Tissue characterization using quantitative biomarkers based on the orientation of blood vessels

Abstract

Vessel orientation feature data are generated from blood vessel image data acquired from a subject. The vessel orientation feature data indicate penetrating and circumferential blood vessels in a region-of-interest (ROI) based on quantifying orientations of blood vessels in the ROI. A blood vessel image is provided to a computer system. An ROI image is generated by applying an ROI mask to the blood vessel image, where the ROI mask corresponds to the ROI in the blood vessel image. The ROI image is converted to a binary image, where the binary image contains pixels with first binary values and pixels with second binary values, where pixels with the first binary value correspond to segmented vessels. The vessel orientation feature data are generated from the binary image.

Description

Mayo 2024-087 630666.01598 TISSUE CHARACTERIZATION USING QUANTITATIVE BIOMARKERS BASED ON THE ORIENTATION OF BLOOD VESSELS CROSS-REFERENCE TO RELATED APPLICATIONS

[0001] This application claims the benefit of U.S. Provisional Patent Application Serial No.63 / 668,666, filed on July 8, 2024, and entitled “TISSUE CHARACTERIZATION USING QUANTITATIVE BIOMARKERS BASED ON THE ORIENTATION OF BLOOD VESSELS,” which is herein incorporated by reference in its entirety. STATEMENT OF FEDERALLY SPONSORED RESEARCH

[0002] This invention was made with government support under CA239548 awarded by the National Institutes of Health. The government has certain rights in the invention. BACKGROUND

[0003] Analysis of blood vessels or microvessels has played a crucial role in diagnosing various diseases, like eye diseases (glaucoma, diabetic retinopathy, macular edema), and tumor classification from different modalities (magnetic resonance imaging (MRI), computed tomography (CT), optical coherence tomography (OCT), and ultrasound) for different tissue regions such as breast, lymph node, thyroid, liver, eye, skin, etc. The most common way to quantify the changes in the behavior of blood vessels is through morphological features such as tortuosity, area, bifurcation angle, etc. SUMMARY OF THE DISCLOSURE

[0004] It is an aspect of the present disclosure to provide a method for generating vessel orientation feature data from blood vessel image data acquired from a subject. The method includes providing a blood vessel image to a computer system, the blood vessel image having been acquired with an imaging system from a subject. A region-of-interest (ROI) image is generated from the blood vessel image using the computer system by applying an ROI mask to the blood vessel image, where the ROI mask corresponds to an ROI in the blood vessel image. The ROI image is converted to a binary image using the computer system, where the binary image contains pixels with first binary values and pixels with second binary values, where pixels with the first binary value correspond to segmented vessels. Vessel orientation feature data are generated from the binary image using the computer system, where the vessel 1 QB\630666.01598\97090978.4Mayo 2024-087 630666.01598 orientation feature data indicate penetrating and circumferential blood vessels in the ROI based on quantifying orientations of blood vessels in the ROI. The vessel orientation feature data are output using the computer system. Other embodiments of this aspect include corresponding systems (e.g., computer systems), programs, algorithms, and / or modules, each configured to perform the steps of the methods. BRIEF DESCRIPTION OF THE DRAWINGS

[0005] FIG. 1 illustrates a Cartesian coordinate to polar coordinate conversion of a binary image.

[0006] FIG.2 illustrates histogram features with (a) a flat distribution, (b) an irregular and dense distribution, and (c) a peaky distribution.

[0007] FIG. 3 illustrates histogram features and an illustration of features with only penetration vessels (top row), only circumferential vessels (middle row), and both penetrating and circumferential vessels (bottom row).

[0008] FIG. 4 illustrates calculating a polar gradient angle of (a) circumferential and (b) penetrating structures in polar coordinates.

[0009] FIG.5 illustrates calculating a vessel angle as part of calculating an angle-based penetration density (APD) feature.

[0010] FIG. 6 illustrates calculating a penetration to circumferential density (PCD) feature, in which (a) illustrates a central portion of an ROI and (b) illustrates computation of a penetration distance measurement.

[0011] FIG.7 is a flowchart of an example method for generating and analyzing vessel orientation feature data.

[0012] FIG. 8 is an example workflow for generating a binary image from a medical image (e.g., IQ ultrasound data) using, for example, a high-definition microvasculature imaging (HDMI) technique.

[0013] FIG.9 shows examples of (a) benign and (b) malignant breast tumors: B-mode images, ROI images (HDMI), and binary images in Cartesian and polar coordinates for (a) a benign breast tumor subject and (b) a malignant breast tumor.

[0014] FIGS.10A and 10B illustrate results from examples of an in vivo study, benign and malignant breast tumors, respectively.

[0015] FIGS. 11A and 11B illustrate box plots (left), ROC curves (middle), and confusion plots (right) for each orientation feature of vessels tested in an example in vivo study. 2 QB\630666.01598\97090978.4Mayo 2024-087 630666.01598

[0016] FIG. 12 shows results of an in-vivo study for selected benign and malignant samples. Row 1: lesion type. Row 2: B-mode image of each case with ROI and dilation mask. Row 3: binary image of polar HDMI image. Row 5: radial and tangential histogram plots along with the biomarker values: MRH, kurtosis^^, MTH, and kurtosis^^. Row 6: skeleton of microvessel structure in polar coordinate along with PGA values. Row 7: microvessel structure in contact with lesion boundaries for each case, along with the value of angle (θ) of microvessel segments and APD. Row 8: small and large microvessel structure, characterized as circumferential and penetrating using the PCD approach along with the values of PCD.

[0017] FIGS.13A–13F show two representative cases of reactive axillary lymph nodes in (a) and (d). (b) and (e) display the corresponding microvessel images derived from the axillary lymph nodes in (a) and (d), respectively. (c) and (f) present the polar-transformed representations of the microvessel images shown in (b) and (e), respectively.

[0018] FIGS. 14A–14F show two representative cases of metastatic axillary lymph nodes in (a) and (d). (b) and (e) display the corresponding microvessel images derived from the nodes in (a) and (d), respectively. (c) and (f) present the polar-transformed representations of the microvessel images shown in (b) and (e), respectively

[0019] FIGS.15A–15M show box plots of orientation biomarkers for differentiation of reactive and metastatic lymph nodes. (a) MRH, (b). Kurtosisr, (c) MTH, (d) Kurtosisθ, (e) PGA, (f) APD, (g) PCD, (h) NPV, (i) NCV, (j) WPD, (k) Diameter APD. (l) Diameter PCD, and (m) Orientation score. Biomarkers with * sign indicate statistically significant biomarkers, and the scoring method.

[0020] FIG. 16 illustrates an example description for computing orientation and extended orientation biomarkers (e.g., NPV, NCV, WPD, Diameter APD and Diameter PCD). APD^^^ൌ Number of penetrating microvessels identified using APD method. PCD^^^ൌ Number of penetrating microvessels identified using PCD method.

[0021] FIGS. 17A–17C illustrate malignant examples of hepatic tumor cases: hepatocellular carcinoma (FIG. 17A, FIG. 17B) and cholangiocarcinoma (FIG.17C). 1(a)-(c) B-mode image, 2(a)-(c) HDMI image, 3(a)-(c) Binary HDMI, 4(a)-(c) Binary polar HDMI, 5(a)-(c) APD algorithm based selected microvessel segment with their angles, 6(a)-(c) PCD algorithm based extracted large circumferential microvessel segment, and 7(a)-(c) PCD algorithm based extracted large penetrating microvessel segment; (a) HCC1, (b) HCC2, (c) CCA. 3 QB\630666.01598\97090978.4Mayo 2024-087 630666.01598

[0022] FIGS. 18A–18B illustrate example benign hepatic tumors. 1(a)-(b) B-mode image, 2(a)-(b) HDMI image, 3(a)-(b) Binary HDMI, 4(a)-(b) Binary polar HDMI, 5(a)-(b) based selected microvessel segment with their angles, 6(a)-(b) PCD algorithm based extracted large circumferential microvessel segment, and 7(a)-(b) PCD algorithm based extracted large penetrating microvessel segment.

[0023] FIGS. 19A–19M show box plots of benign and malignant lesions; a) MRH, b) Kurtosisr, c) MTH, d) Kurtosisθ, e) PGA, f) APD, g) PCD, h) NPV, i) NCV, j) WPD, k) Diameter APD, l) Diameter PCD, and m) orientation score.

[0024] FIGS. 20A–20M show box plots of benign and HCC lesions; a) MRH, b) Kurtosisr, c) MTH, d) Kurtosisθ, e) PGA, f) APD, g) PCD, h) NPV, i) NCV, j) WPD, k) Diameter APD, l) Diameter PCD; and m) orientation score.

[0025] FIG. 21 is a block diagram of an example system for generating and analyzing vessel orientation feature data.

[0026] FIG. 22 is a block diagram of example components that can implement the system of FIG.21.

[0027] FIG. 23 is a block diagram of an example ultrasound system, which can implement some embodiments described in the present disclosure. DETAILED DESCRIPTION

[0028] Described here are systems and methods for quantifying or otherwise characterizing blood vessel orientation based on blood vessel ultrasound imaging. Advantageously, the blood vessel orientation features generated using the techniques described in the present disclosure can be used to assess tumor malignancy and other disease states and / or pathological conditions. As a non-limiting example, the penetration and tangential behavior of blood vessels (e.g., microvessels) can be quantified in order to assess tumor malignancy. Several orientation-based features can be generated as quantitative biomarkers of tumor condition. The orientation features can be computed in polar or Cartesian coordinates, and can be generated without the need for a contrast agent.

[0029] Advantageously, the blood vessel orientation features generated using the disclosed techniques show significance in differentiating malignant from benign breast tumors, amongst others. Vessel orientation can be affected by tumor malignancy before changes in vessel density and morphology; thus, the disclosed techniques can provide complementary 4 QB\630666.01598\97090978.4Mayo 2024-087 630666.01598 information to those methods based on vessel density and morphology, and can therefore enable earlier diagnoses.

[0030] It is another advantage that the disclosed techniques can be used in applications other than ultrasound imaging and tumor assessment. For example, the disclosed methods can be used for retinal disease classification using optical coherence tomography (OCT) and / or fundus image data. Additionally or alternatively, the disclosed methods can be adapted for use in lesion classification for diagnosis of malignancy using different imaging modalities, including magnetic resonance imaging (MRI) and computed tomography (CT).

[0031] The penetration and tangential behavior of blood vessels can be quantified, which can further help in classifying abnormal behavior of blood vessels. As a non-limiting example, seven orientation features can be generated as quantitative biomarkers. Four of the orientation features are histogram features in polar coordinates to analyze radial (or penetrating) vessels and tangential (peripheral) vessels. The combined histogram features (e.g., mean and kurtosis) can reflect the density distribution of vessels based on orientation. The polar gradient angle (PGA) is another orientation feature, which can be calculated in polar coordinates to quantify the mean gradient angle. Additionally or alternatively, two orientation features can be calculated in Cartesian coordinates: namely angle-based penetration density (APD) and penetration to circumferential density (PCD). Both of the Cartesian coordinate- based orientation features categorize penetrating and peripheral vessels based on the angle by which the vessel enters a region-of-interest (ROI) and based on the penetration level of vessels toward the center portion of ROI. In still other examples, other orientation features that may be generated and used as quantitative biomarkers include weighted penetration density (WPD), number of penetrating vessels (NPV), number of circumferential vessels (NCV), diameter APD, and diameter PCD. Additionally, in some cases an orientation score can be computed based on various orientation features.

[0032] In tumor classification applications, it has been observed that the dense penetrated (or radial) orientation of blood vessels is found to be associated with malignancy, and circumferential (or tangential) orientation with sparse distribution is found to be associated with benign lesions. For computing the directionality of vessels, the disclosed orientation features can be computed in both polar coordinate and Cartesian coordinates. The advantage of converting images into a polar coordinate is that the penetrating (radial) vessel can be seen as a horizontal distributed vessel and circumferential vessels as vertical so, this can help in visualizing and understanding the orientation features. 5 QB\630666.01598\97090978.4Mayo 2024-087 630666.01598

[0033] To convert an image from Cartesian coordinates ^x, y^ to polar coordinates^r, θ^ the following relation can be used:r ൌ ^^xଶ ^ yଶ^ (1);θ ൌ tanି^^y⁄ x ^ (2);

[0034] where r varies from 0 to R, where R is half of the maximum of diameter of the ROI; and θ varies from 0 to 360 degrees. For computing diameter, first the mean of the ROI mask can be computed along rows and columns to obtain row and column vectors. The diameter of the ROI along row and column directions can be computed as the number of non- zero values in the obtained vectors. The maximum of the row and column diameters can be used to compute R.

[0035] The center and / or center-of-mass of the ROI can be computed using a region growing technique, or other suitable techniques. FIG.1 shows an example Cartesian image 102 and a corresponding converted polar coordinate image 104. The circle 110 in the Cartesian image 102 can be considered as boundaries of the ROI. Arcs 112 and lines 114 can be considered as circumferential (or tangential) blood vessels 122 and penetrating (or radial) blood vessels 124 in the ROI, respectively. The blood vessels that are penetrating in Cartesian coordinates will be seen like vertical lines in polar coordinates. Blood vessels that are circumferential look like horizontal lines in polar coordinates. Penetrating blood vessels are those vessels that are entering from the boundaries of the ROI and move toward the center of the ROI, whereas circumferential blood vessels enter the ROI boundaries, but move along the periphery of the ROI. As mentioned above, four histogram features and gradient angle can be computed in polar coordinate to estimate the orientation of blood vessels.

[0036] As a non-limiting example, an image matrix (e.g., a binary polar image) can be converted into a histogram that may be represented as a vector where the horizontal axisrepresents bin value which is r (or θ) value (r = [0,1, 2…,R]; θ ൌ ^0,1,2, … ,360^) and thevertical axis represents frequency or number of counts at the r௧^(or θ௧^) row (or column). Radial histogram and tangential histogram vectors can be obtained by averaging a polar image along r or along θ, respectively. Histogram features such as mean and kurtosis can be computed to help visualize the density and spread of blood vessels along radial and tangential directions. Mean of the radial histogram (MRH) and mean of the tangential histogram (MTH) measure the mean density of blood vessels in an image along the radial and tangential directions, respectively. Similarly, kurtosis of radial and tangential histograms can be generated to measure the tailness (or peakedness) in the distribution of blood vessels along the radial and 6 QB\630666.01598\97090978.4Mayo 2024-087 630666.01598 tangential directions. Additionally or alternatively, other histogram features or parameters can be used to describe the distribution of pixels in an image. For example, the histogram features may include other first-order and / or second-order histogram features, such as standard deviation, correlation, entropy, skewness, and so on, or other higher order moments. These additional or alternative histogram features may similarly be used to quantify or otherwise assess blood vessel orientation.

[0037] When a distribution is peaky (e.g., tall and thin) the kurtosis will be high and mean density will likely be low. Kurtosis will be low if the distribution is flat (e.g., having a long tail) and, in that case, general mean density will be high. FIG.2 illustrates an example of the variation in kurtosis and mean value of a flat distribution, an irregular and dense distribution, and a peaky distribution of a histogram. If blood vessels are highly dense and distributed throughout the ROI, the MRH will be flat or irregular and densely distributed, and if blood vessels are concentrated either on circumferential regions or in the central region of ROI, the distribution will be peaky. So, depending upon the distribution of directional blood vessels, the kurtosis and mean distribution of the histogram will change and hence all four features: MRH, MTH, kurtosisr(kurtosis of radial histogram), and kurtosisθ(kurtosis of tangential histogram) together will help in quantifying ROI based on blood vessel orientation. FIG. 3 shows an example of histogram features and an illustration of features with only penetration vessels, only circumferential vessels, and having both penetrating and circumferential vessels.

[0038] The radial and tangential gradients correspond to change of pixel values along vertical and horizontal direction of a polar ROI skeleton image in polar coordinates. A skeleton image can be generated, for example, by reducing the foreground region of a binary image (in polar coordinates) into a single-pixel-wide connected line. For example, a morphological thinning algorithm can be used to generate a skeleton, or skeletonized, image from a binary image. The gradients of the polar image along θ and r can be represented as ^^ఏand ^^^, respectively. As a non-limiting example, a gradient operator such as the Sobel operator can be used for calculating the gradient of the image. Additionally or alternatively, other gradient operators such as Robert or Prewitt operators may be used. The (3×3) filter kernel of the Sobel operator can be represented as: ^1 2 1 ^.Mayo 2024-087 630666.01598

[0039] From the computed ^^ఏand ^^^, the angle Θ can be computed as Θ=tanି^(^^^ / ^^ఏ). To illustrate how PGA reflects the orientation, the blood vessels in the example images of FIG. 1 can be analyzed, as illustrated in FIG. 4, which shows skeleton images (i.e., skeletonized binary images) of circumferential and penetrating blood vessels. In the illustrated example, the circumferential blood vessel (which is horizontal in polar coordinates) has an absolute mean Θ equal to 0 degrees, whereas the radial blood vessels (which are vertical in polar coordinates) have a mean Θ equal to 180 degrees. In general, the PGA will be around 0 degrees if the lesion has only circumferential vessels (or horizontal vessels in polar coordinates). The PGA will be around 180 degrees if the lesion has only radial vessels (or vertical vessels in polar coordinates). PGA of the ROI skeleton image computes the mean gradient angle (MGA) of the skeleton of each ith vessel segment component (MGAi), followed by computing the weighted mean of the mean gradient values, MGAi. The weight (Wi) of each vessel segment depends on the area of the vessel segment. The vessel segment can be obtained using a connected component analysis. Connected component analysis separates group of pixels that are connected and labels them. To eliminate the sudden change of angle at the end points of the skeleton it is possible to remove Θ values at the start and end points of each skeleton connected segment, making the MGAi values invariant to change in the length of the skeleton. The mathematical expression for PGA can be given as: N ^W^MGA.

[0040] can provide complementary information to those generated from polar coordinates. Thus, the systems and methods described in the present disclosure can additionally or alternatively generate vessel orientation features from images having Cartesian coordinates. For example, vessel orientation features can quantify the orientation of blood vessels in terms of angle by which those blood vessels are entering the ROI and based on penetration level of blood vessels inside the ROI.

[0041] One example of a Cartesian coordinate-based orientation feature is angle-based penetration density (APD). The proposed framework to compute APD in Cartesian coordinates is illustrated with respect to FIG. 5. APD can be computed by finding the blood vessels that are entering the ROI and which are penetrating inward or just present in the outer periphery of 8 QB\630666.01598\97090978.4Mayo 2024-087 630666.01598 the ROI (i.e., are circumferential). Using connected components analysis, each blood vessel inside the ROI can be separated out. Then, out of all the vessel segments, only those vessel segments that are touching the ROI are considered for the APD measurement. This can be done by finding the blood vessels that share one or more pixels with the boundaries of the ROI (i.e., that have at least one common boundary point with the ROI). Then, the angle between the ROI boundary points and the selected blood vessel is computed by estimating two straight lines: one representing the tangent to the ROI boundary and the other representing the selected vessel. Several coordinates of the ROI boundary (e.g., 10 boundary points) from the intersection (i.e., common coordinate point) can be used. A curve fitting algorithm can then be applied to estimate a straight line of the ROI tangent to the ROI boundary. For the vessel, the coordinates of skeleton points (obtained by skeletonizing the binary image of the vessel segment in Cartesian coordinates) can be used to estimate a straight line that passes through the intersection point of the ROI and blood vessel. To improve the accuracy of the blood vessel straight line estimation, constraints and conditions can be provided to the curve fitting algorithm, such as the starting point and the slope range for the estimated line. The slopes of both lines (ROI boundary and vessel) are then used to calculate the angle, which can be expressed as follows: ^^tan^ 1 ^ m^ 2 ^ m 1 ^^;

[0042] respectively. If the angle is between some particular range (e.g., 30–150 degrees, as a non- limiting example), then the blood vessel is penetrating, else it is classified as circumferential. The ratio of the area of penetrating components to the sum of the area of total blood vessels is calculated as the APD.

[0043] As a non-limiting example, the slope, ^^1 can be computed as follows. First, skeletonization is performed to extract the skeleton of the blood vessel to achieve a single- pixel-wide representation of the blood vessel while preserving its structure. This is accomplished using a morphological thinning algorithm, which iteratively removes pixels until only the central skeletal structure remains. Next, endpoints and branch points are identified. The endpoints can be identified as pixels on the skeleton that have exactly one neighboring pixel, and the branch points can be identified as pixels with more than two neighboring pixels. Branch isolation is then performed by removing the branch points to isolate the individual branch(es) from the skeleton, resulting in disconnected branches. A straight line is then fit to 9 QB\630666.01598\97090978.4Mayo 2024-087 630666.01598 each disconnected branch using a curve fitting algorithm. The slope and intercept are recorded. The branch that contacts the ROI boundary is then identified by finding a branch sharing a point with the boundary. Then, the slope and intercept of this branch are used as initial conditions for the estimation of the final line fitted to the blood vessel. In cases where a blood vessel segment has multiple entry points, the first contact point during a left-to-right scan can be considered as the point of contact.

[0044] A line is then fit to the binary blood vessel coordinates (bx, by) under the following two constraints: (1) the line must originate at a contact point (ix, iy), where the initial branch intersects the ROI boundary, and (2) the slope of the line must lie within the rangedefined by the minimum and maximum slopes ^m1min , m 1 max ^ of all isolated branches. Then,compute the slope m1 that minimizes the mean squared error (MSE) function, defined as: MSE ^ m1 ^^^^ by ^^ m 1 ^ b x ^ ^ i y ^ m 1 ^ i x ^ ^ ^

[0045] between the slopes ^^1 and ^^2can be estimated. If the angle is between 30° and 150°, the blood vessel is classified as penetrating; otherwise, it is classified as circumferential. Under ideal circumstances, the angle range for a circular ROI would be between 45° and 150° for circular ROI. Due to the irregular shape of the ROI, this range (30° and 150°) was determined using a trial-and-error approach. To quantify the penetrating blood vessels, the APD is calculated, which as described above is defined as the ratio of the penetrating blood vessel area to the total blood vessel area. The area of binary images of blood vessel segments can be computed by summing the number of pixels corresponding to each region.

[0046] Another example of a Cartesian coordinate-based orientation feature is the penetration to circumferential density (PCD). This feature computes the ratio between the area of penetrating vessels to the area of circumferential vessels. Area can be computed by the number of ones in a binary image of penetrating and circumferential vessels. For the PCD feature, the goal is to determine if the blood vessel is penetrating or circumferential by estimating the penetration level of the vessels, which is an indication of how the blood vessel is oriented towards the center portion of the ROI. The estimation can be done by measuring the vessel characteristics like penetration distance (PD) and penetration factor (PF). The penetration distance (PD) can be estimated by computing the difference between the distance from the start point of a vessel skeleton to the center of the ROI (Dstart) and the distance from 10 QB\630666.01598\97090978.4Mayo 2024-087 630666.01598 the end point of a vessel skeleton to the center of the ROI (Dend), as illustrated in FIG. 6. Mathematically, this can be expressed as: PD^ D start ^ D end (6).

[0047] if PD is equal. called penetration factor (PF) can be used, which may be calculated as PF= PD / SL. Practically, an ROI may have an irregular (i.e., non-circular) in shape, so just having a significant PF does not necessarily mean that the blood vessel is approaching toward center of the ROI, because there is a possibility that the vessel is around the periphery of the ROI. Therefore, in addition to having a significant PF, a vessel can be considered as penetrating when a portion of the vessel is contained inside the central portion of the ROI. As an example, conditions for considering a vessel as penetrating are having a PF > 0.7 and such that the vessel is contained in central portion of the ROI (e.g., as a non-limiting example, the central portion can be a 0.7 scaled version of the original ROI as illustrated in FIG.6).

[0048] Another example of a Cartesian coordinate-based orientation feature is the weighted penetration density (WPD), which can be computed as a weighted ratio of the sum areas of penetrating blood vessel segments to the areas of all blood vessel segments. The area of blood vessel segments may be weighted, for example, based on the confidence of being a penetrating segment. The blood vessel segments that are characterized as penetrating based on both APD and PCD techniques will be weighted twice more than the segments characterized based on only one of the techniques.

[0049] Still another example of Cartesian coordinate-based orientation features are the number of penetrating vessels (NPV) and number of circumferential vessels (NCV). NPV is computed based on the maximum NPV in both the APD and PCD algorithms. NPV based on APD and PCD is represented as APD^^^and PCD^^^, respectively. NCV can be calculated by subtracting NPV from the total blood vessel segments.

[0050] A diameter matrix can be used to estimate additional orientation features, such as the diameter APD and diameter PCD. The diameter of blood vessel segments is computed based on a distance metric, such as the Euclidean distance measure, which is applied to a binary image. As an example, the binary image may have zero values for blood vessels (i.e., the blood vessels will appear black) whereas background will be assigned values of one (i.e., the background will be appear white). Alternatively, the binary image can use values of one for blood vessels and values of zero for background. The diameter matrix approach measures the 11 QB\630666.01598\97090978.4Mayo 2024-087 630666.01598 distance between a blood vessel pixel and the nearest background pixel and then stores that distance measure in a pixel location corresponding to the location of the blood vessel pixel. In this way, the diameter matrix measures how far each point in the surrounding area is from the center of the vessel. The distance farther from the vessel edge will lie in the middle of the vessel segment, and the distance value corresponds to the vessel radius. Twice the measure will give the vessel diameter at different cross sections. The diameter measures that may be used as biomarkers include the diameter APD and diameter PCD. The diameter APD and diameter PCD are computed based on the ratio of the maximum penetrating blood vessel diameter to the maximum of the circumferential blood vessel diameter.

[0051] Referring now to FIG. 7, a flowchart is illustrated as setting forth the steps of an example method for generating vessel orientation feature data from medical image data, such as ultrasound image data, OCT image data, MR image data, CT image data, or the like.

[0052] The method includes accessing image data with a computer system, as indicated at step 702. Accessing the image data may include retrieving such data from a memory or other suitable data storage device or medium. Additionally or alternatively, accessing the image data may include acquiring such data with an ultrasound system, or other imaging system, and transferring or otherwise communicating the data to the computer system, which may be a part of the ultrasound system, or other imaging system.

[0053] The image data may be blood vessel image data that includes one or more blood vessel images that depict the vasculature (e.g., blood vessels, microvessels) in a region of a subject from which the images were obtained. In other instances, the image data contain a sequence of images (e.g., ultrasound images, such as a sequence of plane wave ultrasound images). In these instances, one or more blood vessel images are first reconstructed from the sequence of images, as indicated at optional step 704.

[0054] In any such instance, the image data may be acquired with or without the use of an contrast agent (e.g., a microbubbles-based contrast agent). The image data may be two- dimensional image data or three-dimensional image data. As a non-limiting example, the image data can include breast tumor images acquired using an ultrasound system with a linear array running at 8.5 MHz. The images can be acquired using a plane-wave imaging mode used to identify breast masses in the B-mode images. Additionally or alternatively, the image data may include a series of high frame rate data (e.g., about 600 frames per second) captured at each region-of-interest (e.g., each tumor location). 12 QB\630666.01598\97090978.4Mayo 2024-087 630666.01598

[0055] As a non-limiting example, when the image data contain ultrasound plane wave image data, the data can be stored or otherwise converted into an in-phase–quadrature (IQ) format. These data can be characterized by a complex-valued variable that is a function of thelateral dimension ^x ^ , axial dimension ^z ^ , and time ^t ^ (i.e., ultrasound imaging slow time).For instance, the signal can bes^ x, z , t ^^ c ^ x , z , t ^ ^ x , z , t ^ ^ n ^ x , z , t ^ (7);

[0056] n ^ x, z , t ^ isnoise. Spatial andof these threecomponents differ. The additive thermal noise can be considered as zero Gaussian white noise.The signal corresponds to a signal data tensor, S , with dimensions n x^ n z ^ n t , where n x isthe number of spatial samples along the x-direction (i.e., the, n z is the numberof spatial samples along the z-direction (i.e., the axial direction), and n t is the number ofsamples over time. The signal data tensor can be reshaped toCasorati matrix bytransforming the tensor into a two-dimensional spatiotemporal matrix, S C , with dimensions^nx^ n z ^ ^ n t . For example, 3D image data may be converted into reshaped image data byreshaping the 3D data into 2D reshaped image data by vectorizing each image into one column.

[0057] Using a singular value decomposition (SVD) of this spatiotemporal matrix results in, S^† CUΔV (8);

[0058] where Δ is an ^nx^ n z ^ ^ n t non-square diagonal matrix having diagonalelements that are arranged in decreasing order of energy and spatial coherence, U is an^nx^ n z ^ ^ ^ n x ^ n z ^ matrix, V is an n t^ n t matrix, and “†” indicates the conjugatetranspose. The matrices U and V are orthonormal matrices, and the columns of these matricescorrespond to the spatial and temporal singular vectors, respectively, of S C .

[0059] The higher values in Δ correspond to tissue signal. In some implementations,the tissue signal can be suppressed using energy based threshold selection by making the valuesof U , Δ , and V equal to zero. A tissue suppressed (^^^) signal can then be obtained by13 QB\630666.01598\97090978.4Mayo 2024-087 630666.01598 multiplying all three modified matrices. The matrix ^^^ can resized (e.g., to the same size as the original IQ data) and an intensity image can be created. For example, an intensity image can be generated using the following relation: NtI x , z ^ S ^ x , z2 ^^ ^ ^ , kt ^ (9).^

[0060]

[0061] Alternatively, the matrix, C , can be decomposed into a sum of rank onematrix, SC ^^^ i A i ^ ^ ^ i U i ^ V ii i(10); ith

[0062] ordered singular value of S C ; and “^ ” denotes the outer product operation. Each column V iis a temporal signal witheach column U i is a spatial signal withn x^ n z (i.e., a two-dimensional spatial image, I i ); hence,rank^SC ^s ^ x , z , t ^^^^ i I i ^ x , z ^ V i ^ t ^(11).

[0063] framework, the tissue component can be considered to correspond to the first few dominant singular values and vectors, while the blood signal can be formed by the subsequent singular values when sorted in a descending order. Based on these assumptions on tissue and blood signals, clutter removal is performed using athreshold, n , on the number of singular vectors removed from s ^ x, z , t ^ . Therefore, the bloodsignal can be derived as follows,n s^ x , z , t ^^s ^ x , z , t ^^^^ I ^ x , z ^ V ^ t ^(12).

[0064] , a threshold on the slope ofthe second order derivative of the eigenvalues decay. The filtered blood sblood ^ x, z , t ^, can be used to produce a power Doppler image as,K ^2^ ^^^ ^;Mayo 2024-087 630666.01598

[0065] where T is the sampling time between two successive ultrasound frames, whichmay be successive ultrafast ultrasound frames. The clutter removal performance can be further enhanced by enforcing a unilateral Doppler shift, which is expected to occur from the unidirectional flow in vessels, before forming the intensity image in Eqn. (13). In these instances, the final image can be formed as, I^ x, z ^^ I p ^ I n (14);

[0066] where Ipis the energy at the positive frequency side of the spectrum, ^ IS x , z , 2p^^blood ^ f ^ df;

[0067] sblood ^ x, z , t ^ , and I n is the energy at the negative frequency side of the spectrum,0 In^^S blood ^ x , z , f2 ^df(16).

[0068] be filtered to remove background noise. As oneexample, the image can be filtered using a top hat filter (THF). A THF implements a background estimation followed by a background subtraction operation. The output image ofthe THF can be denoted as IT ^ x, z ^ .

[0069] Theare then preprocessed, as generally indicated at step 706. In some instances, the blood vessel images may have already been preprocessed (e.g., the blood vessel images stored in the memory or other data storage device or medium are preprocessed blood vessel images), in which case step 706 can be skipped. In general, preprocessing the blood vessel images can include vessel filtering to enhance the structure of vessels depicted in the blood vessel images, and to provide adequate background separation for segmentation.

[0070] As one example of preprocessing a blood vessel image, to enhance the visibility of the blood vessel image in the presence of strong background signals, morphological filtering based on a THF can be used. Due to background noise, however, random patterns may also be present at the output of a THF. Hence, in some implementations, vessel enhancement filters can be used to penalize background noise and further enhance vessel structure. Enhancement 15 QB\630666.01598\97090978.4Mayo 2024-087 630666.01598 filters based on the analysis of eigenvalues of a Hessian matrix applied on a two-dimensional image selectively amplify a specific local intensity profile or structure in an image. Hessian- based filters distinguish between different local structures by analyzing the second order intensity derivatives at each point in the image. To enhance the local structures of various sizes, the analysis is typically performed on a Gaussian scale space of the image.

[0071] A region-of-interest (ROI) image is generated from the blood vessel image, as indicated at step 708. For example, the ROI image can be obtained by multiplying a mask (e.g., an ROI mask) with the blood vessel image and / or enhanced blood vessel image. The ROI mask can be a binary image where pixels within boundaries of ROI are ones and all other pixels are zero.

[0072] Finally, the blood vessel image, enhanced blood vessel image, and / or ROI image can be converted to a binary image by a binarization operation, as indicated at step 710. An example workflow for generating a binary image is also illustrated in FIG.8.

[0073] Blood vessel orientation feature data are then generated from the binary image(s), blood vessel image(s), and / or ROI image(s), as indicated at step 712. In general, the vessel orientation feature data are indicative of the orientation of blood vessels in tissue. For example, the vessel orientation feature data can quantify or otherwise characterize the orientation of blood vessels in tissue. In this way, the vessel orientation feature data can provide one or more quantitative biomarkers for assessing the condition of a tissue. Vascular-based pathological conditions (e.g., tumor malignancy, other tumor classification, other vascular disease conditions or states, etc.) can thus be assessed based on the vessel orientation feature data.

[0074] The vessel orientation feature data may include polar coordinate features (e.g., histogram features, polar gradient angles (PGAs)), Cartesian coordinate features (e.g., angle- based penetration density(APD), penetrating to circumferential density (PCD)), or other such features that quantify or otherwise characterize blood vessel orientation. Histogram features may include histogram mean, kurtosis, or other statistical measures that can be computed or otherwise derived from a histogram. For example, histogram features may include the mean of the radial histogram (MRH), which measures the mean distribution of blood vessels along the radial direction; the mean of the tangential histogram (MTH), which measures the mean distribution of blood vessels along the tangential direction; kurtosis of radial histogram, which measures tailness (or peakedness) in the distribution of blood vessels along the radial direction; 16 QB\630666.01598\97090978.4Mayo 2024-087 630666.01598 and / or kurtosis of tangential histogram, which measures tailness (or peakedness) in the distribution of blood vessels along the tangential direction.

[0075] The vessel orientation feature data can then be displayed to a user, stored for later use or further processing, or both, as indicated at step 714. For instance, the vessel orientation feature data may be displayed in connection with a graphical user interface that enables a user to visualize the vessel orientation feature data (e.g., as images, maps, or other display or textual elements), manipulate the vessel orientation feature data, or otherwise interact with the vessel orientation feature data. As described above, the vessel orientation feature data can be analyzed to classify a disease type in a body organ. For instance, the vessel orientation feature data can be analyzed to differentiate between malignant and benign lesion. As another example, the vessel orientation feature data can be analyzed to stage a disease progression of a tissue from which the blood vessel image data was acquired. As yet another example, the vessel orientation feature data can be analyzed to monitor a response to a treatment provided to a tissue from which the blood vessel image data was acquired.

[0076] In some cases, an orientation score can be computed for the subject based on the orientation feature data. As a non-limiting example, the following strategy can be used to compute an orientation score from multiple biomarkers in the orientation feature data: Table 1: Example Orientation Score Entry Criteria Entry Criterion Comments 1 a) MRH>10 and b) a) Indicates a uniform distribution of blood vessels along the ). ar ar an of an on he he of s.QB\630666.01598\97090978.4Mayo 2024-087 630666.01598 8 Diameter PCD > 1 Indicates high penetrating pattern; a larger diameter of penetrating blood vessels than circumferential blood vessels. Characterization of orientation based on PCDan eight-element binary vector, where each entry flags a penetrating (1) and circumferential (0) pattern based on the entry criteria shown in Table 1. The binary entries in this vector are summed to generate the orientation score, which in this example has values between 0 and 8, where 0 signifies a purely circumferential structure and 8 a purely penetrating structure. Values near 8 indicate dominating penetrating pattern and value near 0 indicate dominating circumferential pattern.

[0078] In an example study, a total of 70 subjects with ultrasound-identifiable breast lesions were imaged, with 35 having benign breast tumor and 35 having malignant breast tumor. Following an ultrasound examination, core needle breast biopsies were performed on all patients, and the pathology reports were used to make the final diagnosis. The B-mode (ROI marked), ROI image, binary image in Cartesian coordinates, and binary image in polar coordinates for one benign breast lesion and one malignant breast lesion are shown in FIG.9.

[0079] The illustration of all twelve biomarkers for one benign and malignant case (lesion shown in FIG. 9) is shown in FIGS. 10A and 10B. FIG. 10A shows the biomarker corresponding to benign case whereas FIG.10B corresponds to the malignant case. Histogram features, PGA, APD, PCD values and an illustration for both lesions are shown in FIGS.10A and 10B. MRH and MTH are mean distribution of blood vessels in radial and tangential direction whereas, values of kurtosis^and kurtosisѲreflect the distribution of blood vessels in the ROI along radial and tangential directions. PGA values show the angle of weighted mean gradient direction. APD and PCD show the penetration density and should be low (minimum 0) for a benign case and high value for a malignant case due to more penetrating blood vessels compared to benign lesion.

[0080] To find the similarity between the features, correlation coefficients plot between twelve biomarkers and orientation score is computed and is shown in Table 2, the smaller the correlation value the lesser they are dependent on each other; therefore, their contributions are added in the diagnosis of breast cancer. Table 2: Correlation matrix of all twelve features and orientation score. MRH Kurtosis MTH Kurtosis PGA APD PCD NPV NCV WPD Diameter Diameter Orientation APD PCD score18 QB\630666.01598\97090978.4Mayo 2024-087 630666.01598 Kurtosis1 -0.21 0.86 -- - - -0.23 -0.29 -0.27 -0.23 0.17 0.23 0.22 0.10 0.23 MTH 1 -0.34 0.28 0.28 0.51 0.79 0.60 0.73 0.46 0.72 0.67 Kurtosis 1 - - - - -0.25 -0.43 -0.27 -0.39 -0.28gly correlated and the value below 0.45 it is considered as highly independent features. Negative correlation indicated two features tend to move opposite. The Wilcoxon rank-sum test is used to compute p-value and visualize the statistical significance of these biomarkers. The box plots, ROC curves, and confusion plots of all twelve features in the decreasing order of rank and the orientation score are shown in FIGS.11A and 11B.

[0082] To differentiate between benign and malignant breast tumors, the disclosed systems and methods utilize orientation-based vessel features as quantitative tumor biomarkers; namely, MRH, Kurtosis^, MTH, KurtosisѲ, PGA, APD, PCD, WPD, NPV, NCV, diameter APD, diameter PCD, and orientation score. The biomarkers were studied on in vivo breast tissue lesion data to classify benign class from malignant class.

[0083] To explain each biomarker for quantification of benign and malignant case, reference can be made to FIGS. 10A, 10B, 11A, and 11B. FIGS. 10A and 10B illustrate biomarker values for one benign and one malignant case, respectively. FIGS. 11A and 11B show distributions of values of features in both cases (in vivo study). The histogram features MRH, MTH, Kurtosis୰and KurtosisѲdemonstrate the variation of density and distribution of blood vessels polar image i.e., in radial (r) and circumferential direction (^^). From FIGS.10A, 10B, 11A, and 11B, the benign case (generally have circumferential and sparsely distributed blood vessels) has higher Kurtosis୰than malignant cases. For angular (^^) histograms, the reverse is expected; that is, KurtosisѲwill be higher for malignant lesions than benign, but may not always be true due to sparse distribution, so combining distribution (kurtosis) with density (MRH and / or MTH) can help in correctly classifying a lesion. The histogram mean density features (MTH and MRH) will be higher for malignant cases than benign cases as blood vessels are highly distributed both circumferentially and radially in malignant cases. 19 QB\630666.01598\97090978.4Mayo 2024-087 630666.01598

[0084] The PGA feature calculates the mean gradient direction of the polar image. For the benign cases, the mean gradient angle is expected to be around 0 degrees as they will have majority of circumferential blood vessels, so the gradient angle should be in the horizontal direction. On the other hand, in malignant cases the mean gradient direction can have both circumferential and penetrating blood vessels with high density, so it is expected to around 180 degrees (vertical) or between 0 and 180 degrees (diagonal), which is reflected in the in vivo study PGA around 36 degrees for benign and 113 degrees for malignant. From FIGS.11A and 11B, it can be seen that the distribution of PGA values for malignant lesions is higher than the benign lesions.

[0085] The APD feature is used to distinguish the blood vessels which are entering the ROI and are penetrating toward center of ROI or are circumferential to the ROI. If ^^^is between 30^to 150^than blood vessel ^^ will be considered as penetrating blood vessel. As penetrating blood vessels will only be there in malignant lesions, APD will be high compared to benign lesions, which can be seen in FIGS.10A, 10B, 11A, and 11B.

[0086] PCD measures the ratio of the area of penetrating blood vessels to the area of circumferential blood vessels. PCD distinguishes penetrating blood vessels based on penetrating factor of blood vessels and approaching toward the center of the ROI.

[0087] In another example study, 70 subjects, each having at least a single ultrasound- identifiable breast lesion, mostly classified as BI-RADS 4 and up, and recommended for core needle breast biopsy were imaged. Following the ultrasound examination, core needle breast biopsies were performed on all patients, and the pathology reports were used to make the final diagnosis.

[0088] Image acquisition and processing: In vivo images of breast tumor microvessels were acquired using a high-definition microvessel imaging (HDMI) method. An ultrasound system equipped with a linear array running at 8.5MHz was used to acquire the images.

[0089] Microvessel image formation included the following steps. First, plane-wave imaging mode was employed to identify the breast mass in the B-mode image. The ROI was marked by an expert sonographer on the plane-wave B-mode image during acquisition. Subsequently, a series (~2000) of high-frame-rate images at 600 frames per second were captured at each tumor site, where each frame was formed using 5-angle coherent plane-wave compounding. After applying singular value decomposition (SVD) processing, background noise was removed and vessel enhancement filtering was performed. This resulted in the final HDMI image of the tumor microvessels. Then, through a series of morphological filtering and 20 QB\630666.01598\97090978.4Mayo 2024-087 630666.01598 vessel segmentation, the HDMI image was converted into a binary image, the skeleton was identified, and the vessel segments were extracted. Orientation features were than calculated using the methods described in the present disclosure.

[0090] For illustrating the biomarkers obtained from the in vivo study, four subjects were considered: two with benign and two with malignant lesions, encompassing both small and large lesion sizes. Their results are illustrated in FIG.12. The four lesions, benign 1, benign 2, malignant 1, and malignant 2, were 7 mm, 25 mm, 8 mm, and 43 mm in size, respectively. In the B-mode image, the lesion is outlined in blue by the sonographer. The blue outline was dilated by 2 mm and shown as the green border enclosing the actual lesion. Dilation makes a second boundary to include peritumoral vascularity, which is prominently seen in malignant tumors. It is noted that the disclosed techniques use the pattern of tumor microvessels for tumor classification, not the boundary features. The 2mm dilation was empirically chosen, as studies have shown that most peritumoral vasculature typically falls within this range from the tumor boundaries. FIG.12 displays the B-mode image, HDMI microvessel image, and binarized polar coordinates for each lesion in the first three rows. The following rows of FIG. 12 present the values of seven biomarkers: histogram features, PGA, APD, PCD values, and illustrations of the ROI and central ROI for all four lesions. In the PGA section, the images display the lesion microvessel skeleton in red in the polar coordinates with their respective PGA values. In the APD section, the images show the binary image of the microvessels (in white) that are in contact with the ROI boundaries (depicted in grey) in Cartesian coordinates. The angle ^^i represents the computed angles for each microvessel segment (i) in contact with ROI, calculated using the APD algorithm. In the PCD row, each lesion is represented by four images. The top two images in the PCD section illustrate small microvessels oriented either circumferentially or penetrating the ROI. In contrast, the bottom row shows large microvessels with similar circumferential or penetrating orientations.

[0091] In another example study, the orientation features and orientation score described in the present disclosure were analyzed as diagnostic features for classifying reactive and metastatic lymph nodes, providing a quantitative and objective framework to inform clinical decision-making in breast oncology. As described above, various orientation features can be computed and used as quantitative biomarkers. In some cases, the twelve biomarkers described above may include NaN or infinite values. The NaNs mean there is structural missingness, i.e., no microvessel segments (NPV=NCV=0). In such cases, the NaN value can be replaced with zero, reflecting the absence of measurable values. Infinite values can appear 21 QB\630666.01598\97090978.4Mayo 2024-087 630666.01598 in metrics like PCD and diameter biomarkers when no circumferential vessels are detected. These infinite values can be handled by substituting Inf with the maximum value in that row, yielding a finite value that still signals a strong penetration vessel.

[0092] The twelve biomarkers were used to generate an eight-element binary vector, where each entry flags a penetrating (1) and circumferential (0) pattern based on criteria. The entry criteria and their meaning are shown in Table 1. Summing the binary entries in this vector will give an orientation score between 0 and 8, where 0 signifies a purely circumferential structure and 8 a purely penetrating structure. Values near 8 indicate dominating penetrating pattern and value near 0 indicate dominating circumferential pattern.

[0093] Another study included 68 subjects with recently diagnosed breast cancer who had suspicious axillary lymph nodes (ALNs), identified in clinical ultrasound imaging. The pathology findings from fine needle aspiration biopsy were used as ground truth for the label.

[0094] Ultrasound imaging was performed using an ultrasound system equipped with a linear array transducer operating at 8.5 MHz. Plane-wave imaging mode was used to acquire high frame rate data (~600 frames per second) on ALN sites. To reduce motion artifacts, subjects were asked to hold their breath for three seconds and remain still during the data acquisition phase. Further to improve repeatability, two acquisitions were made at each orientation. The ALNs were segmented manually in a B-mode image (obtained by reconstructing acquired in-phase quadrature data).

[0095] The acquired in-phase quadrature frames were then used to generate HDMI. This process involved singular value decomposition-based clutter filtering, followed by denoising using top hat filtering and vessel enhancement using the Hessian filter. The biomarkers were then acquired from the binarized and skeletonized version of HDMI.

[0096] FIGS.13A–13F and 14A–14F illustrate two cases of reactive and two cases of metastatic ALNs, respectively. For each case, the figures display: (a) and (d) shows the conventional B-mode image with lesion boundaries (blue outline) and 2 mm dilation (green out line); (b) and (e) correspond to HDMI based microvessel image; and (c) and (f) correspond to the polar-coordinate HDMI representation. Below each case, the calculated values of the twelve quantitative biomarkers and the overall orientation score for that node are presented.

[0097] FIGS.15A–15M show the boxplot of each biomarker and orientation score. The statistically significant biomarkers are highlighted using the * sign. The statistical analysis suggest seven (MRH, MTH, NPV, NCV, WPD, diameter APD, and diameter PCD) out of twelve quantitative biomarkers were significant. The orientation biomarker which is a weighted 22 QB\630666.01598\97090978.4Mayo 2024-087 630666.01598 matrix based on all twelve biomarkers was also found out to be statistically significant in differentiating reactive and metastatic ALNs.

[0098] In another example study, the orientation features and orientation score described in the present disclosure were analyzed as diagnostic features for discriminating hepatocellular carcinoma (HCC) and / or malignant liver lesions from benign cases. FIG. 16 illustrates an example description for computing orientation and extended orientation biomarkers according to the methods described in the present disclosure.

[0099] Biomarkers computed using the techniques described in the present disclosure may include instances of not a number (NaN) and infinity (Inf). NaN values arise in situations of structural missingness, for example, when no microvessel segments are present, resulting in the absence of NPV and NCV values. For all density and diameter biomarkers, NaN values can be replaced with zeros, as this appropriately represents the lack of measurable density and diameter in such cases. When kurtosis yields NaN values, indicating it is undefined, these cases can be excluded from the univariable analysis. Infinite values may appear in metrics like PCD and diameter biomarkers due to the absence of circumferential microvessel segments. In these instances, the Inf values are replaced with the maximum value in that row, yielding a finite value that still signals a strong penetration vessel.

[0100] In this study, a total of 22 subjects (10 male and 12 female) with 29 suspected hepatic masses (during screening ultrasound) were imaged. Out of 29 lesions, 10 cases were benign and 19 were malignant. Out of the 19 malignant cases, 11 were HCC, 7 were cholangiocarcinoma (CCA), and 1 was metastatic adenocarcinoma. The label of the lesion was based on clinical diagnosis through pathology or cross-sectional imaging. The study was done based size of the lesion, so the entire dataset was divided into three categories: 1) size less than 30 mm, 2) size less than 50 mm, and 3) no constraint.

[0101] The ultrasound scanning for each subject was performed using an ultrasound system equipped with a curved array transducer C1-6 at 3.6 MHz frequency. The system was used to capture raw in-phase and quadrature data at a very high frame rate for the duration of 3 s. Each frame of data is formed using 5-angle coherent plane-wave compounding. To reduce the breathing / motion artifact during acquisition the subject was instructed to halt breathing during the whole 3 s acquisition period. The acquired frames were then further used to generate HDMI. This process involved operations like clutter filtering using singular value decomposition followed by series filtering operations to reduce background noise and enhance 23 QB\630666.01598\97090978.4Mayo 2024-087 630666.01598 vessels. A binarization and skeletonization operation was then performed to obtain a binarized and skeletonized version of HDMI which was used for quantification analysis.

[0102] The illustration of biomarker values and orientation score value of microvessel lesions from the in vivo study are shown in FIGS. 17A–17C and 18A–18B. Results from five subjects are illustrated: two HCCs (HCC1 and HCC2), one CCA, and two benign cases (benign1 and benign2). FIGS. 17A–17C show the biomarker values of three malignant cases, and FIGS.18A–18B show the biomarker values for two benign cases. For each subject, seven images have been shown. Image 1 is a B-mode image with two outlines marked in blue and red. The blue curve is a lesion boundary marked under the supervision of the radiologist, and the red curve is a 10 mm dilation mask to include perilesional vascularity. Images 2, 3, and 4 are HDMI images, binary images of HDMI, and binary polar HDMI images, respectively. Image 5 shows the microvessel segment in contact with the ROI boundaries and the corresponding angle it makes with the boundaries. Images 6 and 7 illustrate a large microvessel segment which are classified as circumferential and penetrating using PCD algorithms. The light brown region is ROI, and the dark brown is the central region of ROI (0.6 scaled version of ROI). The values of each biomarker, the criteria for penetrating pattern, and the orientation score for each case are shown below images.

[0103] The box plot of all biomarkers for two different cases: (1) Malignant liver tumors from benign liver lesions and (2) HCC from benign tumors, is shown in FIGS. 19A– 19M and FIGS.20A–20M, respectively. For benign-malignant, 7 biomarkers were significant (MRH, MTH, NPV, NCV, WPD, diameter APD, and diameter PCD), and for benign-HCC cases, 6 biomarkers were statistically significant (MTH, NPV, NCV, WPD, diameter APD, and diameter PCD). The orientation score was also found to be statistically significant for both cases.

[0104] FIG.21 shows an example of a system 2100 for generating and analyzing vessel orientation feature data in accordance with some embodiments described in the present disclosure. As shown in FIG. 21, a computing device 2150 can receive one or more types of data (e.g., image data, blood vessel image data) from data source 2102. In some embodiments, computing device 2150 can execute at least a portion of a vessel orientation feature generation and analysis system 2104 to generate and analyze vessel orientation feature data from data received from the data source 2102.

[0105] Additionally or alternatively, in some embodiments, the computing device 2150 can communicate information about data received from the data source 2102 to a server 2152 24 QB\630666.01598\97090978.4Mayo 2024-087 630666.01598 over a communication network 2154, which can execute at least a portion of the vessel orientation feature generation and analysis system 2104. In such embodiments, the server 2152 can return information to the computing device 2150 (and / or any other suitable computing device) indicative of an output of the vessel orientation feature generation and analysis system 2104.

[0106] In some embodiments, computing device 2150 and / or server 2152 can be any suitable computing device or combination of devices, such as a desktop computer, a laptop computer, a smartphone, a tablet computer, a wearable computer, a server computer, a virtual machine being executed by a physical computing device, and so on. The computing device 2150 and / or server 2152 can also reconstruct images from the data.

[0107] In some embodiments, data source 2102 can be any suitable source of data (e.g., measurement data, images reconstructed from measurement data, processed image data), such as an imaging system, another computing device (e.g., a server storing measurement data, images reconstructed from measurement data, processed image data), and so on. In some embodiments, data source 2102 can be local to computing device 2150. For example, data source 2102 can be incorporated with computing device 2150 (e.g., computing device 2150 can be configured as part of a device for measuring, recording, estimating, acquiring, or otherwise collecting or storing data). As another example, data source 2102 can be connected to computing device 2150 by a cable, a direct wireless link, and so on. Additionally or alternatively, in some embodiments, data source 2102 can be located locally and / or remotely from computing device 2150, and can communicate data to computing device 2150 (and / or server 2152) via a communication network (e.g., communication network 2154).

[0108] In some embodiments, communication network 2154 can be any suitable communication network or combination of communication networks. For example, communication network 2154 can include a Wi-Fi network (which can include one or more wireless routers, one or more switches, etc.), a peer-to-peer network (e.g., a Bluetooth network), a cellular network (e.g., a 3G network, a 4G network, etc., complying with any suitable standard, such as CDMA, GSM, LTE, LTE Advanced, WiMAX, etc.), other types of wireless network, a wired network, and so on. In some embodiments, communication network 2154 can be a local area network, a wide area network, a public network (e.g., the Internet), a private or semi-private network (e.g., a corporate or university intranet), any other suitable type of network, or any suitable combination of networks. Communications links shown in FIG.21 25 QB\630666.01598\97090978.4Mayo 2024-087 630666.01598 can each be any suitable communications link or combination of communications links, such as wired links, fiber optic links, Wi-Fi links, Bluetooth links, cellular links, and so on.

[0109] Referring now to FIG. 22, an example of hardware 2200 that can be used to implement data source 2102, computing device 2150, and server 2152 in accordance with some embodiments of the systems and methods described in the present disclosure is shown.

[0110] As shown in FIG.22, in some embodiments, computing device 2150 can include a processor 2202, a display 2204, one or more inputs 2206, one or more communication systems 2208, and / or memory 2210. In some embodiments, processor 2202 can be any suitable hardware processor or combination of processors, such as a central processing unit (“CPU”), a graphics processing unit (“GPU”), and so on. In some embodiments, display 2204 can include any suitable display devices, such as a liquid crystal display (“LCD”) screen, a light-emitting diode (“LED”) display, an organic LED (“OLED”) display, an electrophoretic display (e.g., an “e-ink” display), a computer monitor, a touchscreen, a television, and so on. In some embodiments, inputs 2206 can include any suitable input devices and / or sensors that can be used to receive user input, such as a keyboard, a mouse, a touchscreen, a microphone, and so on.

[0111] In some embodiments, communications systems 2208 can include any suitable hardware, firmware, and / or software for communicating information over communication network 2154 and / or any other suitable communication networks. For example, communications systems 2208 can include one or more transceivers, one or more communication chips and / or chip sets, and so on. In a more particular example, communications systems 2208 can include hardware, firmware, and / or software that can be used to establish a Wi-Fi connection, a Bluetooth connection, a cellular connection, an Ethernet connection, and so on.

[0112] In some embodiments, memory 2210 can include any suitable storage device or devices that can be used to store instructions, values, data, or the like, that can be used, for example, by processor 2202 to present content using display 2204, to communicate with server 2152 via communications system(s) 2208, and so on. Memory 2210 can include any suitable volatile memory, non-volatile memory, storage, or any suitable combination thereof. For example, memory 2210 can include random-access memory (“RAM”), read-only memory (“ROM”), electrically programmable ROM (“EPROM”), electrically erasable ROM (“EEPROM”), other forms of volatile memory, other forms of non-volatile memory, one or more forms of semi-volatile memory, one or more flash drives, one or more hard disks, one or 26 QB\630666.01598\97090978.4Mayo 2024-087 630666.01598 more solid state drives, one or more optical drives, and so on. In some embodiments, memory 2210 can have encoded thereon, or otherwise stored therein, a computer program for controlling operation of computing device 2150. In such embodiments, processor 2202 can execute at least a portion of the computer program to present content (e.g., images, user interfaces, graphics, tables), receive content from server 2152, transmit information to server 2152, and so on. For example, the processor 2202 and the memory 2210 can be configured to perform the methods described herein (e.g., the method of FIG.7).

[0113] In some embodiments, server 2152 can include a processor 2212, a display 2214, one or more inputs 2216, one or more communications systems 2218, and / or memory 2220. In some embodiments, processor 2212 can be any suitable hardware processor or combination of processors, such as a CPU, a GPU, and so on. In some embodiments, display 2214 can include any suitable display devices, such as an LCD screen, LED display, OLED display, electrophoretic display, a computer monitor, a touchscreen, a television, and so on. In some embodiments, inputs 2216 can include any suitable input devices and / or sensors that can be used to receive user input, such as a keyboard, a mouse, a touchscreen, a microphone, and so on.

[0114] In some embodiments, communications systems 2218 can include any suitable hardware, firmware, and / or software for communicating information over communication network 2154 and / or any other suitable communication networks. For example, communications systems 2218 can include one or more transceivers, one or more communication chips and / or chip sets, and so on. In a more particular example, communications systems 2218 can include hardware, firmware, and / or software that can be used to establish a Wi-Fi connection, a Bluetooth connection, a cellular connection, an Ethernet connection, and so on.

[0115] In some embodiments, memory 2220 can include any suitable storage device or devices that can be used to store instructions, values, data, or the like, that can be used, for example, by processor 2212 to present content using display 2214, to communicate with one or more computing devices 2150, and so on. Memory 2220 can include any suitable volatile memory, non-volatile memory, storage, or any suitable combination thereof. For example, memory 2220 can include RAM, ROM, EPROM, EEPROM, other types of volatile memory, other types of non-volatile memory, one or more types of semi-volatile memory, one or more flash drives, one or more hard disks, one or more solid state drives, one or more optical drives, and so on. In some embodiments, memory 2220 can have encoded thereon a server program 27 QB\630666.01598\97090978.4Mayo 2024-087 630666.01598 for controlling operation of server 2152. In such embodiments, processor 2212 can execute at least a portion of the server program to transmit information and / or content (e.g., data, images, a user interface) to one or more computing devices 2150, receive information and / or content from one or more computing devices 2150, receive instructions from one or more devices (e.g., a personal computer, a laptop computer, a tablet computer, a smartphone), and so on.

[0116] In some embodiments, the server 2152 is configured to perform the methods described in the present disclosure. For example, the processor 2212 and memory 2220 can be configured to perform the methods described herein (e.g., the method of FIG.7).

[0117] In some embodiments, data source 2102 can include a processor 2222, one or more data acquisition systems 2224, one or more communications systems 2226, and / or memory 2228. In some embodiments, processor 2222 can be any suitable hardware processor or combination of processors, such as a CPU, a GPU, and so on. In some embodiments, the one or more data acquisition systems 2224 are generally configured to acquire data, images, or both, and can include an imaging system, such as an ultrasound system, an OCT system, an MRI system, a CT system, or the like. Additionally or alternatively, in some embodiments, the one or more data acquisition systems 2224 can include any suitable hardware, firmware, and / or software for coupling to and / or controlling operations of an imaging system. In some embodiments, one or more portions of the data acquisition system(s) 2224 can be removable and / or replaceable.

[0118] Note that, although not shown, data source 2102 can include any suitable inputs and / or outputs. For example, data source 2102 can include input devices and / or sensors that can be used to receive user input, such as a keyboard, a mouse, a touchscreen, a microphone, a trackpad, a trackball, and so on. As another example, data source 2102 can include any suitable display devices, such as an LCD screen, an LED display, an OLED display, an electrophoretic display, a computer monitor, a touchscreen, a television, etc., one or more speakers, and so on.

[0119] In some embodiments, communications systems 2226 can include any suitable hardware, firmware, and / or software for communicating information to computing device 2150 (and, in some embodiments, over communication network 2154 and / or any other suitable communication networks). For example, communications systems 2226 can include one or more transceivers, one or more communication chips and / or chip sets, and so on. In a more particular example, communications systems 2226 can include hardware, firmware, and / or software that can be used to establish a wired connection using any suitable port and / or 28 QB\630666.01598\97090978.4Mayo 2024-087 630666.01598 communication standard (e.g., VGA, DVI video, USB, RS-232, etc.), Wi-Fi connection, a Bluetooth connection, a cellular connection, an Ethernet connection, and so on.

[0120] In some embodiments, memory 2228 can include any suitable storage device or devices that can be used to store instructions, values, data, or the like, that can be used, for example, by processor 2222 to control the one or more data acquisition systems 2224, and / or receive data from the one or more data acquisition systems 2224; to generate images from data; present content (e.g., data, images, a user interface) using a display; communicate with one or more computing devices 2150; and so on. Memory 2228 can include any suitable volatile memory, non-volatile memory, storage, or any suitable combination thereof. For example, memory 2228 can include RAM, ROM, EPROM, EEPROM, other types of volatile memory, other types of non-volatile memory, one or more types of semi-volatile memory, one or more flash drives, one or more hard disks, one or more solid state drives, one or more optical drives, and so on. In some embodiments, memory 2228 can have encoded thereon, or otherwise stored therein, a program for controlling operation of data source 2102. In such embodiments, processor 2222 can execute at least a portion of the program to generate images, transmit information and / or content (e.g., data, images, a user interface) to one or more computing devices 2150, receive information and / or content from one or more computing devices 2150, receive instructions from one or more devices (e.g., a personal computer, a laptop computer, a tablet computer, a smartphone, etc.), and so on.

[0121] In some embodiments, any suitable computer-readable media can be used for storing instructions for performing the functions and / or processes described herein. For example, in some embodiments, computer-readable media can be transitory or non-transitory. For example, non-transitory computer-readable media can include media such as magnetic media (e.g., hard disks, floppy disks), optical media (e.g., compact discs, digital video discs, Blu-ray discs), semiconductor media (e.g., RAM, flash memory, EPROM, EEPROM), any suitable media that is not fleeting or devoid of any semblance of permanence during transmission, and / or any suitable tangible media. As another example, transitory computer- readable media can include signals on networks, in wires, conductors, optical fibers, circuits, or any suitable media that is fleeting and devoid of any semblance of permanence during transmission, and / or any suitable intangible media.

[0122] As used herein in the context of computer implementation, unless otherwise specified or limited, the terms “component,” “system,” “module,” “framework,” and the like are intended to encompass part or all of computer-related systems that include hardware, 29 QB\630666.01598\97090978.4Mayo 2024-087 630666.01598 software, a combination of hardware and software, or software in execution. For example, a component may be, but is not limited to being, a processor device, a process being executed (or executable) by a processor device, an object, an executable, a thread of execution, a computer program, or a computer. By way of illustration, both an application running on a computer and the computer can be a component. One or more components (or system, module, and so on) may reside within a process or thread of execution, may be localized on one computer, may be distributed between two or more computers or other processor devices, or may be included within another component (or system, module, and so on).

[0123] In some implementations, devices or systems disclosed herein can be utilized or installed using methods embodying aspects of the disclosure. Correspondingly, description herein of particular features, capabilities, or intended purposes of a device or system is generally intended to inherently include disclosure of a method of using such features for the intended purposes, a method of implementing such capabilities, and a method of installing disclosed (or otherwise known) components to support these purposes or capabilities. Similarly, unless otherwise indicated or limited, discussion herein of any method of manufacturing or using a particular device or system, including installing the device or system, is intended to inherently include disclosure, as embodiments of the disclosure, of the utilized features and implemented capabilities of such device or system.

[0124] FIG.23 illustrates an example of an ultrasound system 2300 that can implement the methods described in the present disclosure. The ultrasound system 2300 includes a transducer array 2302 that includes a plurality of separately driven transducer elements 2304. The transducer array 2302 can include any suitable ultrasound transducer array, including linear arrays, curved arrays, phased arrays, and so on. Similarly, the transducer array 2302 can include a 1D transducer, a 1.5D transducer, a 1.75D transducer, a 2D transducer, a 3D transducer, and so on.

[0125] When energized by a transmitter 2306, a given transducer element 2304 produces a burst of ultrasonic energy. The ultrasonic energy reflected back to the transducer array 2302 (e.g., an echo) from the object or subject under study is converted to an electrical signal (e.g., an echo signal) by each transducer element 2304 and can be applied separately to a receiver 2308 through a set of switches 2310. The transmitter 2306, receiver 2308, and switches 2310 are operated under the control of a controller 2312, which may include one or more processors. As one example, the controller 2312 can include a computer system. 30 QB\630666.01598\97090978.4Mayo 2024-087 630666.01598

[0126] The transmitter 2306 can be programmed to transmit unfocused or focused ultrasound waves. In some configurations, the transmitter 2306 can also be programmed to transmit diverged waves, spherical waves, cylindrical waves, plane waves, or combinations thereof. Furthermore, the transmitter 2306 can be programmed to transmit spatially or temporally encoded pulses.

[0127] The receiver 2308 can be programmed to implement a suitable detection sequence for the imaging task at hand. In some embodiments, the detection sequence can include one or more of line-by-line scanning, compounding plane wave imaging, synthetic aperture imaging, and compounding diverging beam imaging.

[0128] In some configurations, the transmitter 2306 and the receiver 2308 can be programmed to implement a high frame rate. For instance, a frame rate associated with an acquisition pulse repetition frequency (“PRF”) of at least 100 Hz can be implemented. In some configurations, the ultrasound system 2300 can sample and store at least one hundred ensembles of echo signals in the temporal direction.

[0129] A scan can be performed by setting the switches 2310 to their transmit position, thereby directing the transmitter 2306 to be turned on momentarily to energize transducer elements 2304 during a single transmission event according to a selected imaging sequence. The switches 2310 can then be set to their receive position and the subsequent echo signals produced by the transducer elements 2304 in response to one or more detected echoes are measured and applied to the receiver 2308. The separate echo signals from the transducer elements 2304 can be combined in the receiver 2308 to produce a single echo signal.

[0130] The echo signals are communicated to a processing unit 2314, which may be implemented by a hardware processor and memory, to process echo signals or images generated from echo signals. As an example, the processing unit 2314 can generate vessel orientation feature data using the methods described in the present disclosure. Images produced from the echo signals, vessel orientation feature data, and other relevant data generated by the processing unit 2314 can be displayed on a display system 2316.

[0131] The present disclosure has described one or more preferred embodiments, and it should be appreciated that many equivalents, alternatives, variations, and modifications, aside from those expressly stated, are possible and within the scope of the invention. 31 QB\630666.01598\97090978.4

Claims

Mayo 2024-087 630666.01598 CLAIMS 1. A method for generating vessel orientation feature data from blood vessel image data acquired from a subject, the steps of the method comprising: (a) providing a blood vessel image to a computer system, the blood vessel image having been acquired with an imaging system from a subject; (b) generating a region-of-interest (ROI) image from the blood vessel image using the computer system by applying an ROI mask to the blood vessel image, wherein the ROI mask corresponds to an ROI in the blood vessel image; (c) converting the ROI image to a binary image using the computer system, wherein the binary image contains pixels with first binary values and pixels with second binary values, wherein pixels with the first binary value correspond to segmented vessels; (d) generating vessel orientation feature data from the binary image using the computer system, wherein the vessel orientation feature data indicate penetrating and circumferential blood vessels in the ROI based on quantifying orientations of blood vessels in the ROI; and (e) outputting the vessel orientation feature data using the computer system.

2. The method of claim 1, wherein the vessel orientation feature data comprise polar coordinate orientation features calculated in a polar coordinate system.

3. The method of claim 2, further comprising converting the binary image to a polar coordinate system and generating the vessel orientation feature data from the converted binary image.

4. The method of claim 2, wherein the vessel orientation feature data comprise histogram features generated from a histogram of vessel orientations generated from the binary image.

5. The method of claim 4, wherein the histogram features comprise a histogram mean. 32 QB\630666.01598\97090978.4Mayo 2024-087 630666.01598 6. The method of claim 5, wherein the histogram mean comprises a radial histogram mean of blood vessels oriented along a radial direction.

7. The method of claim 5 or 6, wherein the histogram mean comprises a tangential mean of blood vessels orientated along a tangential direction.

8. The method of claim 4, wherein the histogram features comprise a kurtosis.

9. The method of claim 8, wherein the kurtosis comprises a radial kurtosis indicating a tailness in a distribution of blood vessels oriented along a radial direction.

10. The method of claim 8 or 9, wherein the kurtosis comprises a tangential kurtosis indicating a tailness in a distribution of blood vessels oriented along a tangential direction.

11. The method of claim 4, wherein the histogram features comprise at least one of a first-order histogram feature, a second-order histogram feature, or a higher order moment.

12. The method of claim 11, wherein the histogram features comprise at least one of standard deviation, correlation, entropy, or skewness.

13. The method of claim 2, wherein the vessel orientation feature data comprise a polar gradient angle.

14. The method of claim 13, wherein the polar gradient angle is computed by: computing a radial gradient along a radial direction of the binary image; computing an angular gradient along an angular direction of the binary image; and computing the polar gradient angle as a weighted mean of an inverse tangent of a ratio between the radial gradient and the angular gradient for all blood vessel segments in the binary image. 33 QB\630666.01598\97090978.4Mayo 2024-087 630666.01598 15. The method of claim 14, wherein the radial gradient and the polar gradient are computed using a gradient operator.

16. The method of claim 15, wherein the gradient operator comprises one of a Sobel operator, a Robert operator, or a Prewitt operator.

17. The method of claim 14, wherein the blood vessel segments in the binary image are identified using a connected components analysis of the binary image to determine groups of adjacent pixels associated with the blood vessel segments.

18. The method of claim 14, wherein the weighted mean is computed based on an area of the blood vessel segments.

19. The method of claim 14, wherein the weighted mean is computed based on a length of the blood vessel segments.

20. The method of claim 1, wherein the vessel orientation feature data comprise Cartesian coordinate orientation features calculated in a Cartesian coordinate system.

21. The method of claim 20, wherein the vessel orientation feature data comprise an angle-based penetration density (APD) feature.

22. The method of claim 21, wherein the APD feature is calculated by determining an angle between each blood vessel contacting a boundary of the ROI, counting a blood vessel as a penetrating blood vessel when the angle is between a first limit and a second limit, and calculating the APD feature as a ratio of penetrating blood vessel area to total blood vessel area in the ROI.

23. The method of claim 22, wherein the first limit is 30 degrees and the second limit is 150 degrees.

24. The method of claim 22, wherein determining the angle between each blood vessel contacting the boundary of the ROI comprises estimating a blood vessel line based on 34 QB\630666.01598\97090978.4Mayo 2024-087 630666.01598 the binary image of the blood vessel and a boundary line at a boundary point of the ROI contacted by the blood vessel.

25. The method of claim 20, wherein the vessel orientation feature data comprise a movement-based penetration to circumferential density (PCD) feature.

26. The method of claim 25, wherein the PCD feature is calculated as a ratio of penetrating blood vessel area to circumferential blood vessel area in the ROI.

27. The method of claim 26, wherein a blood vessel is identified as a penetrating blood vessel by computing a penetration factor and determining whether the penetration factor is greater than a penetration level.

28. The method of claim 27, wherein the penetration level is 0.

7.

29. The method of claim 27, wherein the penetration factor is computed as a ratio between a penetration distance and a skeleton length for the blood vessel.

30. The method of claim 29, wherein the penetration distance is computed as a difference between a distance from the blood vessel start point and a center of the ROI and a distance from the blood vessel end point and the center of the ROI.

31. The method of claim 27, wherein the blood vessel is identified as the penetrating blood vessel when it is also contained with a central region of the ROI.

32. The method of claim 31, wherein the central region of the ROI is determining by scaling the ROI by a factor that is less than 1.

33. The method of claim 20, wherein the vessel orientation feature data comprise an angle-based penetration density (APD) feature and a movement-based penetration to circumferential density (PCD) feature. 35 QB\630666.01598\97090978.4Mayo 2024-087 630666.01598 34. The method of claim 33, wherein the vessel orientation feature data further comprise at least one of number of penetrating vessels (NPV), number of circumferential vessels (NCV), weighted penetration density (WPD), diameter APD, and diameter PCD.

35. The method of claim 34, wherein NPV is calculated as a maximum number of vessel counts identified as penetrating using the APD feature and the PCD feature.

36. The method of claim 35, wherein NCV is calculated by subtracting the NPV from a total number of vessel segments.

37. The method of claim 34, wherein WPD is computed as a weighted sum of an area of penetrating vessel segments divided by a total vessel area.

38. The method of claim 37, wherein a higher weight is applied to vessel segments identified as penetrating based on both the APD feature and the PCD feature, as compared to vessel segments not identified as penetrating based on both the APD feature and the PCD feature.

39. The method of claim 34, wherein diameter APD is calculated as a ratio of a maximum diameter of penetrating microvessels to a maximum diameter of circumferential microvessels using the APD.

40. The method of claim 34, wherein diameter PCD is calculated as a ratio of a maximum diameter of penetrating microvessels to a maximum diameter of circumferential microvessels using the PCD.

41. The method of claim 39 or 40, wherein the diameter of microvessel segments is determined by applying a Euclidean distance transform to a binary vessel image and doubling a maximum radial distance from a centerline to a boundary of the microvessel segments. 36 QB\630666.01598\97090978.4Mayo 2024-087 630666.01598 42. The method of claim 1, further comprising calculating an orientation score by applying a set of predefined orientation threshold criteria to orientation features contained in the vessel orientation feature data.

43. The method of claim 42, further comprising converting results of the threshold criteria to a multi-element binary vector.

44. The method of claim 43, wherein the orientation score is computed by summing elements in the multi-element binary vector, wherein higher values of the orientation score indicate a more penetrating vessel orientation pattern and lower values of the orientation score indicate a more circumferential vessel orientation pattern.

45. The method of claim 1, wherein the imaging system is an ultrasound system.

46. The method of claim 1, wherein the imaging system is an optical coherence tomography (OCT) system.

47. The method of claim 1, wherein the imaging system is a magnetic resonance imaging (MRI) system.

48. The method of claim 1, wherein the imaging system is a computed tomography (CT) system.

49. The method of claim 1, wherein the vessel orientation feature data are analyzed using the computer system to classify a disease type in a body organ.

50. The method of claim 49, wherein the vessel orientation feature data are analyzed to differentiate between malignant and benign lesions.

51. The method of claim 1, wherein the vessel orientation feature data are analyzed using the computer system to stage a disease progression in a tissue from which the blood vessel image data was acquired. 37 QB\630666.01598\97090978.4Mayo 2024-087 630666.01598 52. The method of claim 1, wherein the vessel orientation feature data are analyzed using the computer system to monitor a treatment response to a treatment provided to a tissue from which the blood vessel image data was acquired. 38 QB\630666.01598\97090978.4

Images