. 9A is a graphical representation illustrating detection of stroke in accordance with an exemplary embodiment;
[0018] FIGs. 9B-9C illustrate effectiveness of detecting stroke affected region in accordance with an exemplary embodiment;
[0019] FIGs. 10A-10C are images illustrating detection of an organ from dynamic contrast image data in accordance with an exemplary embodiment; and
[0020] FIG. 11 is a flow chart illustrating an example method for analyzing dynamic data in accordance with an exemplary embodiment.
DETAILED DESCRIPTION
[0021] Embodiments of a method and a system for analysis of dynamic data are disclosed. More specifically, the disclosed techniques are related to decomposition of observed dynamic data into components corresponding to a plurality of image sources. The technique includes receiving dynamic contrast image data corresponding to a subject from an imaging modality. The dynamic contrast image data is processed using a data conditioning operation to generate physiologically relevant image data. A plurality of source signature signals corresponding to the physiological data is generated using a non-negative source separation technique. A plurality of source images corresponding to the plurality of source signature signals is determined based on the dynamic contrast image data. Further, a diagnostic information is generated based on at least one of the plurality of source images and the plurality of source signature signals. The diagnostic information is used for detecting a health condition of the subject.
[0022] The term “dynamic data” refers to time varying image data set having a plurality of 2 dimensional (2D) images having a plurality of pixels, or 2.5D image data set having a plurality of 2D image volumes indexed by time or a plurality of 3D image volumes having a plurality of voxels, or 4D image data sets having a plurality of 3D image volumes indexed by time. The term “dynamic contrast image data” refers to dynamic data acquired from a perfusion scan. The term “perfusion” refers to passage of fluid through the lymphatic system or the blood vessels to an organ or a tissue. The term “arterial input function (AIF)” refers to perfusion response recorded at a tissue when a contrast agent is introduced in the circulating fluid. A graph representative of the AIF as a function of time characterized by an initial peak response is referred herein as “bolus response” and a reduced flat response at a later portion referred herein as “washout phase.” The initial “kick-in” of the bolus response occurs at a time referred herein as ‘bolus arrival time’ (BAT). The term “voxel signal” refers to a time series representative of evolution of a voxel in dynamic data. The term “source decomposition” or “source separation” refers to decomposition of the dynamic data into a plurality of source images and a plurality of source signature signals. Each of the image frames of the dynamic data is a linear combination of the plurality of source images weighted by samples of the plurality of source signature signal at a specified time instant. The term ‘diagnostic information’ includes at least one of tissue characteristic information, a health condition related information, a prediction information related to the health condition in the form of a parameter, a signal, an image or a statistic.
[0023] FIG. 1 is a schematic representation 100 illustrating a system 102 for analyzing dynamic data in accordance with an exemplary embodiment. The schematic 100 includes an imaging modality 104 acquiring dynamic contrast image data, represented generally by reference numeral 110, from a subject 106. In one embodiment, the imaging modality 104 is a dynamic contrast enhancing magnetic resonance imaging (DCE-MRI) machine. The dynamic contrast image data 110 is a four dimensional dynamic contrast enhanced magnetic resonance imaging (DCE-MRI) data. In other embodiments, the dynamic contrast image data is one of a 2.5D or four dimensional dynamic susceptibility contrast magnetic resonance imaging (DSC-MRI) data or dynamic positron emission tomography (PET) or dynamic ultrasound. The imaging modality may be a DCE-CT system or a DSC-MRI system or dynamic-PET. The system 102 is configured to receive the dynamic contrast image data 110 and generate a diagnostic information 130 which is further used for detecting a health condition of the subject 106. The system 102 includes a data acquisition unit 112, a data conditioning unit 114, a source separator unit 116, and a diagnosis unit 118. The system 102 further includes a processor unit 120 and a memory unit 122 communicatively coupled with each other through a communication bus 132.
[0024] The data acquisition unit 112 is communicatively coupled to the imaging modality 104 and configured to acquire the dynamic contrast image data 110. In one embodiment, the dynamic contrast image data 110 is dynamic contrast enhancing image data acquired from an image modality such as, but not limited to, an MRI machine, a CT machine, a PET machine, an ultrasound machine, or combinations thereof. By way of example, the dynamic contrast enhancing image data may be acquired from a PET-CT machine, or a PET-MR machine. The dynamic contrast image data includes a plurality of images corresponding to a plurality of time instants. Each of the plurality of the images includes a plurality of voxels representative of signal intensity changes over time. In one embodiment, the signal intensity changes may be related to concentration values of a contrast agent introduced intravenously into the subject 106. The concentration values are indicative of concentration of the contrast agent corresponding to the time instant at which the image is acquired at a specified spatial location. In some embodiments, the plurality of voxels may be representative of time-course changes of some derived normalized quantitative properties such as concentration of contrast agent. In one embodiment, the plurality of voxels is representative of a physiological parameter. The physiological parameter may be an intravascular parameter, such as but not limited to, perfusion, relative blood volume and mean transit time or an extravascular parameter such as, but not limited to, vascular permeability and the extravascular volume.
[0025] The data conditioning unit 114 is communicatively coupled to the data acquisition unit 112 and configured to perform a data conditioning operation on the received dynamic contrast imaging data 124. The data conditioning unit 114 is further configured to determine physiological data set having a plurality of physiologically relevant voxels 126 among the plurality of voxels based on one or more parameters of arterial input function (AIF). In one embodiment, the data conditioning unit 114 is configured to identify a noise voxel among the plurality of voxels. It may be noted that hereinafter the terms “rogue voxel” and “noise voxel” are used equivalently and interchangeably. The rogue voxel in the dynamic contrast image data manifests as a noisy signal in the plurality of source signature signals. In another embodiment, the data conditioning unit 114 is configured to remove motion artifacts from the dynamic contrast image data. The motion artifacts are generated from voluntary and/or involuntary motion of the subject during acquisition of the imaging data and manifests as a step change or spikes in one or more of the plurality of source signature signals.
[0026] In one embodiment, the physiologically plausible voxels 126 are identified in the voxel signal based on a bolus arrival time value (BAT). The voxel signal corresponding to AIF is considered to have a physiological origin if time instant at which a peak value is observed corresponds to the BAT. Any spike in the washout phase of the voxel signal is rejected as being originating from artifacts. Further, any step change or spikes in the voxel signal are considered to be from motion from the subject.
[0027] In certain embodiments, the data conditioning unit 114 is further configured to exclude voxels corresponding to randomly fluctuating components based on a feature parameter determined in a frequency domain of the dynamic contrast image data. In one embodiment, a first feature parameter and a second feature parameter are determined based on a frequency spectrum of a voxel signal from the dynamic contrast image data. The frequency spectrum includes a main lobe and a plurality of side lobes. The first feature parameter refers to a ratio of a peak magnitude value of the frequency spectrum to a variance of the plurality of side lobes of the frequency spectrum. The second feature parameter refers to logarithm of a ratio of an area of a main lobe of the spectrum to an area of a side lobe among the plurality of side lobes of the spectrum. If the first feature parameter has a relatively higher value than the second feature parameter, the voxel signal is considered to be a physiological signal. A ratio of the first feature parameter to the second feature parameter is compared with a pre-determined threshold to determine the nature of the voxel signal. Further, if a value of the second feature parameter is negative, the voxel corresponding to that particular voxel signal is considered as a rogue voxel and is excluded from further processing. In one example, for a given voxel, the first feature parameter has a value of +26.68 and the second feature parameter has a value of -22.71. In this example, the second feature parameter is negative, hence the given voxel is considered to be a rogue voxel. In another example, for another given voxel, the first feature parameter has a value of +913.73 and a second feature parameter has a value of +5.474, the voxel signal corresponding to this given voxel is considered to be a physiological signal as the first feature parameter is very high and the second feature parameter is a non-negative number. In another embodiment, the rouge voxels may be identified by using one of a supervised machine learning technique, an un-supervised machine learning technique or a deep learning technique or combinations thereof (e.g. deep learning followed by machine learning). In the supervised learning technique, initially a voxel classifier is determined based on training with a plurality of training voxels identified apriori with labels of “good voxel”(that is, physiologically relevant voxels) and “rogue voxel.”. Subsequently, when a new voxel is received, the voxel classifier automatically classifies the voxel as either a good voxel or a bad voxel.
[0028] The source separator unit 116 is communicatively coupled to the data conditioning unit 114 and configured to generate a decomposed data set 128. In one embodiment, the decomposed data set 128 is determined based on the physiologically relevant voxels 126. In an alternative embodiment, the decomposed data set 128 is determined based on the received dynamic contrast imaging data 124. The decomposed data set includes a plurality of source signature signals corresponding to a plurality of underlying image sources. The decomposed data set 128 also includes a plurality of source images corresponding to the plurality of source signature signals. In one embodiment, the source separator unit 116 is further configured to perform source separation using a convex analysis of mixture of non-negative source (CAMNS) technique. In one embodiment, the plurality of source signature signals is determined using CAMNS based on the physiologically relevant voxels 126. Further, in this embodiment, the plurality of source images is determined based on the received dynamic contrast imaging data 124 using least squares minimization technique. It may be noted that use of the physiologically relevant voxels 126 allows robust determination of source signature signals and the consequent least squares minimization technique provides a filtering effect on the plurality of source signature signals. In the same embodiment, imaging data may be reconstructed based on the plurality of source signature signals and the plurality of source images. It may be noted that the reconstructed imaging data is a filtered version of the dynamic contrast imaging data 124.
[0029] The source separator unit 116 is further configured to fit a pharmacokinetic model for at least one of the plurality of source signature signals. The pharmacokinetic model may be in the form of a pharmacokinetic equation and is representative of tracer concentration. In one embodiment, the Toft equation is used as a pharmacokinetic equation. The Toft equation is given by Equation (1):
Equation (1)
where, C(t) is representative of concentration at time t, ? is tissue density, Hct is representative of Hematocrit value indicative of the proportion of total blood volume that is composed of red blood cells. The term Cp is representative of plasma concentration, k2 is representative of outflux constant, K1 is blood to tissue transportation constant and vp is representative of a plasma volume. In another embodiment, a modified gamma variate may be used as the pharmacokinetic equation. The modified gamma variate function is given by Equation (2):
Equation (2)
where, C(t) is representative of concentration at time t, Cp is representative of peak concentration from recorded dilution curve, Ta is arrival time, a, ß are parameters of gamma variate distribution function. The source separator unit 116 is further configured to generate a plurality of modified source signature signals based on fitting of pharmacokinetic model to at least one of the source signature signals. In other embodiment, a plurality of modified source signature signals may be determined by fitting a physiological model for at least one of the plurality of source signature signals. The plurality of modified source signature signals have reduced noise components compared to the plurality of source signature signals. Consequently, determination of source images with modified source signature signals has reduced noise. In embodiments disclosed herein, either of the plurality of source signature signals or the plurality of modified source signature signals may be used for further processing.
[0030] The source separator unit 116 is further configured to generate a plurality of source images corresponding to the dynamic contrast image data based on the plurality of source signature signals. In one embodiment, the source separator unit 116 is configured to generate the plurality of source images based on the plurality of modified source signature signals. It may be noted that one or more of the plurality of source signature signals, the plurality of modified source signature signals may be representative of a vascular input function of a vascular region in the dynamic contrast image data. In one embodiment, at least one of the plurality of source signature signals may be representative of an arterial input function of an artery in a dynamic contrast image data.
[0031] The diagnosis unit 118 is communicatively coupled to the source separator unit 116 and configured to generate the diagnostic information 130 indicative of a health condition of the subject 106. In one embodiment, the diagnosis unit 118 is configured to generate a diagnostic image representative of an abnormality such as, but not limited to, a tumor in an organ of interest of the subject 106. In another embodiment, the diagnosis unit 118 is configured to determine a health anomaly condition in the subject 106 based on at least one of the plurality of source signature signals. The diagnosis unit 118 is further configured to characterize the at least one source signature signal as one of an AIF type, a leakage type and a washout type based on phase information of the frequency domain representation of the at least one source signature signal. For example, when the phase information includes a saw tooth type of signal, the source signature signal is categorized as an AIF-curve. In another example, when the phase information resembles a step response, the source signature signal is categorized as a leakage-curve. In another example, when the phase information exhibits transition from step response to saw tooth response, the source signature signal is categorized as a washout-curve. In another embodiment, the diagnosis unit 118 is further configured to determine a volume transfer coefficient indicative of vascular permeability corresponding to a source signature signal. In another embodiment, the diagnosis unit 118 is further configured to determine a dominant image based on the plurality of source images and the plurality of source signature signals.
[0032] In some embodiments, when at least one of the plurality of source signature signals is modelled as a pharmacokinetic equation, a parameter representative of transfer rate constant from the intravascular system to the extravascular-extracellular region, denoted by symbol Ktrans, is determined. In other embodiments, other posteriori statistical technique such as analysis of variance (ANOVA), or Scheffe’s method may be employed to determine the diagnostic information. It may be noted that techniques such as regression analysis and significance tests may also be used to determine the diagnostic information. Statistical tests having a test statistic of F-distribution under null hypothesis may be used to compare statistical models. In one embodiment, a Ktrans map corresponding to the reconstructed imaging data may be obtained as the diagnostic information 130.
[0033] In certain embodiments, the diagnosis unit 118 is configured to identify an organ, such as but not limited to, a liver, based on the plurality of source signature signals and the plurality of source images. In instances where the liver is identified as the organ of interest by the diagnosis unit 118, a plurality of dominant images corresponding to the plurality of source images is determined. In one embodiment, determining the plurality of dominant images includes determining a sum image is formed as a sum of the plurality of source images. Further, each dominant image of the plurality of dominant images is determined as a ratio of pixel values of a corresponding source image and corresponding pixel values of the sum image. Further, a plurality of 2-norm values corresponding to the plurality of source signature signals is determined. A maximum 2-norm value is determined among the plurality of 2-norm values. A source signature signal among the plurality of plurality of source signature signals corresponding to the maximum 2-norm value is selected. Subsequently, the liver organ is detected based on a dominant image corresponding to the selected source signature signal.
[0034] The processor unit 120 may include one or more processors. The terms ‘processor’, ‘one or more processors’ and ‘processor’ are used equivalently and interchangeably throughout this application. The one or more processors include at least one arithmetic logic unit, a microprocessor, a general purpose controller, or a processor array to perform the desired computations or run the computer program. In certain embodiments, the processor unit 120 may be configured to acquire the dynamic contrast image data 110. In same or different embodiments, the processor unit 120 may be configured to perform the data conditioning operations on the dynamic contrast image data 110. Further, the processor unit 120 may be configured to generate a plurality of source signature signals. In one embodiment, the processor unit 120 may be configured to generate the diagnostic information 130 indicative of a health condition of the subject 106. The functionality of the processor unit 120 may be limited to one or more of the data acquisition unit 112, the data conditioning unit 114, the source separator unit 116, and the diagnosis unit 118. While the processor unit 120 is shown as a separate unit, there can be a processor co-located or integrated in one or more of the units 112, 114, 116, and 118. Alternatively, one or more processors can be local or remote, such as a central server or cloud based, and the communications link may be a computer bus, a wired link, or a wireless link or a combination thereof.
[0035] The memory unit 122 may be a non-transitory storage medium. For example, the memory unit 122 may be a dynamic random access memory (DRAM) device, a static random access memory (SRAM) device, flash memory or other memory devices. In one embodiment, the memory unit 122 may include a non-volatile memory or similar permanent storage device, media such as a hard disk drive, a floppy disk drive, a compact disc read only memory (CD-ROM) device, a digital versatile disc read only memory (DVD-ROM) device, a digital versatile disc random access memory (DVD-RAM) device, a digital versatile disc rewritable (DVD-RW) device, a flash memory device, or other non-volatile storage devices. A non-transitory computer readable medium may be encoded with a program to instruct the one or more processors to perform analyze the dynamic contrast image data using a blind source separation techniquede.
[0036] In certain embodiments, one or more of the data acquisition unit 112, the data conditioning unit 114, the source separator unit 116, and the diagnosis unit 118 may be stored in the memory unit 122 and may be executable by the at least one processor 120. At least one of the units 112, 114, 116, 118 may be standalone hardware components. Other hardware implementations such as field programmable gate arrays (FPGA), application specific integrated circuits (ASIC) or customized chip may be employed for one or more of the units of the system 102.
[0037] FIG. 2 is a schematic representation 200 of a model for the observed images in accordance with an exemplary embodiment. The schematic 200 includes a plurality of observed image frames 202 at different instants of time. In the illustrated embodiment, the plurality of observed image frames are extracted from dynamic contrast image data. In an example, thirty image frames are considered. The data model corresponding to the observed image frames 202 is given by Equation (3):
Equation (3)
where, Y is the observation matrix, S is the matrix generated based on source signature signals, and W is representative of source images. The matrix S is also referred herein as mixing matrix and the matrix W is referred as ‘weight matrix’. The dimension of the observation matrix is KoXN with N voxels in each of the plurality of observed image frames, and Ko is the number of observed image frames. The dimension of the mixing matrix S is KoXKw, where Kw is the number of source images and the dimension of the weight matrix W is KwXN. As an example, if the number of observed images Ko=30, and the number of sources Kw=3, in a plurality of signature signals 204, each signature signal having Ko samples on time axis 212 are represented as columns of the matrix S. Further, a plurality of source images 206 having the same dimensions as that of the observed image frames are used to generate the matrix W. An observed voxel vector 208 representative of time evolution of a voxel (or a voxel signal) forms a column in the matrix Y. Similarly, a source image 210 forms a column in the matrix W. In one exemplary embodiment, the convex analysis of mixture of non-negative (CAMNS) technique is used to determine a plurality of source signature signals. In one embodiment, the CAMNS technique includes determining a convex hull of the observation image frames. Further, the CAMNS technique includes determining a plurality of extreme points of a convex hull. In one embodiment, the plurality of extreme points is determined using a simplex volume maximization technique. The extreme points of the convex hull determine the plurality of source signature signals. In another embodiment, linear programming (LP) techniques are used to determine the extreme points of the convex hull. Further, the matrix S is determined based on the plurality of source signature signals. A plurality of source images are determined based on the plurality of observed dynamic contrast image data and the plurality of the source signature signals. In one embodiment, a constrained optimization operation such as non-negative least squares (NNLS) is performed for obtaining W. The columns of the matrix are used to reconstruct the plurality of source images.
[0038] FIG. 3 is a schematic representation illustrating a signal flow 300 of a non-negative source separation technique in accordance with an exemplary embodiment. The signal flow 300 is initiated with dynamic contrast image data having a plurality of voxels representative of signal intensity changes over time 302. The plurality of voxels is processed and data conditioning is performed to generate physiologically relevant image data 304. The data conditioning includes selection of a region of interest within the dynamic contrast image data. The data conditioning also includes noise (or artifact) reduction in the plurality of voxels to significantly reduce the complexity of subsequent source separation processing. In one embodiment, the noise reduction includes identification and exclusion of rogue voxels. In same or different embodiment, the noise reduction includes identification of voxels affected by motion of the subject. Further, the motion affected voxels may be excluded from further processing, or modified to reduce the effect of motion.
[0039] A non-negative source separation technique is used to process the physiologically relevant image data 304. In one embodiment, the non-negative source separation technique is a convex analysis mixture non-negative source separation (CAMNS) 306 technique performed by the source separator unit (116 of FIG. 1). The CAMNS technique generates a plurality of source signature signals 308 representative of temporal patterns of weights used for mixing the plurality of source images to generate the plurality of observed image frames. Further, a plurality of source images 312 are estimated at source identification block 310 performed by the source separator unit (116 of FIG. 1) based on the plurality of voxels and the plurality of source signature signals. The plurality of source images 312 and the plurality of source signature signals 308 are suitably combined at block 314 to generate a plurality of components 316 of the observed dynamic image data. In one embodiment, at least one of the plurality of source signature signals 308 is used to determine the diagnostic information. In another embodiment, at least one of the plurality of source images 312 is used to determine the diagnostic information. In another embodiment, at least one of the plurality of components 316 is used to determine the diagnostic information. In another embodiment, one or more of the plurality of source signature signals 308, the plurality of source images 312 and the plurality of components 316 are combined to generate the diagnostic information.
[0040] FIGs. 4A-4B are graphs illustrating effect of removal of rogue voxels on the plurality of source signature signals in accordance with an exemplary embodiment. The graph of FIG. 4A includes an x-axis 402 representative of time index and a y-axis 404 representative of amplitude. The graph of FIG. 4A also includes a plurality of source signature signals 408 exhibiting temporal fluctuations in a portion 406 of the plurality of source signature signals. The fluctuations are due to presence of rogue voxels in the dynamic contrast data set. The graph of FIG. 4B includes x-axis 410 that is representative of time index and y-axis 412 that is representative of amplitude. The graph of FIG. 4B includes a plurality of source signature signals 414 obtained based on modified dynamic contrast data with rogue voxels excluded. It may be observed that temporal fluctuations in the portion 406 is reduced and the plurality of source signals 414 are representative of physiological characteristics that are typically noticed in tumor pathology.
[0041] FIGs. 5A-5C are graphs illustrating categorization of a source signature signal in accordance with an exemplary embodiment. In the illustrated embodiment, the source signature signal is characterized based on a phase response of Fourier transform of the source signature signal. FIG 5A is a graph 502 representative of a source signature signal representative of AIF. FIG. 5A also includes a phase response 508 of Fourier transform of the source signature signal 502. The phase response 508 is a regular saw tooth curve characterizing the AIF-type of source signature signal. FIG 5B is a graph 504 representative of a source signature signal representative of a leakage function. FIG. 5B also includes a phase response 510 of Fourier transform of the source signature signal 504. The phase response 510 is a step response curve characterizing the leakage-type of a source signature signal. FIG 5C is a graph 506 representative of a source signature signal representative of a washout function. FIG. 5C also includes a phase response 512 of Fourier transform of the source signature signal 506. The phase response 128 exhibits characteristics of saw tooth curve and step response curve characterizing the washout-type of source signature signal.
[0042] FIG. 6A and FIG. 6B are images illustrating effectiveness of the non-negative source separation technique for detecting a tumor in accordance with an exemplary embodiment. FIG. 6A is a source image 602 generated based on dynamic contrast enhancing MRI data corresponding to prostrate region of a subject. The CAMNS technique is used to generate the plurality of source signature signals and corresponding plurality of source images. The source image corresponding to a washout signature signal includes a high blood volume tumor region 604. FIG. 6B illustrates a histopathological image 606 acquired from the subject. A region 608 in the histopathological image 606 is representative of the tumor region. It may be noted that the region 604 determined from the disclosed technique has a good correlation with the region 608 determined from the histopathological studies.
[0043] FIG. 7A and FIG. 7B are graphs illustrating effectiveness of the non-negative source separation technique to distinguish healthy cells from non-healthy cells in accordance with an exemplary embodiment. The graphs of FIGs 7A-7B are obtained using posteriori statistical techniques designed to find patterns and/or relationships between subgroups of sampled populations. FIG. 7A is a graph 700 illustrating performance of the posteriori statistical technique corresponding to the DCE-MRI data. The graph 700 includes x-axis 702 representative of bar charts corresponding to tumor and contralateral side data and a y-axis 704 representative of DCE data values. Further, the graph 700 shows a first bar chart 706 corresponding to observation image data in area of prostrate with tumor and a second bar chart 708 corresponding to the contralateral portion of the observation image data. FIG.7B is a graph 710 illustrating the performance of the posteriori statistical technique corresponding to the plurality of source signature signals. The graph 710 includes x-axis 712 representative of bar charts corresponding to tumor and contralateral side data and a y-axis 714 representative of tracer concentration values. Further, the graph 710 includes a first bar chart 716 representative of tracer concentration determined based on the washout signature signal corresponding to the tumor portion and a second bar chart 718 representative of tracer concentration determined based on the washout signature signal corresponding to the contralateral portion. The tracer concentration changes are modelled as pharmacokinetic equation and an estimation of the tracer concentration in the voxels (Ktrans values) are determined. In FIG. 7B, the first bar chart is more pronounced compared to the second bar chart indicating generation of enhanced diagnostic information by the source separated signals in comparison with the diagnostic information generated by the DCE-MRI data.
[0044] FIGs. 8A-8C illustrate CAMNS based filtering technique in accordance with an exemplary embodiment. FIGs 8A-8C are obtained based on a representative dynamic contrast enhanced MRI acquisition corresponding to liver organ. FIG. 8A is graph 810 illustrating filtering of a concentration curve corresponding to a selected voxel in a dynamic contrast image data using CAMNS technique. The graph 810 includes x-axis 802 representative of time and y-axis representative of amplitude. The graph 810 further includes a first curve 806 having random fluctuations and a second curve 808 having smooth characteristics. The first curve 806 corresponds to a concentration curve obtained from the received dynamic contrast image data and the second curve 808 corresponds to a concentration curve of a corresponding voxel in a reconstructed dynamic contrast image data. It may be observed that the second curve 808 is a CAMNS filtered version of the first curve 806.
[0045] FIG. 8B illustrates an image Ktrans map generated by pharmacokinetic modelling of the dynamic contrast image data. The image of FIG. 8B includes a voxel 812 which is used to generate the graph of FIG. 8A. The image of FIG. 8B also illustrates a region 814 having structures masked by noise and other artifacts. FIG. 8C illustrates an image Ktrans map generated by pharmacokinetic modelling of the reconstructed dynamic contrast image data. The image of FIG. 8C illustrates the region 816 corresponding to the region 814 of FIG. 8B. The region 816 exhibits structural details that are not seen in the image of FIG. 8B. An R2 metric representative of goodness of fit for pharmacokinetic modelling is evaluated for the images of FIGs. 8B-8C. The R2 metric for the image of FIG. 8B is 0.99 and the R2 metric for the image of FIG. 8C is approximately 0.7 indicating effectiveness of CAMNS based filtering for modelling purposes.
[0046] FIGs.9A-9C illustrate effectiveness of detecting stroke affected region in accordance with an exemplary embodiment. FIG. 9A illustrates a graph 900 of delayed source signature signal representative of stroke affected region in accordance with an exemplary embodiment. The graph 900 illustrates a diagnostic information useful in detecting stroke affected area. The graph 900 includes an x-axis 902 representative of time samples and a y-axis 904 representative of amplitude values. The graph 900 includes a plurality of curves 906, 908, 910 representative of a first source signature signal, a second source signature signal and a third source signature signal determined using the disclosed technique. The first source signature signal 906, and the second source signature signal 908 are representative of healthy tissues in the image. The curve 910 corresponding to the third source signature signal is delayed in time and is representative of stroke affected tissues in the image.
[0047] FIG. 9B is an image 916 representative of TMAX map used by radiologists to identify stroke an affected region in the brain from dynamic contrast image data. The image 916 includes a region 912 affected by stroke. FIG. 9C is an image 918 representative of a dominant image corresponding to the third source signature signal 910. The dominant image 910 is determined based on a plurality of source images corresponding to the plurality of sources signature signals 906, 908, 910 generated based on the CAMNS technique. The dominant image 910 corresponding to the delayed source signature signal 910 includes a region 914 representative of stroke affected region. It may be observed that the region 914 in the dominant image 918 has sharper boundaries compared to the region 912 of the TMAX map image 916.
[0048] FIGs. 10A-10C are images illustrating detection of an organ from dynamic contrast image data in accordance with an exemplary embodiment. In the illustrated embodiment, the dynamic contrast image data corresponds to a liver organ. The dynamic contrast image is processed using the convex analysis of mixture of non-negative (CAMNS) technique to generate three source signature signals and corresponding three source images. A plurality of dominant images is determined as explained in a previous paragraph based on the three source images. FIG. 10A is representative of a first dominant image corresponding to a first source image. The first dominant image provides an estimate of dominance of the first source image at a plurality of voxel positions. FIG. 10B is representative of a second dominant image corresponding to a second source image at the plurality of voxel positions. The second dominant image provides an estimate of dominance of the second source image. FIG. 10C is representative of a third dominant image corresponding to a third source image. The third dominant image provides an estimate of dominance of the third source image at the plurality of voxel positions. It may be observed that liver organ 1000 is detectable from FIG 10C. In one embodiment, a plurality of standard image processing techniques, such as but not limited to, thresholding, detecting largest connected component, and hole-filling, are used to detect the organ 1000.
[0049] FIG. 11 is a flow chart 1100 illustrating a method for analyzing dynamic data in accordance with an exemplary embodiment. The method includes receiving dynamic contrast image data corresponding to a subject from an imaging modality in step 1102. The dynamic contrast image data includes a plurality of voxels representative of concentration values. In one embodiment, the receiving step further comprises acquiring four dimensional dynamic contrast enhancement magnetic resonance imaging (DCE-MRI) data. The method further includes performing the data conditioning operation on the dynamic contrast image data to generate physiologically relevant image data in step 1104. In one embodiment, the data conditioning operation includes identifying a noise voxel among the plurality of voxels. In another embodiment, the data conditioning operation includes removing motion artifacts from the dynamic contrast image data. Further, embodiments of the disclosed technique includes the data conditioning operation for determining physiologically relevant voxels among the plurality of voxels based on one or more parameters of arterial input function. In a further embodiment, the data conditioning operation also includes excluding voxels corresponding to randomly fluctuating components based on a feature parameter determined in a frequency domain of the dynamic contrast image data.
[0050] The method further includes determining a plurality of source signature signals corresponding to the physiologically relevant image data using a non-negative source separation technique in step 1106. In one embodiment, determining the plurality of source signature signals includes performing source separation using a convex analysis of mixture of non-negative sources (CAMNS) technique. In other embodiments, other similar techniques such as independent component analysis (ICA), or non-negative matrix factorization (NMF) or their variants may be used for determining the plurality of source signature signals.
[0051] The method also includes determining a plurality of source images corresponding to the dynamic contrast image data based on the plurality of source signature signals in step 1110. In an alternative embodiment, the method includes fitting a pharmacokinetic model to at least one of the plurality of source signature signals at step 1108 to determine a plurality of modified source signature signals. In the same embodiment, the plurality of modified signature signals are used to determine the plurality of sources images at step 1110. In one embodiment, the plurality of source signature signals are used to determine a mixing matrix, subsequently a pseudo inverse of the mixing matrix is determined. The pseudo inverse matrix is post multiplied with the observation data matrix to get the source matrix. A plurality of source images is derived from the source matrix. In one embodiment, a plurality of component signals are determined by multiplying the plurality of source images with the corresponding source signature signals.
[0052] The method further includes generating a diagnostic information based on at least one of the plurality of source images and the plurality of source signature signals as illustrated in step 1112. The diagnostic information is used for detecting a health condition of the subject. In one embodiment, determining the diagnostic information includes characterizing the source signature signal as one of an AIF type, a leakage type and a washout type based on phase information of Fourier transformation of the source signature signal. In another embodiment, determining the diagnostic information includes generating a diagnostic image indicative of an abnormality (such as a tumor) in an organ of interest of the subject. In a further embodiment, the source signature signal is used to detect a health anomaly condition in the subject. In another embodiment, the diagnostic information is determined based on a volume transfer coefficient indicative of vascular permeability corresponding to the source signature signal. In yet another embodiment, the diagnostic information is determined by determining a dominant image based on plurality of source images and the plurality of source signature signals.
[0053] It is to be understood that not necessarily all such objects or advantages described above may be achieved in accordance with any particular embodiment. Those skilled in the art will recognize that the systems and techniques described herein may be embodied or carried out in a manner that achieves or improves one advantage or group of advantages as taught herein without necessarily achieving other objects or advantages as may be taught or suggested herein.
[0054] While the technology has been described in detail in connection with only a limited number of embodiments, it should be readily understood that the claimed inventions are not limited to such disclosed embodiments.