WO2009023626A1

WO2009023626A1 - Systems and methods for tissue characterization and border detection

Info

Publication number: WO2009023626A1
Application number: PCT/US2008/072774
Authority: WO
Inventors: Amin Katouzian; Babak Baseri; Elisa E. Konofagou; Andrew F. Laine
Original assignee: The Trustees Of Columbia University In The City Of New York
Priority date: 2007-08-10
Filing date: 2008-08-11
Publication date: 2009-02-19

Abstract

Systems and methods for tissue characterization and border detection are disclosed herein. An exemplary method of the disclosed subject matter for tissue characterization and border detection includes receiving image or radiofrequency data from a predetermined region of interest, extracting features from the data, and categorizing the features, thereby determining tissue characterization and tissue border location. In some embodiments, receiving image or radiofrequency data further includes positioning a transducer near said predetermined region of interest and utilizing said transducer to collect data regarding said region of interest. In some embodiments, utilizing of the transducer to collect data regarding the region of interest includes utilizing the transducer to perform ultrasound.

Description

SYSTEMS AND METHODS FOR TISSUE CHARACTERIZATION AND

BORDER DETECTION

CROSS REFERENCE TO RELATED APPLICATIONS This application claims priority to U.S. Provisional Application No.

60/955,249 entitled "Intravascular ultrasound (IVUS) Tissue Characterization and Border Detection Through Backscattered Signals by Wavelet Packet Analysis", filed on August 10, 2007 and U.S. Provisional Application No. 61/081,669 entitled "Texture Derived Atherosclerotic Tissue Characterization and Border Detection", filed on July 17, 2008, which are incorporated by reference in their entirety herein.

BACKGROUND

FIELD

The present application relates to systems and methods for tissue characterization and border detection.

BACKGROUND ART

The atherosclerotic tissue characterization and detection of vulnerable plaques in coronary arteries have been of extensive interest in the field of interventional cardiology. Intravascular ultrasound (IVUS) provides morphologic as well as pathologic cross- sectional grayscale images of the arterial lumen and occluded lesions throughout the arteries. The IVUS radiofrequency (RF) signals provide adequate spatial resolution and sufficient penetration while other comparable imaging techniques, such as near infrared (NIR), and optical coherence tomography (OCT) with excellent resolution, lack adequate penetration or clinical applicability.

Previous efforts at tissue characterization have deployed IVUS elastography (IVE) using cross-correlation between the RF signals in consecutive frames to assess the local mechanical properties of the plaques in vitro and classified tissues based on their mean strain values to fibrous, fibro-fatty and fatty. Additional IVE validation that employed the LaGrangian speckle model estimator (LSME) along with the scaling factor estimator (SFE) to compute the radial strain elastograms has been performed. Integrated backscatter (IB) has been combined with IVUS (IB- IVUS) to classify tissues and delineate lipid, fibrous and calcified tissues in vivo. Contrary to IVE and IB-IVUS that could not detect the necrotic core directly previous researchers presented a spectrum-based technique in combination with a classification tree to classify tissues into fibrous, fibro-lipid, necrotic and calcified. They extracted eight spectral features (intercept, slope, mid-band-fit (MBF), IB, minimum and maximum powers and their corresponding frequencies) from a linear regression fit to the calibrated tissue spectra using 30 MHz, single element rotating transducers. The first two spectral signatures (intercept, slope) are indicative of scatterer concentration and size. Currently, this methodology, known as IVUS-virtual histology (IVUS-VH), has been implemented in the Volcano (Rancho Cordova, CA) IVUS clinical scanners using 20 MHz 64 element phased array transducers.

There is a need for an effective texture-derived tissue characterization algorithm using a discrete wavelet packet frame (DWPF) and two-dimensional (2D) envelope detection.

SUMMARY

Systems and methods for tissue characterization and border detection are disclosed herein.

An exemplary method of the disclosed subject matter for tissue characterization and border detection includes receiving image or radiofrequency data from a predetermined region of interest, extracting features from the data, and categorizing the features, thereby determining tissue characterization and tissue border location. In some embodiments, receiving image or radiofrequency data further includes positioning a transducer near said predetermined region of interest and utilizing said transducer to collect data regarding said region of interest. In some embodiments, utilizing of the transducer to collect data regarding the region of interest includes utilizing the transducer to perform ultrasound.

In some embodiments, extracting features from the data further includes processing the data, applying a tensor product extension to the processed data, computing one or more output signals, and constructing one or more feature representation matrices. In the same or another embodiment, the processing of the data includes decimating the data in an axial direction and interpolating the data in a lateral direction. In some embodiments, categorizing of the features includes clustering the features into clusters, computing the center of the clusters, assigning pixels of feature representation matrices to the clusters, re-computing the centers of the clusters and repeating the assigning pixels of the feature representation matrices to the clusters and the re-computation of the cluster centers one or more times. In some embodiments, clustering the features into one or more clusters comprises assigning labels to the pixels in the feature representation matrices.

In some embodiments, assigning one or more labels to one or more pixel in said one or more feature representation matrices includes utilizing a K-mean classifier to assign labels to each pixel in the feature representation matrices. In some embodiments, assigning of the pixels of the feature representation matrices to the clusters includes determining the shortest distance between each pixel and the center of each cluster, and assigning the pixels to the nearest cluster. In some embodiments, the computation of the center of the clusters and the assignment of the pixels to the clusters is repeated until there is no further change in the assignment of the pixels to clusters.

An exemplary system of the disclosed subject matter for intravascular tissue characterization and border detection in an artery or vein includes a processor and a memory operatively coupled to the processor, the memory storing program instructions that when executed by the processor, cause the processor to extract features from the data, and categorize the features, thereby determining tissue characterization and tissue border location. In some embodiments, an exemplary system further comprises a transducer operatively coupled to the processor, and the memory storing program instructions, when executed by the processor, further cause the processor to utilize the transducer to collect data regarding regions of interest. In some embodiments, an exemplary system further includes a catheter for transporting the transducer to the regions of interest and a catheter guide wire for guiding a catheter to regions of interest. In some embodiments, an exemplary system further includes a tissue box for placing an organ for measurement, at least one hemostatis valve, at least one pressure monitor, at least one pressure regulator, a fluid reservoir, and a pump for circulating fluid from the fluid reservoir though an organ. In some embodiments, the fluid is phosphate buffered saline. In some embodiments, the fluid is blood. BRIEF DESCRIPTION OF THE DRAWINGS

The accompanying drawings, which are incorporated and constitute part of this disclosure, illustrate some embodiments of the disclosed subject matter.

FIG. l(a) is a diagram of a system for in vivo data collection, tissue characterization and border detection in accordance with an embodiment of the disclosed subject matter.

FIGS. l(b)-(d) illustrate a catheter located in three different areas of a human heart in accordance with an embodiment of the disclosed subject matter.

FIG. 2 is a diagram of a system for in vitro data collection, tissue characterization and border detection in accordance with an embodiment of the disclosed subject matter.

FIG. 3 illustrates an exemplary method for catheter insertion and ultrasound measurement of an artery or vein in accordance with an embodiment of the disclosed subject matter. FIG. 4 illustrates an exemplary method for multi-channel wavelet analysis in accordance with an embodiment of the disclosed subject matter.

FIG. 5 is a tree-structure diagram for discrete wavelet packet frames in accordance with an embodiment of the disclosed subject matter.

FIG. 6(a) is a graph of a Lemarie-Battle filter of order 18 in accordance with an embodiment of the disclosed subject matter.

FIG. 6(b) is a graph of a constructed filter bank at level 4 in accordance with an embodiment of the disclosed subject matter.

FIG. 7(a) is an intravascular ultrasound image in the (r,θ) domain in accordance with an embodiment of the disclosed subject matter. FIG. 7(b) is an intravascular ultrasound image in the Cartesian domain in accordance with an embodiment of the disclosed subject matter.

FIG. 8(a) is an intravascular ultrasound B-mode image with manually traced vessel wall and lumen borders in accordance with an embodiment of the disclosed subject matter. FIG. 8(b) is a classified tissue color map defined as prognosis histology (PH) image generated in accordance with an embodiment of the disclosed subject matter. FIG. 8(c) is a Movat Pentachrome histology image in accordance with an embodiment of the disclosed subject matter.

FIGS. 9(a)-(b) is a intravascular ultrasound images showing border detection in accordance with an embodiment of the disclosed subject matter.

DETAILED DESCRIPTION

The systems and methods described herein are useful for tissue characterization and border detection. Although the description is focused on the example of atherosclerotic tissue analysis, the systems and methods herein are useful for tissue characterization in other body locations, such as the liver, prostate, or breast. An exemplary embodiment of the system 10 is illustrated in Figure l(a), and includes signal or image acquisition equipment 20. For example, any known intravascular ultrasound equipment having a transducer 25, e.g., a 40 MHz mechanically rotating single element Boston Scientific Atlantis transducer or a 45 MHz mechanically rotating single element Volcano Revolution transducer, may be used for acquiring the images of the intravascular plaque structure of a patient P, whether in vivo or in vitro (not shown). Image acquisition equipment may include video/signal capture equipment 30, e.g., two 12-bit Acquiris circuit boards, and data storage equipment 40, e.g., a hard drive or other storage medium, to store the resulting video images/signals. The video images may be written onto a tape, memory card, or other medium by an appropriate recording device 45. Image processing equipment 50 is used to process the images in accordance with the invention. Image or RF processing may be performed by a personal computer 55, such as a IBM with 32 bit Intel® Core ™ 2 CPU, 6600 @ 2.4GHz and 6 GB RAM memory, or other computer, having a central processing unit or processor 57 and memory 59 storing program instructions for execution by the processor 57, an input device 60, such as tape drive, memory card slot, etc., for receiving the digital images and a keyboard 70 for receiving user inputs, and an output device, such a monitor 75, a printer 80, or a recording device 90 for writing the output onto a tape, memory card, or other medium. Image processing equipment 50 may also located on several computers, which are operating in a single location or which are connected as a remote network. Figures l(b)-(d) show three possible positions that a transducer bearing catheter 102 can be placed in a human heart, e.g., coronary arteries, whether in vitro or in vivo. The transducer can be inserted into the heart by well known means, for example, insertion through a femoral artery using a guide wire, e.g., a 0.014 inch guide wire, and a catheter 102, e.g. , a 40 MHz rotating single-element Boston Scientific transducer bearing catheter. Figure l(b) shows a configuration wherein the catheter 102 has been inserted into the left coronary artery. Figure l(c) shows a configuration wherein the catheter 102 has been fed into the right coronary artery. Figure l(d) shows a configuration wherein the catheter 102 has been feed into the circumflex branch of the left coronary artery. The transducer bearing catheter 102 can also be fed into any one of the numerous smaller artery branches.

In one exemplary embodiment, measurements of the artery can be made utilizing a transducer to produce sounds waves and detect the reflections of those waves. This technique is known as intravascular ultrasound (IVUS). The transducer can be, for example, a 40 MHz rotating single-element Boston Scientific transducer bearing catheter. The measurements can be made utilizing an automatic or manual pullback technique, for example pulling back the transducer from distal to proximal. In one exemplary embodiment the catheter pullback speed and frame rate can be set to 0.5 mm/second and 30 frames/second, respectively. Each frame can consist of 256 lines and 2048 samples per line, for example.

Figure 2 shows an exemplary system for in vitro imaging and analysis in accordance with the disclosed subject matter. The in vitro system of Figure 2 can be used in connection with the computer equipment represented in Figure l(a). In an exemplary embodiment, an organ or specimen, e.g., an artery (not shown), can be placed in the tissue fixture box 210, wherein catheter guide wire 224 can be fed into the specimen through proximal hemostatis valve 226. The catheter guide wire 224 can be used to guide the catheter, and accompanying transducer (not shown), into the organ's interior to a predetermined location for conducting measurements, for example, IVUS. Figure 2 further shows a distal hemostatis valve 212 on the opposite side of the tissue box fixture 210. Inflow pressure monitor 222 monitors the inflow 228, and outflow pressure monitor 214 monitors the outflow 230, and pressure regulator clamp 216 allows the fluid pressure in the organ to be adjusted as necessary. Pump 218 pumps the fluid from the fluid reservoir 220 through the organ. In an exemplary in vitro embodiment, the fluid in the fluid reservoir 220 can be phosphate buffered saline (PBS) and it can be circulated through the organ at a constant pressure, for example, 100 mmHg. In an alternative embodiment, blood, e.g., human blood, can be used instead in the in vitro system shown in Figure 2. Figure 3 illustrates an exemplary method for in vivo data acquisition in accordance with the disclosed subject matter. In an exemplary embodiment, a catheter can be inserted into the subject's cardiovascular system, for example from a femoral vessel near the groin. Alternatively, the catheter can be inserted into carotid or jugular vessels at the bend of the elbow and the neck, respectively. To insert the catheter, an incision is made 312 at the point of entry, and a guide wire is inserted 314 into the artery or vein. The guide wire can be, in an exemplary embodiment, a 0.014 inch flexible guide wire. The guide wire can be threaded through the artery or vein until it reaches a portion of the heart of interest. The catheter can then be threaded along the guide wire 316 until it has reached the region of interest. In an exemplary embodiment, the catheter can be equipped with a small ultrasound probe attached to the distal end of the catheter. The catheter can be the type of catheter typically used for IVUS, for example, a 40 MHz mechanically rotating single element Boston Scientific Atlantis transducer or a 45 MHz mechanically rotating single element Volcano Revolution transducer. The proximal end of the catheter can be connected to a computerized system for collecting and digitizing the radiofrequency (RF) data, for example using two 12-bit Acquiris circuit boards. Once the distal end of the catheter is near the coronary arteries it can be used to perform an angiogram in the well-known manner, e.g., by x-ray imaging, thus producing an image of the heart that can be used to determine regions of interest 322 for further imaging and analysis in accordance with the disclosed subject matter.

Once an region of interest is determined the ultrasound probe can be utilized to conduct IVUS of the region of interest. The method for conducting the measurements can be, in one exemplary embodiment, the technique of pushing the transducer beyond the region of interest 324 and then performing automated or manual pullback 326 of the probe at a predetermined rate, e.g., 0.5 mm/second, taking measurements 326 at a predetermined rate, e.g., 30 frames/second. In an exemplary embodiment, each frame can consist of 256 lines and 2048 samples per line. The radiofrequency (RF) data can be collected using a computerized system, e.g., two 12- bit Acquiris circuit boards, and digitized using the same at periodic time intervals, e.g., every 2.5 nanoseconds.

Figure 4 illustrates an exemplary method, wherein after the data has been collected, a multi-channel wavelet analysis 400 can be performed on the data to extract useful information. The spatial-frequency-localized expansions and their generalization to two dimensions can be used to discern the patterns on constructed images from backscattered IVUS signals while the geometrically oriented decompositions are provided at this dimension, e.g., two or three. Unlike with discrete wavelet transform (DWT) and discrete wavelet packet transform (DWPT), decompositions can be translation invariant in the discrete wavelet frame (DWF) or discrete wavelet packet frame (DWPF) since no decimation occurs between expansion levels. The DWPF can be a benefit to signal representations such as textures in at least two respects: 1) it can result in less restriction on filter selection; and 2) it can prevent aliasing. The principles underlying the multi-channel wavelet analysis will now be explained in detail. Wavelet packets are orthonormal in space of summable- integrable functions L² (R) and can be described by a collection of functions ξ_j

0,p ≠ q} obtained from:

2^lξ_2k (2^ιx-n)= ∑hl₂βξ_k (2^Mx-m) (l) meZ L LU

2H_u+λ (2^ιx-n) = ∑ g_m ^> __2n2 * ξ_k {2^Mx-_m) (2) meZ where / , n , k , ξ_ϋ (x) = φ(x) and ξ_λ {x) = ψ (x) are the scale index, translation index, channel index, scaling function and basic wavelet, respectively. The selection of discrete filters h_n and g_n are described in more details below. The wavelet packets at different scales can also be found by the inverse relationship as follows:

2 (2'x-n) (3)

Any function f(x) <≡ H (R) can be decomposed onto a wavelet packet basis by the inner product \f(x) ,ξ_k \ 2^ι x-n vj . Equation (3) can be used to write:

=∑i,2^]_/(^_a(^-f₊χi₂;^|_/(^_M(^-«)ώ (4)

Defining the decomposition coefficients as:

equation (4) can be rewritten as:

PK¹ m = Σ Hn-InAk* + Σ gm' -2n pL₊l,n (⁶) n n

Using equations (1) and (2), the coefficients can be calculated by:

m

In the example of a standard wavelet transform, the index k is restricted to k = 0 and only two wavelet packets ξ_ϋ and ξ_] are used. Consequently, only the leftmost nodes (pΛ' are decomposed into high and low frequency bands.

However, in the example of wavelet packets, the decompositions can be performed on both low and high frequency components. Figure 5 illustrates the result of an exemplary application of decompositions performed on both low and high frequency components for the case of wavelet packets. As shown in Figure 5, a tree- structure multiband extension of the standard wavelet transform can be constructed utilizing both low and high frequency components.

A Fourier transform can be used to represent this tree-structure in the frequency domain and the tree-structure can be seen as sub-band filtering and implemented using iterated constructed highpass and lowpass filters in frequency domain. Taking the Fourier transform of both sides of equations (7) and (8) yields:

Y- H = G' {ω)Υ[ {ω) (9)

Υ[t_x {ω) = H' {ω)Υ[ {ω) (lθ)

where Y^ (o) is the Fourier transform of the frame coefficients at channel k and level / . Since the IVUS signals can be sampled at the rate of f_s , the original discrete signal can be considered as the set of frame coefficients at the first scale (/ = θ) .

Returning to Figure 4, the multi-channel wavelet analysis can be performed by first selecting 410 a set of high and low pass filters. In one exemplary embodiment, the highpass G' (ω) and lowpass H' (ω) filters at each level / can be realized by: G' (ω) = G⁰ (2' ω) and H¹ {ω) = H° (l^!ω) . Consequently, the multichannel wavelet schematic in Figure 5 can behave like a filter bank with channel filters JF/ {ω)\ 0 < k < 2' -l| , where F/ (ω) can be derived recursively as follows:

F- Η = G'⁺¹ Η Fl Η = G° (2^M ω) F/ (ω) (12)

(ω) (13)

The selection of the filters G° (<y) and H⁰ (ø) can have an impact on texture classification performance. In an exemplary embodiment, the filter candidates should satisfy necessary criteria such as symmetry as well as boundary accuracy and have optimal frequency response. In the same or another embodiment, the filters can be Lemarie-Battle wavelets that are symmetric (e.g., have linear phase response) and quadrature mirror filters (QMF), as shown in Figure 6(a). The Lemarie-Battle wavelet property alleviates boundary effects through simple methods of mirror extension. The discrete highpass filter g° can be obtained by g° = (-1)" fy⁰. or G⁰ Η = H⁰ (ω + π) in the frequency domain. Figure 6(b) illustrates the constructed filter bank at level 4 generated by a Lemarie-Battle wavelet of order 18, shown in Figure 6(a). The wavelets using QMF as well as constructed filter bank |F/ (fi>)| can cover the frequency domain and satisfy the property:

The procedure for extracting features 420 will now be described in detail. In an exemplary embodiment, the IVUS signals from each raw data frame, represented in the {r,θ) domain, containing 256 lines that span over 360 degrees with 2048 samples per line can be processed 422. In order to have an optimized frame size with respect to its computational complexity and textural resolution, the signals can be processed 422 by decimating and spline interpolating the signals in the axial and lateral directions, respectively, to generate the square M = 512 pixels frame. Figure 7(a) illustrates the B-mode (brightness-mode) images of an IVUS frame in the (r, θ) domain and Figure 7(b) illustrates the same frame in the (x,y)

Cartesian domain.

In one exemplary embodiment, for each frame, a tensor product extension can be applied 424, in which the channel filters were denoted by

F,'_XJ {ω_r , ω_θ ) = F₁' (ω_r ) Fj (ω_θ ) . Consequently, such an extension will lead to the orientation selectivity in the decomposition tree. Four possible orientations can be considered excluding the root node, which is omnidirectional. The node last filtered by G' (co_r)H^l ((O₆) corresponds to the vertical orientation. The highpass filter G¹ and lowpass filter H' can be applied in the axial and lateral directions, respectively. The node last filtered by H¹ (o)_r)G' (co_θ) corresponds to the horizontal orientation. The lowpass filter H' and highpass filter G¹ can be applied in the axial and lateral directions, respectively. The node last filtered by G' (ω_r)G^! (ω_θ) corresponds to the diagonal orientation. The highpass filter G' and lowpass filter H¹ can be applied in the axial and lateral directions, respectively. The node last filtered by

H' {co_r)H^l (ω_θ) has the same orientation as its parent. The lowpass filter H¹ and lowpass filter H¹ can be applied in the axial and lateral directions, respectively.

Due to narrowband characteristic of IVUS signals, the envelope of output signals from channel filters can be computed 426 using the corresponding two dimensional analytical signals. Finally, the feature matrices can be constructed 428 as follows: ^ = {<# |θ ≤ * ≤ (2' -l), UJ = \,...,M] (16)

where e^l'_} ^k represents the envelope value of pixel (i,j) for the k -th channel at level

/ .

The process of tissue classification and border detection 430 will now be described in detail. The overall justification of in vivo real-time plaque characterization is made by the interventional cardiologists through the use of classified tissues. Traditionally, researchers build the training data sets by marking regions of interest (ROIs) in the arteries and taking relative cross-sectional histology images to label the plaque compositions. Once the training data set is developed, a supervised classifier can be used to differentiate the tissue types. The overall performance of such a classifier can be evaluated through cross-validation.

In an exemplary embodiment, for every feature representation matrix, V_MxM , the pixels can be clustered 432 by assigning 434 a label to each pixel in feature representation matrix, V_UxM , by modulo N_c , where N_c represented the number of classes of tissue, e.g., four classes of tissue for identifying plaque (fibrotic, lipidic, calcified and no-tissue) and two classes for identifying a lumen border (the boundary between blood and non-blood). In one embodiment, a K-mean classifier can be used to assign 434 the labels for each feature representation matrix V_MyM . The center of clusters ic_κ

≤ /c ≤ N_c l] can be computed 436 by calculating the mean vector for each class. In the same or another embodiment, the pixel Iv₁ ^ /, j = 1,..., M ) can be assigned 438 to a cluster, e.g., the class K , if the Euclidean distance between the corresponding pixel and the cluster (class) center C_κ is a predetermined distance. In an exemplary embodiment, if the distance between a given pixel and a given class center is the shortest distance among the distances between that pixel and all the class centers then that pixel is assigned 438 to that class cluster. The centers of the clusters can be updated in an iterative fashion by re-computing 440 the relative mean vectors.

The procedure can be terminated 442 once no further change in assigning 438 occurs. In an exemplary embodiment, the above described algorithm can be an unsupervised algorithm, e.g., an algorithm that does not need to be taught which feature belongs to which class. In the same or another embodiment, a K-mean classifier can be used to partition the class clusters in an unsupervised fashion. The above described algorithm can assign features to classes based solely on the feature's

Euclidean distance from the center of the class cluster.

In an exemplary embodiment, the tissue characterization was performed using the above algorithm on 512-by-512 frames with Lemarie-Battle filter of order 18, at expansion level, e.g., 1 = 2 and number of classes, e.g., tissue types,

N_c = 4. Figure 8(a) shows an IVUS B-mode image with vessel wall 810 and the lumen border 812 manually traced onto the image. Figure 8(b) shows an IVUS image produced using the above algorithm, illustrating that the above algorithm is capable of discerning plaque constitutes. Figure 8(c) shows the corresponding Movat Pentachrome histology image. Figure 9(a) shows an example where an expansion level of / = 5 and class number of N_c = 2 can be used to detect the lumen border 910 automatically, represented by green line 911 compared to manually drawn red line 912. Figure 9(b) shows an example where an expansion level of / = 5 and class numbers of N_c = 2 , N_c = 3 , or N_c = 4 can be used to estimate the location of the vessel wall 914, represented by automatically produced green line 915 and manually drawn yellow line 916.

It will be understood that the foregoing is only illustrative of the principles described herein, and that various modifications can be made by those skilled in the art without departing from the scope and spirit of the disclosed subject matter. For example, the system and methods described herein are used for tissue characterization and border detection . It is understood that that techniques described herein are useful in connection with tissue characterization and border detection .

Moreover, features of embodiments described herein may be combined and/or rearranged to create new embodiments.

Claims

CLAIMS We claim:

1. A method for tissue characterization and border detection, comprising:

(a) receiving image or radiofrequency data from a predetermined region of interest;

(b) extracting features from said data; and

(c) categorizing said features, thereby determining tissue characterization and tissue border location.

2. The method of claim 1 , wherein said receiving image or radiofrequency data further comprises:

(a) positioning a transducer near said predetermined region of interest; and

(b) utilizing said transducer to collect data regarding said region of interest.

3. The method of claim 2, wherein said utilizing of said transducer to collect data regarding said region of interest comprises utilizing said transducer to perform ultrasound.

4. The method of claim 1, wherein said extracting features from said data further comprises:

(a) processing said data;

(b) applying a tensor product extension to said processed data;

(c) computing one or more output signals; and

(d) constructing one or more feature representation matrices.

5. The method of claim 4, wherein said processing of said data comprises decimating said data in an axial direction and interpolating said data in a lateral direction.

6. The method of claim 1 , wherein said categorizing of said features comprises: (a) clustering said features into one or more clusters;

(b) computing the center of said one or more clusters;

(c) assigning one or more pixels of one or more feature representation matrices to said one or more clusters; and

(d) re-computing said center of said one or more clusters;

(e) repeating (c) and (d) one or more times.

7. The method of claim 6, wherein said clustering said features into one or more clusters comprises assigning one or more labels to one or more pixels in one or more feature representation matrices.

8. The method of claim 7, wherein said assigning one or more labels to one or more pixel in said one or more feature representation matrices comprises utilizing a K-mean classifier to assign one or more labels to each pixel in said one or more feature representation matrices.

9. The method of claim 6, wherein said assigning one or more pixels of one or more feature representation matrices to said one or more clusters comprises:

(a) determining the shortest distance between said one or more pixels and said center of said one or more clusters; and

(b) assigning said one or more pixels to the nearest of said centers of said one or more clusters.

10. The method of claim 6, wherein said repeating of (c) and (d) is performed until there is no further change in the assignment of said one or more pixels to said one or more clusters.

11. A system for tissue characterization and border detection comprising:

a processor,

a memory operatively coupled to said processor, said memory storing program instructions that when executed by said processor, cause said processor to: extract features from said data;

categorize said features, thereby determining tissue characterization and tissue border location; and

output tissue characterization data to a computer readable medium.

12. The system of claim 11 , further comprising:

a transducer operatively coupled to said processor, and

wherein said memory storing program instructions, when executed by said processor, further cause said processor to utilize said transducer to collect data regarding one or more regions of interest.

13. The system of claim 12, further comprising:

a catheter for transporting said transducer to said one or more regions of interest, and

a catheter guide wire for guiding said catheter to said one or more regions of interest.

14. The system of claim 13, further comprising:

a tissue box for placing an organ for measurement,

one or more hemostatis valves,

one or more pressure monitors,

one or more pressure regulators,

a fluid reservoir, and

a pump for circulating fluid from said fluid reservoir though said organ.

15. The system of claim 14, wherein said fluid is phosphate buffered saline.

16. The system of claim 14, wherein said fluid is blood.