Identification and analysis of lesions in medical imaging

Abstract

A method for automated classification of curve patterns associated with dynamic image data of a lesion in a subject in order to determine characteristics of the lesion. The method comprising the steps of loading the image data into an electronic memory means, producing a plot of signal intensity profile, converting the signal intensity profile into a contrast enhancement profile, detecting a reference enhancement profile having a highly positive slope over an initial enhancement period followed by a decreasing profile portion, deriving signature curve types based on the reference enhancement profile, classifying an enhancement curve for each pixel in a selected ROI into one of the derived signature curve types using all available time points and displaying a grid-plot of the classified enhancement curves for all pixels in the selected ROI, wherein the overall display of curves and heterogeneity provides visual indication of the characteristics of the lesion.

Claims

1. A method for automated classification of curve patterns associated with dynamic image data of a lesion in a subject in order to determine characteristics of the lesion by operating a computer program in a computer, comprising the steps of: (a) loading the image data into an electronic memory means; (b) producing a plot of signal intensity profile; (c) converting the signal intensity profile into a contrast enhancement profile; (d) detecting a reference vascular enhancement profile having a highly positive slope over an initial enhancement period followed by a decreasing profile portion; (e) deriving signature curve types based on the reference vascular enhancement profile; (f) classifying an enhancement curve for each pixel in a selected region of interest (ROI) into one of the derived signature curve types using all available time points based on correlation analysis; and (g) displaying a characteristic grid-plot of the individually classified enhancement curves of all pixels within the selected ROI, wherein the distribution pattern and heterogeneity of the curve types provide visualization of the characteristics of the lesion.

2. A method according to claim 1 wherein prior to detecting a reference vascular enhancement profile, the method comprises the steps of detecting a pre-contrast baseline and the initial enhancement period by said computer program: (a) calculating the mean signal intensity of the entire volume of contrast at each time point to obtain a mean signal intensity time-curve S(t); (b) estimating a mean enhancement time-curve Y(t)=S(t)S(t.sub.0), where t.sub.0 represents the time point for a first pre-contrast scan; (c) finding a maximum uptake slope at time point t.sub.s and a maximum enhancement Y.sub.max=max{Y(t)}; (d) searching backward from t.sub.s to find the first occurring time point t.sub.1 such that t.sub.0t.sub.1<t.sub.s and Y(t.sub.1)<a %*Y.sub.max where a % is a fraction value that is adjustable; and (e) setting the initial enhancement period between t.sub.1 and t.sub.2=t.sub.s+(t.sub.st.sub.1) and find an initial enhancement index time point at t.sub.2.

3. A method according to claim 1 wherein the step of detecting the reference vascular enhancement profile includes: (a) calculating a mean baseline image from all available time points; (b) calculating an initial enhancement index map Y.sub.i, and a washout map Y.sub.O representing the decreasing profile portion; (c) finding all pixels collectively noted as XYZ such that Y.sub.i(XYZ)>f*max(Y.sub.i) and Y.sub.O(XYZ)>0, where f is an adjustable masking threshold.

4. A method according to claim 1 wherein for each pixel with an initial enhancement above a predetermined threshold, the step of correlation analysis includes a rank correlation with respect to the signature curves in order to obtain correlation coefficients and statistical p-values. The enhancement curve is classified into the signature curve type that corresponds to the maximum correlation coefficient value.

5. A method according to claim 1 wherein prior to the displaying step the method includes applying segmentation analysis for the lesion and providing information including the lesion volume and the percentage contribution from pixels with each type of enhancement curve respectively.

6. A method according to claim 5 wherein the results of the segmentation analysis and curve classification of all pixels are stored in a color-coded curve pattern map for display as a color image overlay on top of a raw image of the lesion.

7. A method according to claim 1 wherein the enhancement curves of all pixels displayed in the grid-plot are highlighted by different color and line-thickness to reflect the type of enhancement curve and whether it is statistically significant respectively for visual identification.

8. A method according to claim 1 wherein the pattern and heterogeneity of enhancement curve types in the lesion reflect the characteristics of the lesion including benign or malignant tumor.

9. A computer program embodied on a computer-readable medium for automated classification of curve patterns associated with dynamic image data of a lesion in a subject, wherein the computer program instructs a processor to: (a) load the image data into an electronic memory means; (b) produce a plot of signal intensity profile; (c) convert the signal intensity profile into a contrast enhancement profile; (d) detect a reference vascular enhancement profile having a highly positive slope over an initial enhancement period followed by a decreasing profile portion; (e) derive signature curve types based on the reference vascular enhancement profile; (f) classifying an enhancement curve for each pixel in a selected region of interest (ROI) into one of the derived signature curve types using all available time points based on correlation analysis; and (g) displaying a characteristic grid-plot of the individually classified enhancement curves of all pixels within the selected ROI, wherein the distribution pattern and heterogeneity of the curve types provide visualization of the characteristics of the lesion.

10. A system for automated classification of curve patterns associated with dynamic image data of a lesion in a subject in order to determine characteristics of the lesion, the system comprising: A scanner for providing a dynamic imaging scan of the subject; A processor linked to the scanner for retrieving the image data from the scan; the processor further by operating a computer program in a computer: (a) loads the image data into an electronic memory means; (b) produces a plot of signal intensity profile; (c) converts the signal intensity profile into a contrast enhancement profile; (d) detects a reference vascular enhancement profile having a highly positive slope over an initial enhancement period followed by a decreasing profile portion; (e) derives signature curve types for classification of the characteristics based on the reference vascular enhancement profile; (f) classifying an enhancement curve for each pixel in a selected region of interest (ROI) into one of the derived signature curve types using all available time points based on correlation analysis; and (g) displaying a characteristic grid-plot of the individually classified enhancement curves of all pixels within the selected ROI, wherein the distribution pattern and heterogeneity of the curve types provide visualization of the characteristics of the lesion.

11. A system according to claim 10 wherein prior to detecting a reference vascular enhancement profile, the processor detects a pre-contrast baseline and the initial enhancement period comprising the steps of: (a) calculating the mean signal intensity of the entire volume of contrast at each time point to obtain a mean signal intensity time-curve S(t); (b) estimating a mean enhancement time-curve Y(t)=S(t)S(t.sub.0), where t.sub.0 represents the time point for a first pre-contrast scan; (c) finding a maximum uptake slope at time point t.sub.s and a maximum enhancement Y.sub.max=max{Y(t)}; (d) searching backward from t.sub.s to find the first occurring time point t.sub.1 such that t.sub.0t.sub.1<t.sub.s and Y(t.sub.1)<a %*Y.sub.max where a % is a fraction value that is adjustable; and (e) setting the initial enhancement period between t.sub.1 and t.sub.2=t.sub.s+(t.sub.st.sub.1) and find an initial enhancement index time point at t.sub.2.

12. A system according to claim 10 wherein the detection of the reference vascular enhancement profile includes the processor: (d) calculating a mean baseline image; (e) calculating an initial enhancement index map Y.sub.i, and a washout map Y.sub.O representing the decreasing profile portion; (f) finding all pixels collectively noted as XYZ such that Y.sub.i(XYZ)>f*max(Y.sub.i) and Y.sub.O(XYZ)>0, where f is an adjustable masking threshold.

13. A system according to claim 10 wherein for each pixel with an initial enhancement above a predetermined threshold, the step of correlation analysis includes a rank correlation with respect to the signature curves in order to obtain correlation coefficients and statistical p-values. The enhancement curve is classified into the signature curve type that corresponds to the maximum correlation coefficient value.

14. A system according to claim 10 wherein prior to the displaying of the grid-plot the processor applies a segmentation analysis for the lesion and provides information including the lesion volume and the percentage contribution from pixels with each type of enhancement curve respectively.

15. A system according to claim 10 wherein the enhancement curves of all pixels displayed in the grid-plot are highlighted by different color and line-thickness to reflect the type of enhancement curve and whether it is statistically significant respectively for visualization.

16. A system according to claim 10 wherein the pattern and heterogeneity of enhancement curve types in the lesion reflect the characteristics of the lesion including benign or malignant tumor.

17. A method according to claim 1 wherein prior to classifying an enhancement curve, the method further comprises the steps of selecting a enhancement curve from a pixel or a ROI, and setting it as one of the signature curves.

18. A system according to claim 10 wherein prior to the processor classifying an enhancement curve further comprises the steps of the processor selecting a enhancement curve from a pixel or a ROI, and setting it as one of the signature curves.

Description

BRIEF DESCRIPTION OF THE DRAWINGS

[0046] The invention will hereinafter be described in a preferred embodiment, by way of example only, with reference to the drawings wherein:

[0047] FIG. 1 illustrates five types of enhancement curves of signal intensity, as a function of time, in different breast tissues. Specifically curve I being no enhancement; curve II being slow sustained enhancement; curve III being rapid initial and sustained late enhancement; curve IV being rapid initial and stable late enhancement; and curve V being rapid initial and decreasing late enhancement;

[0048] FIG. 2 is a block diagram showing a communication network including a number of scanners linked to a data storage system and a processing system;

[0049] FIG. 3 is a plot of a typical averaged signal intensity time-curve from all pixels of a breast dynamic MRI data;

[0050] FIG. 4 is a series of plots showing a dotted line representing a typical vascular time-curve with rapid initial enhancement and fast washout, together with derived solid curves representing four types of signature enhancement curves as illustrated in FIG. 1;

[0051] FIG. 5(a) shows a graphical interface displaying an enhancement index map in MIP mode;

[0052] FIG. 5(b) shows a graphical interface displaying a color coded overlay of a pixel map of different types of enhancement curves on top of a raw image of a portion of the subject;

[0053] FIG. 5(c) is a gird-plot of curves for all pixels within the ROI of a displayed slice where different colors and line thicknesses are used to differentiate the enhancement curve type and whether it is statistically significant;

[0054] FIG. 5(d) is a plot of the mean enhancement curve of the ROI average signal intensity; and

[0055] FIG. 6 is a flow diagram showing steps performed by a computer program to enable the processing of image data and display of results according to an embodiment of the invention.

DETAILED DESCRIPTION OF PREFERRED EMBODIMENT

[0056] The present invention is particularly applicable to dynamic MRI, CT, NM and PET imaging systems, for example dynamic MRI of the breast. Raw data and/or images collected by a scan, such as from a MRI scanner 20, CT scanner 25, NM scanner 30 or PET scanner 35 are forwarded to a data storage system 40 in the form of a Picture Archiving Communications System (PACS) in FIG. 2. A computer program operating on a processor 50, in the form of a computer, is used to retrieve or receive the various images or raw data from any one of the scanners 20, 25, 30 or 35 or from the data storage system 40. The program then processes those images to provide an improved data set for medical personnel to use, particularly in relation to evaluate both morphologic and kinetic information for diagnosis of disease and optimal discrimination between benign and malignant disease. The computer program need not reside on computer 50, but may reside in a console computer linked to any one of the scanners 20, 25, 30 or 35. Alternatively the program may reside in a workstation (stand-alone or in a system) or in the PACS 40.

[0057] The user will first use the computer software to instruct the computer processor or workstation 50 to load the dynamic image data into a memory of the computer processor 50. The processor may apply image registration to the data if it is applicable by configuration to a particular organ, such as of the brain, or by user instruction. Image registration is where images are aligned to account for motion correction in situations where a ROI of the subject moves due to respiratory or other functions of the subject. The processor then detects the pre-contrast baseline and the rapid initial enhancement phase, and converts the signal intensity profile of the data to the contrast enhancement profile depending on the type of image data loaded. The processor further detects a subject-dependent reference time-curve and then derives signature curves of enhancement patterns for a classification system. Thereafter automated classification and segmentation analyses are applied to a region of interest of the subject, such as a suspicious lesion in the breast. The results are then displayed on the processor to the user.

[0058] The method can be configured to apply a motion correction method depending on the specific function and organ being studied. A typical dynamic imaging data consisting of a series of 3-dimensional (3D) scans are acquired over a period of time. During acquisition, at least one scan is acquired before contrast injection and at least two scans after the injection. A dynamic imaging data acquisition usually takes several tens of seconds to several tens of minutes depending on the specific protocol used, and patient motion during the acquisition period can occur. To accurately evaluate the contrast enhancement patterns of different tissue types, image alignment or registration of the imaged volume over time is desired, particularly for accurate diagnosis of small-size lesions. Different kinds of motion correction method can be applied depending on the specific organ being studied. For example, a conventional rigid body image registration method can be applied automatically for the brain. However, for some body organs such as the breast, kidney or liver, non-uniform distortions are normally expected due to local tissue stretching associated with cardiac and/or respiratory motion. Some nonlinear motion compensation methods recognizable by persons skilled in the art may be used such as a non-rigid registration well known in medical imaging and remote sensing fields.

[0059] The process of detecting the pre-contrast baseline and the rapid initial enhancement phase involves the following steps of:

[0060] (1) Calculating the mean signal intensity of the entire volume at each time point to obtain a mean signal intensity time-curve S(t), as shown by 76 in FIG. 3;

[0061] (2) Estimating a mean enhancement time-curve Y(t)=S=S(t)S(t.sub.0), where t.sub.0 represents the time point for the first pre-contrast scan;

[0062] (3) Finding the maximum uptake slope at time point t.sub.s and the maximum enhancement Y.sub.max=max{Y(t)};

[0063] (4) Searching backward from t.sub.s to find the first occurring time point t.sub.1 such that t.sub.0t.sub.1<t.sub.s and Y(t.sub.1)<a %*Y.sub.max where a % is used for example and the fraction value can be adjusted by configuration, such as 10%;

[0064] (5) Setting the initial enhancement period between t.sub.1 and t.sub.2=t.sub.s+(t.sub.st.sub.1) and find the initial enhancement index time point at t.sub.2>t.sub.s.

[0065] The process of further detecting a subject-dependent reference time-curve and derives signature enhancement patterns includes the following steps of:

[0066] (6) Calculating the mean baseline image S.sub.b(x, y, z) covering the time points from t.sub.0 to t.sub.1 for every pixel of (x, y, z) respectively;

[0067] (7) Calculating the initial enhancement index map as Y.sub.i(x, y, z)=S(x, y, z, t.sub.2)S.sub.b(x, y, z) and the washout map as Y.sub.O(x, y, z)=S(x, y, z, t.sub.2)S(x, y, z, t.sub.3), where t.sub.3 represents the last time point of the dynamic data. Alternatively by configuration, the relative enhancement maps Y.sub.i(x, y, z)=S(x, y, z, t.sub.2)/S.sub.b(x, y, z)1 and Y.sub.O (x, y, z)=S(x, y, z, t.sub.2)/S(x, y, z, t.sub.3)1 can be used for the same purpose;

[0068] (8) Finding all the pixels collectively noted as XYZ satisfying the criteria of Y.sub.i(XYZ)>f*max(Y.sub.i) and Y.sub.O(XYZ)>0, where f can be adjusted as a masking threshold. For example, a reference time-curve Y.sub.ref(t) can be calculated by averaging those pixels with f=80% to represent the mean vascular enhancement profile with a strong and rapid initial uptake followed by late washout, as shown by the thick dotted curve 80 in FIG. 4.

[0069] (9) Deriving four types of signature curves to represent the enhancement patterns as illustrated in FIG. 4 by:

[00001]$Y_4\left(t\right)=\{\begin{array}{cc}{Y}_{\mathrm{ref}}\left(t\right)& \left(t{t}_2\right)\\ Y_{\mathrm{ref}}\left(t_2\right)& \left(t>t_2\right)\end{array}.Math..Math.Y_5\left(t\right)=\{\begin{array}{cc}{Y}_{\mathrm{ref}}\left(t\right)& \left(t{t}_2\right)\\ Y_{\mathrm{ref}}\left(t_2\right)-\frac{1}{2}\left[Y_4\left(t_3\right)-Y_{\mathrm{ref}}\left(t_3\right)\right].Math.\left(t-t_2\right)/\left(t_3-t_2\right)& \left(t>t_2\right)\end{array}.Math..Math.Y_3\left(t\right)=\{\begin{array}{cc}{Y}_{\mathrm{ref}}\left(t\right)& \left(t{t}_2\right)\\ Y_{\mathrm{ref}}\left(t_2\right)+\frac{1}{2}\left[Y_4\left(t_3\right)-Y_{\mathrm{ref}}\left(t_3\right)\right].Math.\left(t-t_2\right)/\left(t_3-t_2\right)& \left(t>t_2\right)\end{array}.Math..Math.Y_2\left(t\right)=\{\begin{array}{cc}0& \left(t{t}_1\right)\\ Y_{\mathrm{ref}}\left(t_3\right).Math.\left(t-t_1\right)/\left(t_3-t_1\right)& \left(t>t_1\right)\end{array}$

Where Y.sub.2(t) represents type II curve 82 with slow sustained enhancement (Slow), Y.sub.3(t) represents type III curve 84 with rapid initial and sustained late enhancement (Sustained/Persistent), Y.sub.4(t) represents type IV curve 86 with rapid initial and stable late enhancement (Plateau) and Y.sub.5(t) represents type V curve 88 with rapid initial and decreasing late enhancement (Washout).

[0070] The initial enhancement index time point (t.sub.2) may be fixed by configuration for a specific data acquisition protocol. Further, the user may have the option to select a preferred time-curve from a pixel or a ROI, and set it as one of the signature curves described above.

[0071] The automated classification is applied to all pixels in a region of interest of the subject, such as a suspicious lesion in the breast. Pixels whose initial enhancement are below certain threshold (for example, Y.sub.i(XYZ)<10%*max(Y.sub.i) where the threshold percentage value can be adjusted by configuration) are classified as no enhancement type I and not included for further classification analysis. The time-curve of each candidate pixel is subject to a Spearman's rank correlation with each of the four signature curves (Y.sub.2, Y.sub.3, Y.sub.4 and Y.sub.5) to obtain correlation coefficients (cc) and statistical probability (p-value) respectively. The cc value ranges from 1 to 1, representing complete negative and positive linear relationships between two variables while a cc value of 0 means no linear relationship between the two variables. The smaller the p-value, the more significant the result is said to be. If the p-value is less than the significance level, the result is terms as statistically significant. The default level of significance (p value) is 5%, and the user can select other popular levels (1% or 0.1%) by configuration For each pixel, a determination is made of the unique curve type corresponding to the maximum cc value. In the situation where multiple curve types produce the same maximum cc value, a further Pearson's correlation is applied to further determine the type also corresponding to the maximum Pearson's cc value. The classified curve type and corresponding p-value are recorded for each candidate pixel. The results are then displayed on the processor to the user.

[0072] The process of displaying the results involves an intuitive interface displaying the enhancement index map in MIP mode as shown in FIG. 5(a). The user can place a ROI 92 over a suspicious lesion on the MIP image. The processor automatically finds the slice containing the maximum amount of signal enhancement within the ROI, then display the detected slice. The processor further automatically detects a cluster of connected pixels belonging to one of the four signature types, and then calculates the volume of the cluster and the percentage contribution of each type with the results displayed in a tabular form 94 in the image window. The results of the segmentation and pattern classification of all pixels of the detected cluster can be stored in a color coded curve pattern map for display as a color overlay 98 on top of the raw image as shown in FIG. 5(b) for easy visualization of both the enhancement and morphology of the enhancing lesions. In addition, a grid-plot 95 of the curves for all individual pixels within the ROI of the displayed slice can be displayed simultaneously as shown in FIG. 5(c), wherein different colors and line-thickness can be used to highlight the type of enhancement curve and whether it is statistically significant for enhanced visual contrast effects. For example, each type of the four enhancing curves can be represented by different colors (e.g., red for type V, yellow for type IV, cyan for type III and blue for type II) while the line thickness differentiates whether the classification is significant (e.g., extra thick line for p<0.05 as being statistically significant versus thick line for non-significance), and the non-enhancing type I curve is marked by a white thin line. Alternatively, the user can display the mean curve-plot 96 of the ROI average signal intensity as shown in FIG. 5(d), which does not provide the enhancement heterogeneity information within as clearly shown by the grid-plot 95 in FIG. 5(c).

[0073] The user may have the option to apply the cluster analysis to each suspicious lesion highlighted by a ROI one by one, or to all lesions located within the boundary of a selected ROI.

[0074] With reference to FIG. 6 there is shown a flow diagram 100 of the steps taken by a computer program loaded into memory of a computer processor. Computer program code is written to perform the various steps and functions depicted in the flow diagram 100 whereby upon user instruction, the computer program loads the dataset into the memory of the computer processor at step 102. At step 104, if it is applicable to a particular scan, such as of the brain or breast, or by user instruction, image registration can be applied by the computer program. At step 106 the program detects the baseline and initial uptake/enhancement periods. At step 108 the mean baseline image, initial enhancement index map and washout map are calculated. At step 110 the computer program detects a reference curve from the mean vascular enhancement profile, and then derives the signature curves for a classification system. At step 111, the user may have the option to manually select a preferred time-curve from a pixel or a ROI to replace one of the signature curves detected. At step 112 a correlation analysis is applied to classify every pixel with initial enhancement above a certain threshold. At step 114, the program performs segmentation analysis of a suspicious lesion identified by a user-defined ROI and at 116 the results including a grid-plot and the color-coded curve pattern are then displayed of the whole process to the user for inspection and review.

[0075] This embodiment has been described using an example of a dynamic MRI of the breast. The invention is equally applicable to other disease such as prostate cancer, brain tumor or diseases in other body organs, not just of humans but animals as well, and using CT or NM or PET scans. Further the invention can be extended by combining the detailed kinetic and morphologic information in order to provide optimal discrimination between benign and malignant disease.

[0076] It will be appreciated by persons skilled in the art that numerous variations and/or modifications may be made to the invention as shown in the specific embodiments without departing from the scope of the invention as broadly described. The present embodiments are, therefore, to be considered in all respects as illustrative and not restrictive.