Ultrasonic analysis method and equipment

Abstract

This invention provides an ultrasound data processing method (30) for detecting the presence of intravascular objects in a vascular lumen based on the analysis of acquired intravascular ultrasound data of a lumen. The method includes receiving (32) data comprising multiple frames, each frame containing data for multiple radial lines corresponding to different circumferential positions around the IVUS device body, and reducing (34) the data to a single representative value for each radial line in each frame. These representative values ​​are then processed to derive (36) values ​​representing the probability of object presence within a given frame for at least each frame. Based on the probability values, (38) regions within the data occupied by intravascular objects are determined, for example, a continuous set of frames occupied by objects.

Description

Technical Field

[0001] The present invention relates to an ultrasound processing method and apparatus, in particular, for detecting the presence of an object in intravascular ultrasound data. Background Art

[0002] Intravascular ultrasound (IVUS) imaging is a valuable technique for obtaining internal images of a patient's cardiovascular system, such as the patient's arteries or heart. IVUS images can assist in assessing the condition of the cardiovascular system, such as by detecting and quantifying the size of stenosis, plaque buildup, or assisting in positioning medical implants such as stents.

[0003] To obtain IVUS images, a small non-invasive medical device, such as a catheter or guidewire with an ultrasound probe or set of ultrasound transducers fitted at its tip, is inserted into the patient's cardiovascular system, typically into an artery, after which IVUS images are captured at regular intervals while the small non-invasive medical device is pulled back. In this manner, the captured cross-sectional IVUS images of the cardiovascular system can assist in providing valuable insight into the condition of the cardiovascular system along its imaged length.

[0004] IVUS devices, such as catheters, incorporate one or more ultrasound transducers for imaging the lumen of a blood vessel.

[0005] The transducer transmits and receives ultrasound energy to create an image of the vessel of interest. The ultrasound waves are affected in part by discontinuities arising from tissue structures such as the various layers of the vessel wall, red blood cells, and other features of interest, including such as artificial objects implanted in the vessel.

[0006] Echoes from the reflected waves are received by the transducer and passed to an IVUS imaging system, which may be connected to the IVUS catheter, for example, by way of a patient interface module known as a “PIM.” The imaging system processes the received ultrasound signals to produce, for example, cross-sectional images of a blood vessel.

[0007] There are two types of IVUS catheters commonly used today: rotational and solid state.For a typical rotational IVUS catheter, one or more ultrasound transducer elements are located at the tip of a flexible drive shaft that rotates within a sheath inserted into the vessel of interest.

[0008] In contrast, a solid-state IVUS catheter carries an ultrasound scanner assembly including an array of ultrasound transducers, such as a one-dimensional or two-dimensional array, distributed around the periphery of the device and connected to a set of transducer control circuits. The array and its corresponding circuitry are often referred to as the imaging core or scanner of an IVUS device.

[0009] For conventional image generation, transducer control circuitry can select individual or groups of transducers in the array for transmitting ultrasound pulses and / or for receiving echo signals. By sequentially stepping a sequence of transmitter-receiver pairs around the periphery of the device, the IVUS system is able to synthesize the effects of mechanically rotating transducer elements but without moving parts. Since there are no rotating mechanical elements, the transducer array can be placed in direct contact with the blood and vessel assembly with minimal risk of vascular trauma.

[0010] The data captured at each of the serial positions around the periphery of the device is often referred to as a radial scan line.

[0011] A common way to visualize a fully pulled-back (pulling the IVUS device along the longitudinal axis of the vessel lumen) IVUS data set is to display one or more cross-sectional views along the pull-back (longitudinal) axis. A cross-sectional view means data representing a single radial line along the entire longitudinal pull-back distance. This is done in Figure 1 is described in.

[0012] Figure 1 An exemplary IVUS device 12 having an elongated device body 14 is schematically illustrated. The elongated device body extends along a longitudinal axis oriented from Figure 1 The example IVUS device is shown as including an array 16 of ultrasound transducer elements 20 extending circumferentially around the device with the outer sides facing. Figure 1 illustrates the activation of a transducer element to move along a specific azimuth Directionality, i.e., along a specific rotation angle around the device A single radial acoustic signal line 24 is generated. The single radial signal line 24 captures a single radial line of ultrasound data. Multiple radial signal lines together form a complete cross-sectional IVUS image 25.

[0013] exist Figure 1 , the trajectories of two example radial lines a1 and a2 are schematically illustrated. The corresponding trajectories of these two lines within the complete IVUS scan image 25 are also shown.

[0014] Although a single transducer element is shown as generating each radial line, in practice, each radial scan line may be generated using a subset of transducer elements (in order to enable beamforming for focusing acoustic transmission and reflection along a single narrow radial axis). The subset may, for example, include up to 14 transducer elements that are peripherally or azimuthally aligned. Moreover, the array 16 may in some examples include multiple elements along the A-A' longitudinal direction (i.e., have multiple columns).

[0015] By way of example, in one known embodiment, the array of transducer elements is only elements that are wide in the longitudinal direction and are There are 64 elements in the direction. A subset of 14 transducer elements is used to create 4 radial rays. Indexing the active subset in azimuth through the 64 element positions results in 4*64=256 radial rays per IVUS frame. These figures represent only one exemplary embodiment and are not limiting.

[0016] In operation, the device 12 is pulled longitudinally back along the lumen of the blood vessel 28 into which it is inserted. Figure 2 . The device steps peripherally through sequential transducer elements 20 (or sequential azimuthal subsets of transducer elements), each transducer element (or subset) capturing a different corresponding radial line of ultrasound data. Each complete peripheral cycle of radial lines constitutes data for a single frame f. The longitudinal pull of the device 12 means that sequential frames f of data correspond to sequentially advancing longitudinal positions along the lumen of the blood vessel 28, as indicated by Figure 2 The arrows in the figure illustrate this.

[0017] Figure 3 (Top) shows a sequence of reconstructed IVUS frames and cross-sectional views ( Figure 3 , bottom). Each IVUS frame represents a complete radial cross section, along a cross-sectional plane parallel to the radial axis of the IVUS device and perpendicular to the longitudinal axis. Each radial IVUS frame consists of data for a complete set of radial lines around the entire set of azimuth (rotation) angle directions around the IVUS device.

[0018] exist Figure 3 The cross-sectional view shown at the bottom of represents a cross-section along a cross-sectional plane parallel to the longitudinal axis of the IVUS device and also parallel to a particular (perpendicular) radial axis shown in the image.

[0019] The IVUS frames depicted are 100 frames apart. Overlapping arrows indicate the location and direction of the cross section in each case. In the bottom cross-sectional view, the columns corresponding to each frame are represented by the corresponding four arrows. Note that although the second to fourth frames contain stents, the stents are barely visible in the cross-sectional view and are similarly undetectable from this cross-sectional view.

[0020] The challenge with IVUS is that due to the multiple dimensions of data collection, it is difficult to represent or process the data in a way that provides useful or understandable results to the user. Current solutions typically present the data as cross-sectional planes across the vessel or along its length, such as in Figure 3These may represent only a small portion of the vessel wall at a given time. Since only a subset of the volume is visible at a given time, features such as vessel side branches may be missed during analysis. This is important in the case of stent (or other intravascular object) implantation, as the final position of the stent boundary should preferably not overlap with the side branches.

[0021] Furthermore, if one desires to inspect an object that has been implanted, or seeks to identify the proximal or distal edges of an object such as a stent during implantation, this can be difficult when the entire inner circumference of the vessel across its length cannot be analyzed or inspected at once.

[0022] There is therefore a need for improved solutions for IVUS ultrasound data processing. Summary of the invention

[0023] The invention is defined by the claims.

[0024] According to an example of aspects of the present invention, there is provided an ultrasound data analysis method for analyzing ultrasound data captured by an intravascular ultrasound (IVUS) device, the method comprising:

[0025] A plurality of frames of ultrasound data captured by an IVUS device are received, each frame comprising data for a plurality of radial scan lines, each radial scan line corresponding to a different rotation angle ( ) about a longitudinal axis (A-A') of the device. ) direction of the acoustic signal reflection;

[0026] processing ultrasound data for each radial scan line in each frame to reduce the data for each line to a single representative data value for the line;

[0027] deriving a set of probability values ​​corresponding to the probability of the presence of the object within the artificial blood vessel at least in each frame from the set of representative data values;

[0028] A region within the ultrasound data occupied by an intravascular object is determined based on the probability value.

[0029] Embodiments of the present invention allow for easier analysis and processing of data, as well as a more intuitive presentation of data, by aggregating the data for each entire scan line into a single representative value or number, such as the maximum pixel intensity along the scan line or its standard deviation. This allows an entire line of data to be processed as a single point instead, greatly reducing the complexity of processing operations for identifying objects within the data.

[0030] Embodiments of the present invention are based on the insight that for detecting intravascular objects, particularly artificial intravascular objects, an entire radial scan line of data is not required, since the object will typically cause some significant contrast or change in certain features in the data at the local region in which it is located. For example, intensity may be detected as being higher or lower at a particular region occupied by the object.

[0031] In addition to faster and simpler processing of the data to identify objects, embodiments of the present invention also provide the advantage that an overview of the entire pullback intravascular scan can be presented in a single 2D map, or just a small number of 2D maps as one or more sets of representative points, with respect to the azimuth or rotation angle ( ) and the frame number on the other axis. The data presented in this figure may be the most relevant data for detecting the presence of an object.

[0032] In an example, the representative value may include information related to the spread or range of data values ​​in a given line or information representing the spread or range in a given row, such as the maximum or peak data value in the line, the minimum value in the line, or the standard deviation or variance of the data values. The data value may be an intensity.

[0033] In some examples, the presence of an object within a particular scan line or within a particular frame can be detected by looking for one or more contrast or variation properties (single values) of the data in each line rather than processing the entire set of visual data and searching for the object. The representative value can correspond to such contrast or variation properties for each line.

[0034] The contrast or variation attribute may for example include the standard deviation of the intensity values ​​for a single radial line, or for example the maximum intensity point of a line. For example, the maximum intensity value or the standard deviation of the intensity for a line or set of lines (which may be higher than some surrounding or neighboring lines) may indicate (at least with high probability) that an object is present in that line or set of lines. These may indicate boundaries or edges of objects.

[0035] More generally, the representative value may be associated with a representation of the intensity value of the corresponding radial line.

[0036] Embodiments of the present invention may be used to determine areas within ultrasound data occupied by artificial (foreign or "man-made") intravascular objects, such as, by way of non-limiting example, stents, sheaths, and / or guidewires. These tend to generate relatively large reflections that produce characteristic patterns of high intensity at localized areas in the ultrasound data. Areas occupied by objects may be identified in the data based, at least in part, on detecting such patterns within the data.

[0037] Additionally or alternatively, in some examples, embodiments may be used, for example, to determine areas within the data occupied by anatomical structures or features, such as, by way of non-limiting example, vessel bifurcations (where a vessel divides into two branches) and vessel side branches. Additionally or alternatively, in some examples, embodiments may be used, for example, to determine areas within the data occupied by pathology, such as, by way of non-limiting example, areas of plaque, calcification, or thrombus.

[0038] In some examples more than one region may be detected.The method may include detecting at least one region in the data occupied by an intravascular object.

[0039] As mentioned above, during data collection, the IVUS device is pulled through a subject lumen (e.g., a blood vessel lumen) to capture frames at regular intervals. This produces a data set comprising a plurality of frames, each frame corresponding to a different longitudinal position along the lumen. Each frame includes data for each radial line. Thus, the frame number corresponds to the longitudinal position, and the radial line number corresponds to the circumferential position.

[0040] Azimuth means the rotation angle about the longitudinal axis of the IVUS device. It corresponds to the azimuth coordinate in the standard circular coordinate system.

[0041] The method may further include generating a data output representative of the determined area within the ultrasound data occupied by the object within the artificial blood vessel.

[0042] Determining the area occupied by the intravascular object may include identifying a continuous subset of frames of the data occupied by the object, and / or identifying a continuous subset of radial lines occupied by the object. The object will typically occupy a series of frames (due to its longitudinal extension) and / or a series of radial lines (due to its circumferential extension), and thus the area occupied by the object within the data may include a continuous series of frames and / or a continuous series of radial lines.

[0043] The area occupied by the object may be determined based on the probability value in different ways.

[0044] According to some embodiments, determining the region may include detecting edges of the intravascular object. Edges may appear as areas of high contrast in the data, such as rapidly changing intensity values ​​as one enters the region of the object, making them relatively intuitive to detect. In some cases, edges generate characteristic patterns in the data or characteristic changes in patterns that allow them to be detected.

[0045] For example, an edge may be detected based on a step change in detection probability value (eg between two consecutive frames or a subset of frames) or based on a decrease or increase in probability of detecting a predefined slope / gradient.

[0046] The (longitudinal) near and far edges of the object may be detected. These may occupy different frames separated by the area occupied by the object.

[0047] According to at least one set of embodiments, determining the area occupied by the endovascular object may include identifying a contiguous subset of frames for which the derived probability values ​​are all above a predefined threshold.

[0048] In a specific example, the threshold may be, for example, 50%, or approximately 50%.In some examples, the threshold may be set dynamically depending on the context, or, for example, specified by a user.

[0049] This solution is computationally simple, meaning that it can be performed very quickly and requires relatively low processing power and resources. It may for example comprise applying a simple temporal filtering operation, for example with a rectangular filter kernel applied to the time series of probability values.

[0050] A variation of this solution may include identifying the largest contiguous set of frames with probability values ​​above a predefined threshold. This will take into account that there may be more than one region that meets the criteria, and will apply the assumption that the largest such region is the one most likely to be truly occupied by an object. This may increase the robustness and reliability of the method.

[0051] In some examples a minimum length for a contiguous set of frames may be imposed to filter out false positive results caused, for example, by noise.

[0052] A further variation may include applying a dynamic process that identifies the longest continuous set of frames with the highest probability value, i.e. an optimization process between length and probability. This may be an algorithm that applies different weights to the probability value relative to the length of the subset in order to identify the most likely occupied region.

[0053] According to at least one set of embodiments, determining the area within the data occupied by the intravascular object may include testing a plurality of trial areas occupied within the data. Testing may include calculating, for each trial area, a difference between a set of probability values ​​for the trial area and an equivalent set of probability values ​​representing the area precisely occupied by the object. Precise may mean determined.

[0054] For example, a set of probability values ​​for a test region may be compared to a test probability curve or function or waveform representing the region assumed to be occupied by the object. This may include a simple square signal with a probability height of 1, and a width or extension corresponding to the extent of the test region.

[0055] This solution is more robust and may produce more reliable results than the simpler thresholding solutions listed above.

[0056] Testing may include evaluating a cost function for each trial region.

[0057] Exact occupancy may include a probability value of 1 over the entire area.

[0058] In an example, the test may include calculating a cost function including, as a first additional term, the sum of the above-mentioned set of difference values ​​for the entire set of frames throughout the trial area.

[0059] The cost function may include as a further additional term the summation of probability values ​​for consecutive sets of frames on one or both sides of the trial area.

[0060] According to any embodiment of the present invention, the representative value may represent the maximum intensity value for each radial line, i.e., the representative value for each line represents the maximum intensity value for that line. They may thus correspond to the values ​​of the maximum intensity points along that line, respectively. This process produces a two-dimensional map of intensity values ​​as a function of the radial scan line index or frame index, and is sometimes referred to as a maximum intensity projection.

[0061] The maximum intensity value is ideally suited for the purpose of object presence detection due to the fact that objects tend to generate characteristic high intensity areas over the area of ​​the data where they are located. Thus, the maximum intensity value of each line essentially corresponds to a rough indicator of the presence or absence of an object. If the maximum intensity is, for example, above a certain characteristic threshold, this may be an indication that an object is present at a certain position in the line.

[0062] The method may also include analyzing the set of maximum intensity values ​​for the entire set of lines or the entire set of frames to detect the location where the object exists. This may, for example, include identifying points where a high value drops to a lower value. In some examples, these points can be taken to represent the location of the edge of the object.

[0063] According to certain advantageous examples, the method may further include determining a set of index values ​​representing the position of each identified maximum intensity value along each radial line. Due to the orientation of the radial line projections, this may correspond to the longitudinal position of the maximum intensity point. This allows for a more accurate identification of the longitudinal position of the object along the vessel lumen.

[0064] The maximum intensity represents only one advantageous example. Other example representative values ​​may include, by way of non-limiting example, the standard deviation of the intensity values ​​in each line, an index representing the location along the line where the maximum intensity point is found, an index representing the location along the line where the first pixel exceeding a predefined threshold is found, or any other suitable example. Further examples will be discussed in subsequent sections of this disclosure.

[0065] According to one or more embodiments, the method may include generating a curve or graph of representative values, such as a curve or graph of representative values ​​with respect to frame number and / or radial line number. This curve or graph may be transmitted as a control output to an associated display device for displaying the curve or graph to a user, for example. The curve or graph may show representative values ​​plotted with respect to at least radial line number along one axis.

[0066] According to one or more examples, deriving the probability value may include detecting one or more characteristic patterns within a set of representative values, characteristic of the presence of the object.

[0067] For example, the method may comprise detecting a pattern in a frame-by-frame sequence of probability values, meaning a sequence of probability values ​​over consecutive self-increasing frames.

[0068] For example, the method may include detecting a change in pattern or value in the spread or distribution according to incrementing frame numbers and / or radial line numbers.

[0069] The characteristic pattern to be detected may be predetermined, for example pre-stored in advance.

[0070] In some examples, a characteristic pattern may include consistent areas of high intensity values ​​or high maximum intensity values.

[0071] In some examples, the characteristic pattern may be a pattern representing an edge of an object, such as a rapid (eg, step) change in intensity value or maximum intensity value. In some instances, the characteristic pattern may include a trough (up and down or undulating) pattern in the representative value.

[0072] According to one or more examples, deriving the probability value can include using a classifier algorithm.

[0073] The classifier algorithm may include a machine learning algorithm. The machine learning algorithm may be trained using labeled or annotated training data including multiple example sets of probability values, each labeled to indicate the presence or absence of an object. Using this training data, the algorithm may learn to distinguish, i.e., classify, a set of probability values ​​indicating the presence of an object and a set of values ​​indicating the absence of an object.

[0074] In some examples, the algorithm may be a neural network.

[0075] In other examples, the algorithm may include a random forest (RF) algorithm. In this case, deriving the probability value may further include using a wavelet transform to perform the initial transformation of the aforementioned representative value, and the RF algorithm uses wavelet coefficients to perform the generation of a probability value estimate describing the possible existence of an object in an associated frame or frame range.

[0076] According to any embodiment of the present invention, each frame may correspond to a different longitudinal position along the lumen. This results from the way IVUS data is collected by pulling the IVUS device longitudinally through the lumen of the vessel, capturing consecutive frames at regular intervals along the path.

[0077] An example according to another aspect of the present invention provides an ultrasound data processor for analyzing ultrasound data captured by an IVUS device, the processor being adapted to:

[0078] receiving a plurality of frames of ultrasound data captured by the IVUS device, each frame comprising data for a plurality of radial scan lines, each radial scan line corresponding to a different rotation angle ( ) about a longitudinal axis (A-A') of the device; )'s acoustic reflection;

[0079] processing the ultrasound data for each radial scan line in each frame to reduce the data for each line to a single representative data value for the line;

[0080] deriving a set of probability values ​​corresponding to the probability of the presence of the object within the artificial blood vessel at least in each frame from the set of representative data values;

[0081] A region within the ultrasound data occupied by an intravascular object is determined based on the probability value.

[0082] Implementation options and details for each step performed by the above processor device may be understood and explained based on the explanation and description provided above regarding the method aspects of the present invention.

[0083] Any examples, options or embodiment features or details described above with respect to the data analysis method may be applied or combined or incorporated into the present device (processor) aspect of the present invention after appropriate modifications.

[0084] According to an example of another aspect of the present invention, there is provided an ultrasound system, comprising:

[0085] An intravascular ultrasound (IVUS) device for capturing ultrasound data within a blood vessel lumen along a plurality of different radial scan lines; and

[0086] An ultrasound data processor according to any example or embodiment listed above or described below or according to any claim of the present application, which is operably coupled to the intravascular ultrasound device and is configured to receive ultrasound data captured by the IVUS device.

[0087] The IVUS device may, for example, include an ultrasound probe, such as a catheter or a guidewire.

[0088] These and other aspects of the invention will be apparent from and elucidated with reference to the embodiments described hereinafter. BRIEF DESCRIPTION OF THE DRAWINGS

[0089] For a better understanding of the invention, and in order to more clearly show how the invention may be implemented, reference will now be made, by way of example only, to the accompanying drawings, in which:

[0090] Figure 1 An example IVUS device is depicted, and capture of an example radial scan line is illustrated;

[0091] Figure 2 An IVUS device positioned inside the lumen of a blood vessel is shown;

[0092] Figure 3 shows an example longitudinal cross section generated from IVUS data in a known solution;

[0093] Figure 4 An example data analysis method according to one or more embodiments is shown;

[0094] Figure 5 shows an example 2D map of the interior of a blood vessel that can be generated in accordance with one or more embodiments;

[0095] Figure 6 Schematically illustrates the Figure 5 The curve provides a geometric viewpoint inside the blood vessel;

[0096] Figure 7 An example curve of maximum brightness values ​​for an example data set is shown, illustrating the presence of brackets in the data;

[0097] Figure 8 An example workflow of an example method according to one or more embodiments is shown;

[0098] Fig. 9 and Fig.10 shows a 2D graph representing maximum intensity values ​​and maximum intensity index values ​​of radial lines for an example data set for target presence and target absence, respectively;

[0099] Fig.11 An example flow chart illustrating an example method according to one or more embodiments;

[0100] Fig.12 An example workflow is shown for an example process for deriving a set of probability values ​​for the probability of an object being present in each data frame;

[0101] Fig.13 Shown in Fig.12 The shapes of the basis functions of the Haar wavelet applied during the example process;

[0102] Fig.14 Shows the use of Fig.12 An example process of applying a wavelet transform to multiple signal decomposition nodes;

[0103] Fig.15 An example set of probability values ​​for frames of an example data set is shown;

[0104] Fig.16 Shown according to Fig.15 Example time-averaged probability curve derived from probability values ​​of ;

[0105] Fig.17 and Fig.18 It is described that different trial regions occupied by objects within a set of probability values ​​are tested based on applying a test square wave to the different trial regions;

[0106] Figure 19-21 The pre-calculation based on two cost functions is illustrated. Fig.17 and Fig.18 The cost function corresponds to different areas under the probability curve. DETAILED DESCRIPTION

[0107] The present invention will be described with reference to the accompanying drawings.

[0108] It should be understood that while illustrating exemplary embodiments of the apparatus, system, and method, the detailed description and specific examples are merely and not intended to limit the scope of the invention. The advantages of the apparatus, system, and method of the present invention will become more readily understood through the following description, the appended claims, and the accompanying drawings. It should be understood that the drawings are schematic only and are not drawn to scale. The same reference numerals are used throughout the drawings to identify the same or similar parts.

[0109] The present invention provides an ultrasound data processing method for detecting the presence of an intravascular object in a lumen of a blood vessel based on analysis of collected intravascular ultrasound data of the lumen. The method includes receiving data including a plurality of frames, and each frame contains data for a plurality of radial lines, corresponding to different circumferential positions around the body of an IVUS device, and reducing the data to a single representative value for each radial line in each frame. These representative values ​​are then processed to derive a value representing the probability of the presence of an object in a given frame for at least each frame. Based on the probability value, an area within the data occupied by the intravascular object is determined, such as a set of consecutive frames occupied by the object.

[0110] Thus, the method provides for the detection of veins or arteries or other body lumens, and uses a novel solution whereby the data for each line is aggregated into a single value to allow for faster and easier processing. It may also allow for a representation of the captured data for the user. The solution is based on the view that for the detection of an object, a single representative value for each line provides sufficient information due to the fact that objects clearly appear in ultrasound data as areas that contrast with surrounding areas (e.g., in terms of intensity values). Thus, a single value can capture the extent of the variation or spread of data values ​​in a radial line, or the point of maximum intensity for a line, which itself combined with similar values ​​for other lines in the frame can provide information sufficient to detect whether an object is present in that frame of data.

[0111] Thus, embodiments of the present invention are based on the use of a different representation or processing of the data, which provides an overview of the intravascular scan in all directions.

[0112] Embodiments of the present invention propose applying this simplified data processing for detecting and / or classifying intravascular objects, in particular artificial intravascular objects, which tend to generate large reflections in the data, allowing identification, for example, by searching for certain contrasting properties in the data.

[0113] Using this processing of data to detect intravascular objects, anatomical structures, or pathologies is not known in the art. Examples of objects within artificial blood vessels include, by way of non-limiting examples, stents, sheaths, and guidewires. Examples of anatomical structures include, by way of non-limiting examples, bifurcations. Examples of pathologies include, by way of non-limiting examples, plaques, calcifications, and thrombi.

[0114] Figure 4 An exemplary ultrasound data analysis method 30 according to one or more embodiments of the present invention is shown in block diagram form. The method 30 is used to analyze ultrasound data captured by an intravascular ultrasound (IVUS) device.

[0115] The method 30 includes receiving 32 a plurality of frames of ultrasound data captured by an IVUS device, wherein each frame includes data for a plurality of radial scan lines. Each radial scan line corresponds to ultrasound signals received along a different radial axis of the device, the different radial axes having a particular rotational (or azimuth) angle about the device. Directivity, i.e., the ability to rotate the device at different angles around the body of the device relative to the longitudinal axis (A-A') of the device. The signal received. Figure 1 This can be seen more clearly in the figure, which will be explained below. The direction of the longitudinal axis is Figure 1 Indicated by line AA'.

[0116] The data may be received, for example, from an IVUS ultrasound device. Alternatively, it may be received from a communicatively coupled data storage device or memory, or in some examples may be received from an external or remote computer or server.

[0117] The method 30 also includes processing 34 the ultrasound data for each radial scan line in each frame to reduce the data for each line to a single representative data value for that line. In some examples more than one representative value may be generated, each representative value representing an entire radial line.

[0118] The method 30 further comprises deriving 36 a set of probability values ​​corresponding to the probability of the presence of the object within the artificial blood vessel at least in each frame from the set of representative data values.

[0119] The method 30 also includes determining 38 a region within the ultrasound data occupied by the intravascular object based on the probability value.

[0120] In some examples, the method may further include generating a data output representing the determined area occupied by the intravascular object, such as for transmission to an external device, such as an external computer, or for transmission to a patient monitoring device or a display device. A control output may be generated for controlling an operatively coupled display device to display the results of the object detection.

[0121] The present invention is used to analyze intravascular ultrasound (IVUS) data, meaning data captured by an IVUS device 12. As discussed above (see, e.g., Figure 1 ), an IVUS device typically has an elongated device body 14 and in operation is pulled longitudinally through a blood vessel, capturing frames at regular intervals. An IVUS device typically includes an array 16 of ultrasound transducer elements 20 located at its outer surface, extending circumferentially around the device. Signals received by one or more transducer elements after transmission of ultrasound signals from the one or more transducer elements are combined to create one of a plurality of radial lines of ultrasound data 24. The radial lines of ultrasound data 24 can be combined to form one of a plurality of frames 25, each frame representing a local spatial slice of the lumen being imaged.

[0122] Each radial line may be achieved by transmitting an ultrasound signal (e.g., an ultrasound beam) along a specific radial axis of the device, with a specific azimuthal direction This means that along a specific rotation angle around the device The reflected signals are then received and provide radial lines of data. Each radial line can be generated by transmitting an ultrasound beam or ray along a corresponding radial axis. Each radial line can use a subset of multiple transducer elements for transmission and / or reception to facilitate, for example, beamforming.

[0123] In operation, the device 12 is pulled longitudinally back along the lumen of the blood vessel 28 into which it is inserted. Figure 2 The device steps through the array of transducer elements 20 along a circle, with each transducer element (or subset of elements) around the circumference capturing a different corresponding radial line of ultrasound data. Each completed peripheral sweep of a radial line constitutes data for a single frame f.

[0124] According to the present invention, the data for each radial scan line 24 is converted to a single representative number, such as the maximum pixel intensity or its standard deviation.

[0125] These values ​​can be plotted in a 2D graph. Figure 5 An example is shown in .

[0126] The y-axis of the graph is the azimuth (or rotation) angle of the corresponding radial line. and the x-axis is the frame number in which the radial line is contained. The frame number effectively corresponds to the longitudinal position along the lumen, due to the way the device is pulled longitudinally along the lumen while capturing consecutive frames. This is thus a graph representing values ​​(in this example, the maximum intensity value) at each circumferential position around the device, at each of a series of longitudinal positions along the lumen.

[0127] Such a map can be useful in itself for presentation to a user such as a clinician. For example, vessel side branches are represented in this map as dark areas (denoted by Figure 5 ). In contrast, calcifications can be detected as characteristic bright areas within the data. Note that it is not required to perform a brief segmentation of the data, as the figure provides an intuitive view of the inside of the vessel, with the vertical axis being the azimuth.

[0128] When represented as a 2D image, the resulting reduced data set effectively provides an unrolled view of the inner wall of the imaged vessel as if cut along its length and placed flush with the inner surface, e.g. Figure 6 Schematically illustrated in .

[0129] For each radial scan line, a representative value of the received ultrasound signal for the entire line is calculated.

[0130] By way of non-limiting example, the representative value may include, for example, any one or more of the following:

[0131] The standard deviation (or variance) of the data values ​​(eg, intensity values) of the radial lines.

[0132] The maximum value (eg, intensity value) of a radial line.

[0133] The index representing the position of the maximum data value along the line, that is, the radial position of the maximum value along the line.

[0134] An index representing a data value or (eg radial) position along the line at which the first pixel has a value exceeding a predefined threshold. The data value is an intensity value.

[0135] An index representing the position (eg, radial position) along the line at which the cumulative sum of the data values ​​reaches a predefined percentage (eg, 75%) of the sum of the data values ​​for the row.

[0136] The output of the matched filter has a filter kernel that matches the characteristic shape of the target object to be detected. The matched filter will thus produce an output signal that adopts the (local) maximum value where the target object is present. In this way, the matched filter can be matched, for example, to the size and / or shape of the support supported in the radial direction.

[0137] At least some embodiments of the present invention are used to identify the presence of an intravascular artificial object in ultrasound data. One example of an intravascular artificial object that can be detected in some examples is an intravascular stent.

[0138] Vascular stents are typically formed from a tubular length of metal mesh, wherein the connecting elements of the mesh extending around its circumference are known as struts.

[0139] The metal stent struts produce strong reflections of the incident ultrasound signal. This results in high maxima of the intensity values ​​and / or large variance of the intensity values ​​along the radial scan lines containing the struts. Therefore, where the set of representative values ​​corresponds to the maximum intensity value or variance value for each line, the position of the strut can be detected in the data as an area of ​​relatively high representative values ​​(relative to the surrounding values), i.e., an area of ​​high contrast with the surrounding values.

[0140] This is Figure 7 In the description, Figure 7 A 2D plot of the maximum intensity values ​​for a radial line for an example ultrasound data set is shown. The illustrated region contains a portion of a stent. The x-axis corresponds to the frame number of the scan line and the y-axis corresponds to the azimuth angle of the scan line.

[0141] The grid-cross pattern of the bracket is clearly visible in the data as bright areas of high contrast are set against a background of darker values.

[0142] Processing the data as a set of representative values ​​for each scan line provides advantages over conventional modes of processing and representing captured IVUS data. In typical known solutions, the entire IVUS data set is used in any analysis process, and detection and identification of objects or anatomical regions within the data are performed based on segmentation. Segmentation can utilize anatomical pattern information and includes detecting the outline of identifiable objects based on reference outlines or images.

[0143] However, especially for intravascular ultrasound, lumen segmentation is unreliable and very error-prone. Thus, segmentation of the captured data can only provide a relatively rough representation of the vessel wall, wherein smaller details such as vessel bifurcations can be easily missed.

[0144] The present solution avoids the need for segmentation by instead simply analyzing calculated representative values, such as searching for patterns or high contrast areas in the values. Generating these values ​​in itself performs the extraction of the key data points from the data that are most relevant to the identification of the presence of an object.

[0145] Thus, instead of applying segmentation, embodiments of the present invention follow the most common approach in the field of 3D volume rendering, where a 2D output image is constructed by directly evaluating the 3D volume through a collection of radially directed scan lines. The scan lines extend along the radial depth direction (from the perspective of the generated 2D image of the vessel wall).

[0146] This solution allows the presence of even very small spatial details to be identified based on collecting ultrasound data along an entire radial path. For example, in the case of detecting a vessel bifurcation, multiple adjacent data points along the same radial path effectively "collaborate" in creating a specific local low intensity value for such a bifurcation.

[0147] exist Figure 8 An example workflow of an example data analysis method according to one or more embodiments is shown in the block diagram of .

[0148] First, raw (polarized) ultrasound data is received 52. Alternatively, the raw polarized data can be output to a graphical user interface (GUI) 62 at this point in at least some form for viewing by a user. For example, certain radial lines of the data can be output for a complete set of frames. In some examples, this can be output as a graphical or visual output.

[0149] Processing 54 is then performed to extract a single representative value from each radial scan line. This includes performing a frame-by-frame calculation of a representative value per radial A-line (envelope signal after beamforming) within the set of (radial) ultrasound A-lines that together form a single frame. The representative value can provide a contrast mechanism to create a compressed representation of the data for each frame.

[0150] The use of representative values ​​that represent data according to a contrast or change attribute (contrast mechanism) is ideally suited for subsequent detection of the presence of an object. This is because it allows sufficient information to be retained for detecting the presence of an object, such as a stent, graft, guide ring, guide wire, or any other intravascular object, because objects tend to exhibit high contrast areas compared to the background. Thus, the scheme of an embodiment of the present invention is at least partially guided by the observation that for at least some choices of contrast mechanism, the output set of representative values ​​exhibits retention of sufficient relevant information to achieve successful detection or classification of the object.

[0151] The resulting set of representative values ​​is then buffered 56. Alternatively, the derived set of representative values ​​may be output at this point to a graphical user interface (GUI) 62 for viewing by a user. They may be output in a graphical or visual form. For example, a 2D graph or curve of the generated set of representative values ​​may be generated, such as in Figure 5 , and control signals representing this graphical output transmitted to an associated GUI 62 (eg, a display device or patient monitor).

[0152] An object detection process 58 is then applied to the generated set of representative values, which detects the presence of an object in the data, and preferably determines the area within the data occupied by the object. This is based on deriving a set of probabilistic representative values ​​that derive the probability of the presence of the object in each frame of data, and then determining the area within the data occupied by the object based on these values.

[0153] The results of the detection process 58 are then buffered 60. The results of the detection process may be output to a graphical user interface 62 for display in a graphical, textual or any other form. Alternatively, the results may be stored locally or remotely, for example instead.

[0154] Fig. 9 A further more specific example workflow for an example data analysis method according to one or more embodiments is shown in In this example, the representative value derived from the ultrasound data is the maximum intensity value.

[0155] For applications in the presence detection of artificial objects (e.g. stents), it has been found that the use of so-called Maximum Intensity Projections (MIPs) is a useful representation of line-by-line data. MIPs are known to be computationally efficient methods in 3D volume rendering in order to visualize volume data stored as so-called voxel values ​​in a 3D data buffer. When applied to volume data, MIPs produce images that have the appearance of a semi-transparent projection of a 3D volume on a 2D image plane.

[0156] MIP is based on identifying the maximum intensity voxel or pixel along a projection line through the volume data and then using this value to output a 2D MIP image.

[0157] Below is a similar type of solution in some embodiments of the present invention, where a maximum intensity projection is performed along each radial line. This means that each radial line is converted to a single value corresponding to the maximum intensity value of the scan lines pointed along that radial direction. Thus, effectively, a similar MIP operation is used to convert the raw frame-by-frame 2D ultrasound polarization data into a single value (the maximum intensity value).

[0158] Furthermore, according to a beneficial set of examples, two representative values ​​are derived for each line: the maximum intensity value itself and an index representing the position of the maximum value along the radial scan line (referred to herein as the "maximum intensity index"). Extracting these two pieces of information enables certain more robust object detection processes to be performed, which will be explained in more detail further below.

[0159] The maximum intensity value and the maximum intensity index can be understood as a set of two 1D vectors.

[0160] For a given sequence of acquired frames, the maximum intensity value vector and the maximum intensity index vector can be represented as corresponding 2D graphs, with Figure 5 Similar to the figure in .

[0161] Fig. 9 and Fig.10 Two examples are shown in FIG. The top graph in each figure shows a graph of the maximum intensity value for each radial scan line, and the bottom graph in each figure shows a graph of the maximum intensity index value for each radial line. In each case, the values ​​are plotted against the azimuth of the line on the y-axis and the frame number on the x-axis. Fig. 9 corresponds to the maximum intensity and index value for the example set of data in which the object (particularly the scaffold) is present, and Fig.10 An example set corresponding to data in which the object does not exist.

[0162] Fig. 9 The presence of the mid-stand is clearly demonstrated by the characteristic pattern caused by the cross-grid arrangement of the stand struts. The relatively strong reflections give rise to bright local maximum intensity values ​​(upper curve), and the strut spacing gives rise to distinct trough patterns in the maximum intensity index values ​​(lower curve).

[0163] Fig.11 An example workflow for an example method employing extraction of maximum intensity values ​​and maximum intensity indices is shown.

[0164] As in Figure 8In a workflow of , the method comprises receiving raw polarized ultrasound data 52. Again, optionally, this may be output to a graphical user interface (GUI) at this stage, for example in graphical form. The method may comprise presenting the graphical output, or the GUI may perform the presentation.

[0165] The method then comprises extracting 55 the maximum intensity value and the maximum intensity index for each radial scan of the received data. This is then buffered 56.

[0166] An object detection process 59 is then applied which involves detecting whether an object is present in the data.

[0167] This may be based on deriving a set of probability values ​​representing the probability of the object being present in each frame of the data, and then determining whether the object is present based on these values.

[0168] The results of the object detection process are then buffered 61 .

[0169] If an object is detected, then an edge detection process 62 is applied to determine where the edge of the object is within the data. Again, this may be based on using a set of probability values ​​derived from the extracted maximum intensity and maximum intensity index values.

[0170] This process generates as output the frame index (or frame number) within which the edge has been found to be located. This can then be output to a user display such as GUI 70 for presentation to the user.

[0171] The edge positions themselves provide an indication of the area occupied by an object such as a stent. These essentially indicate the area between two identified edges, i.e., the contiguous set of frames that lie between the two edge position frames.

[0172] An example process for performing each of the above steps is described in more detail below.

[0173] Although the use of the maximum intensity value as the representative value for each radial line is discussed in detail in this example, the present invention is not limited to the use of this particular type of representative value. In other examples, the data for each scan line can be converted to different representative values. A plurality of different example representative values ​​that can be used according to an embodiment are listed above.

[0174] As discussed above, the method includes deriving a set of probability values ​​corresponding to the probability of the presence of an intravascular object, eg, an artificial intravascular object, at least in each frame from a set of extracted representative values ​​for radial scan lines.

[0175] In some examples, these values ​​can be time-averaged or smoothed to generate a continuous sequence or curve of probability values ​​according to frame number. This sequence of probability values ​​can then be processed to detect the region of the data (a set of consecutive frames) occupied by the object. In some examples, more than one region can be detected.

[0176] Generating probability values ​​according to a representative value set can be performed in different ways. Different example methods for generating a probability value set will now be described in detail. Afterwards, various example methods for detecting the presence of an object and determining the occupied area of ​​the object within the data according to the probability value will be described.

[0177] A first example method will be described below. This solution is based on deriving a maximum intensity value from each radial line as a representative value. Briefly, the maximum intensity value will be referred to as a "MIP value".

[0178] An exemplary method for converting frame-by-frame representative values ​​(e.g., MIP values) into probability estimates of objects (e.g., stents) present in associated frames is based on using a random forest (machine learning) classifier algorithm and a discrete wavelet transform (DWT) to reduce the dimensionality of input data. This solution preferably uses MIP values ​​as representative values, and is based in part on observations of characteristic patterns of both MIPs and MIP index values ​​that tend to be generated by the presence of stents. This solution has been shown to have less computational requirements than the second method (described further below) due to a relatively simple algorithm and the resulting reduced computational complexity.

[0179] Note that although the following examples are described with reference to stent detection for illustrative purposes, this is by way of example only. This solution is not limited to detecting stents. It can also be used to detect any other intravascular object, and has been successfully tested for use in detecting other vascular insertions, such as but not limited to guide rings or vascular grafts.

[0180] This solution is based on a combination of Discrete Wavelet Transform (DWT), standard deviation calculation, and Random Forest (RF) classifier. The workflow of the example method for deriving probability values ​​is given in Fig.12 It is shown in block diagram form.

[0181] First, the complete set of IVUS ultrasound data is processed 82 to reduce the data in each radial line to a single representative value in the form of a MIP value, and an additional accompanying index indicating the position along the corresponding line where the maximum intensity value is found.

[0182] These output values ​​are then provided as input to a discrete wavelet transform (DWT) 84. The DWT is used to reduce the dimensionality of the data from 2*256 values ​​to 15 values ​​per frame (corresponding to the 15 radial lines included in each frame). These numbers are exemplary only and do not limit the invention.

[0183] The discrete wavelet transform process 84 generates as output a set of band-wise variances 86 of the resulting wavelet coefficients. These effectively provide the second eigenvectors.

[0184] The band-wise variance or standard deviation 86 is then provided as input to a random forest (RF) classifier algorithm 88. This algorithm is trained to produce, for each frame, a probability value (between 0 and 1) indicating the probability that a stent (or other object) is present in the corresponding frame.

[0185] The RF classifier can be an algorithm that has been trained based on labeled example data. After training, the algorithm includes a set of trained decision trees, which combined together can be used as both a classifier (presence or absence in a frame) and a regressor (regression analysis). The regression function can be configured to generate a continuous probability value or function (or sequence of values) for a set of frames, respectively between 0 and 1.

[0186] As described above, the presence of an object (e.g., a stent) can be identified in the data based on the presence of a characteristic pattern in the data generated by the object. Most artificial objects such as stents generate relatively strong reflections, which result in bright local MIP values, and the support spacing results in different slot patterns in the MIP index. Thus, these different patterns provide a way to detect possible object presence. A classifier algorithm can be trained to detect such a pattern or use such a pattern as part of determining the probability that an object exists in a frame.

[0187] The set of probability values ​​may then be time averaged or filtered using a frame-by-frame filtering operation (temporal filtering operation) 90, which produces a sequence of local time average probability values. From these, the edges of the support (or other object) may be detected using a suitable edge detection process, and the area occupied by the object thus identified. An example process for determining the position of an object based on probability values ​​will be described in further detail below.

[0188] As described above, the solution involves applying a discrete wavelet transform, and generating an output for feeding into the RF classifier, comprising a set of band-wise variances of the wavelet coefficients. This process will now be discussed in more detail.

[0189] The choice of a frequency transform such as a wavelet transform is based on the observation that the pattern of the support of the stent (or other object) produces a characteristic regular pattern in the set of MIP values ​​(and MIP indices). This pattern is quasi-periodic in nature. For this reason, a characteristic pattern (indicating the presence of an object) can be associated with one or more characteristic spatial frequencies.

[0190] By way of example, in an advantageous example, a discrete Haar wavelet transform may be used.

[0191] The use of discrete Haar wavelets has the advantage of being relatively computationally limited, thus making the process faster and more computationally efficient. This simplification corresponds to the square of the Haar basis function. For illustration purposes, this shape is Fig.13 Shown in.

[0192] The continuous Haar basis wavelet function ψ(t) has the form:

[0193]

[0194] The normalized discrete Haar wavelet transform produces a continuous convolution of the input set of MIP values ​​with the following filter kernel:

[0195] and

[0196] There, a high-pass filter h[n] and a low-pass filter g[n] repeatedly convolve the input (MIP value) signal x[n] in a cascade of filtering and downsampling operations.

[0197] This cascade of filtering and downsampling operations is Fig.14 At each decomposition level 110 a , 110b, 110 c , the downsampled low-pass filtered approximation of the signal forms the input for the next signal decomposition. At each level, a high-pass filter h[n] and a low-pass filter g[n] with the kernel depicted above are applied.

[0198] For the case of applying a transform to the stent presence detection, the transform may be applied to the input signal x[n], including the MIP values ​​and the MIP indices. By way of example, a 7-level transform may be applied. This may produce a total of 512 output wavelet coefficients (2*256 wavelet coefficients).

[0199] For each subband, the standard deviation (or alternative variance) is calculated to form a vector of 2*7 subband-by-subband standard deviation values. These values ​​can be understood as spanning a 14-dimensional feature space. These calculated vectors of standard deviation values ​​form the input to the RF classifier algorithm 88.

[0200] A second example method for converting a set of representative values ​​for each radial line in each frame into a frame-by-frame stent presence estimate (probability value) is based on the use of one or more convolutional neural networks (CNNs). Optionally, these use inputs from multiple frames rather than a single frame. As the neural network is further trained, this solution can become more robust and reliable than the first example solution listed above. However, it can sometimes be more computationally complex.

[0201] This solution provides an alternative to using wavelet transforms. CNNs can be trained based on, for example, labeled sample data (a sample set of MIP values ​​and optionally also MIP indices). The algorithm can use deep learning.

[0202] In operation, the CNN is fed frame-by-frame MIP values ​​as input and produces as output a continuous probability value (or sequence of values) between 0 and 1 that the object is present in each corresponding frame.

[0203] Thus, using such a CNN can in some embodiments replace the combination of discrete wavelet transform DWT, standard deviation calculation and RF classifier used in the above solution.

[0204] Thus, using this deep learning solution, the input MIP values ​​(e.g., 2*256 MIP values, using the example above) form the input to the deep convolutional neural network. When trained with representative annotation data, the convolutional neural network can perform the functions of feature extraction / detection (albeit implicitly and hidden) and feature classification. Feature extraction means, for example, detecting the presence of an object. Feature classification means, for example, detecting which object has been detected in a set of possible objects.

[0205] By feeding the CNN with MIP value data rather than the original full polarimetric data set, the size of the neural network can be kept smaller than would otherwise be possible, allowing for a very computationally efficient implementation. Such a small network can, for example, run in real time on a typical CPU.

[0206] Fig.15 An example of a generated sequence of frame-by-frame probability values ​​120 of object presence in each frame is shown, shown against a background of maximum intensity values ​​extracted from each radial scan line. The y-axis of the background map corresponds to the radial line azimuth, and the x-axis corresponds to the frame number of each radial line.

[0207] Each of the above example solutions for deriving a set of probability values ​​includes using a machine learning algorithm, with the first solution using a random forest algorithm and the second solution using a convolutional neural network.

[0208] We will now briefly outline the fundamentals of machine learning algorithms.

[0209] A machine learning algorithm is any self-training algorithm that processes input data in sequence to produce output data. Here, the input data may include a set of probability values ​​for the presence of an object in each frame, or a set of representative values ​​derived for radial lines. The output data includes the presence or absence of an object, and may further include the area of ​​the data occupied by the object.

[0210] Suitable machine learning algorithms for use in the present invention will become apparent to those of ordinary skill in the art. Examples of suitable machine learning algorithms include decision tree algorithms (e.g., the random forest algorithm cited above) and artificial neural networks. Other machine learning algorithms such as logistic regression, support vector machines, or naive Bayes models are suitable alternatives.

[0211] The structure of an artificial neural network (or simply a neural network) is inspired by the human brain. A neural network consists of multiple layers, each layer including multiple neurons. Each neuron includes a mathematical operation. Specifically, each neuron may include a different weighted combination of a single type of transformation (e.g., the same type of transformation, a sigmoid function, etc., but with different weights). During the processing of input data, the mathematical operation of each neuron is performed on the input data to produce a digital output, and the output of each layer in the neural network is sequentially fed to the next layer. The last layer provides the output.

[0212] Methods for training machine learning algorithms are known. Typically, such methods include obtaining a training data set, including training input data entries and corresponding training output data entries. An initialized machine learning algorithm is applied to each input data entry to generate a predicted output data entry. The error between the predicted output data entry and the corresponding training output data entry is used to modify the machine learning algorithm. This process can be repeated until the error converges and the predicted output data entry is sufficiently similar to the training output data entry (e.g., ±1%). This is generally known as a supervised learning technique.

[0213] For example, in the case where the machine learning algorithm is formed by a neural network, the mathematical operation (weights) of each neuron can be modified until the errors converge. Known methods of modifying neural networks include gradient descent, back propagation algorithms, and the like.

[0214] The training input data entries correspond to an example set of object presence probability values ​​or a representative set of values ​​for a complete set of radial lines. The training output data entries correspond to the presence or absence of an object in the data and / or the region of the data occupied by the object (if the object is present).

[0215] In generating a frame-by-frame sequence of object presence probability values ​​(e.g. Fig.15 ), method 30 further comprises processing the values ​​to determine, based thereon, at least one region of the ultrasound data (eg, a continuous subset of the frames) occupied by the intravascular object.

[0216] Detecting the area occupied by an object can take advantage of the fact that, particularly for manually inserted objects such as stents, grafts, guide sheaths, and guidewires, the objects are of separate and varying sizes, which results in the objects appearing in multiple consecutive frames during the pullback of the IVUS device through the vessel lumen. This continuous presence feature throughout the finite width area enables estimation of the occupied area, particularly the longitudinal distal and proximal edges of the object along the longitudinal direction of the pullback.

[0217] Detecting a region of the data (e.g., a set of consecutive frames) occupied by an intravascular object can include one of at least two solutions. A first solution can include searching for all consecutive frame regions having corresponding probability values ​​all above a certain threshold, such as above 50% or above 60% or any threshold that can be defined.

[0218] Another more complex solution involves testing multiple possible occupied regions within the data, evaluating a predefined cost function (based on a derived set of probability values ​​for the complete ultrasound data set) for each, and then determining the most likely occupied region for any one of the one or more objects based on identifying the region with the minimum (or maximum) cost function result.

[0219] In either case, the solution may include identifying the edges of the object and deriving the occupied area based on the edge detection. Alternatively, the solution may include identifying the complete area occupied without first identifying the edges of the object.

[0220] The first solution is explained below.

[0221] According to a first solution, an initial processing step may first be performed including averaging or smoothing the probability values ​​over time. Time means the dimension in which the frame numbers are incremented, i.e., the time domain corresponds to the frame number domain. In operation, since frames are captured at regular intervals, the frame index may effectively be understood as a time index. Averaging with respect to time (or in time) thus corresponds in this example to averaging with respect to incrementing the frame numbers.

[0222] Thus, according to one or more advantageous embodiments, a frame-by-frame filtering operation (temporal filtering operation) is performed, which produces a sequence of locally time-averaged probability values.

[0223] Fig.16An example sequence of filtered probability values ​​produced as an output of an example temporal filtering operation is shown in FIG. Fig.15 Filtered values ​​124 (in the form of a continuous probability curve) of an example set of raw probability values ​​shown in FIG. 1 . The filtered probability curve 124 is shown superimposed on the Fig.15 For easy reference, Fig.16 The original set of probability values ​​120 shown in is located directly below the time-averaged probability curve 124 .

[0224] exist Fig.15 and 16 , the corresponding probability curves 120, 124 are shown as superimposed on a 2D plot of representative values ​​for each radial line (y-axis, azimuth angle) in each frame (x-axis; frame index), from which the probability values ​​are derived. The sequence of two probability values ​​(original and time average) varies between a maximum of 100% (or 1) and 0% (0).

[0225] Temporal Averaging or Filtering Averaging or filtering is performed in the frame index (x-axis) domain to smooth out very rapid local changes in probability.

[0226] According to this first example solution, regions within ultrasound data occupied by artificial intravascular objects can be detected based on applying a simple threshold to a set of averaged probability values. Specifically, the solution comprises identifying a consecutive subset of frames whose time-averaged probability values ​​remain above a predefined threshold.

[0227] In an advantageous example, the predefined threshold may be 50%. In other examples, the predefined threshold may be greater than 50%, such as 60% or 70%. In other examples, it may be less than 50%. The threshold may be set depending on the context, such as the noisiness of the acquired data and the overall level of probability assigned to the frame. For example, in the case where the data is noisy and the resulting probability values ​​vary around a generally low probability region, the threshold probability value may be set lower (and vice versa in the case where the noisy data results in a generally high probability value set).

[0228] Identifying a consecutive set of frames having probability values ​​above a predefined threshold may, in an example, include applying a rectangular low pass filter (in the probability domain), and then applying a threshold to the output.

[0229] According to some examples, the process may include identifying the largest contiguous set of frames within the data set that have a probability value above a predefined threshold. This handles the situation where there is more than one region that meets the threshold criteria. All regions may be selected as regions occupied by the corresponding object, or only the longest set of frames may be selected. There may be pre-stored prior information representing the expected typical width of regions occupied by objects in the data, and this information may be applied to select only one or more of any multiple sets that are found to be regions that pass the probability threshold.

[0230] The process may in some examples include a dynamic process of identifying the longest continuous set of frames with the highest probability value. This may be an algorithm that applies different weights to the probability values ​​for the length of the set in order to identify the most likely region of occupancy.

[0231] Fig.16 An identified region 128 of consecutive frames within the data occupied by an object is shown. The identified proximal edge ("PE") and distal edge ("DE") of the object are indicated by corresponding arrows.

[0232] The above solution has the advantage of being relatively simple and therefore places low computational demands on the processor executing the method.

[0233] A second solution for determining the area within ultrasound data occupied by an intravascular object based on probability values ​​will now be explained. This solution has been found to perform more robust and reliable in accurately identifying the area occupied by the object, such as detecting the edges of the object.

[0234] This second solution is based on a number of trial areas within the test data, the test comprising calculating for each trial area the difference between the set of probability values ​​for the trial area and an equivalent set of probability values ​​representing the area exactly occupied by the object. The test specifically comprises calculating a cost function comprising as a first term the sum of the set of difference values ​​calculated for the entire set of frames throughout the trial area. The cost function also comprises as a further one or more additions the sum of the probability values ​​for consecutive sets of frames on one or both sides of the trial area.

[0235] This solution will now be explained in more detail.

[0236] This solution is based on the observation that the presence of a stent estimate tends to take a rectangular outline depending on the pullback distance (or depending on the frame number). The concept consists in testing multiple hypotheses of the presence of an object, such as a stent, within a region (frame) extending between frame d and frame p, where frame d and frame p contain the distal and proximal object edges, respectively (or vice versa). For each trial region (each trial combination of edge positions d and p), a cost function C(d, p) is calculated, associated with a given hypothesis that an object is present within that given region.

[0237] In some examples all possible regions of occupancy may be tried, or, to save processing resources and reduce processing time, only a subset of the possible regions of occupancy of the object.

[0238] The cost function C(d,p) is calculated based on the difference between the set of frame-by-frame (raw) probability values ​​of the object's presence in each frame (calculated in the steps explained in the example above) and a test wave (e.g., a square wave) covering the test area (representing a 100% (or predefined map) probability at each frame inside the test area and a 0% probability outside the area). This test wave thus corresponds to a certain (or predefined high) probability that the object occupies the test area, and does not occupy any other area.

[0239] This is Fig.17 , a sequence of frame-by-frame probability values ​​142 is shown with an example trial region test wave 144 superimposed thereon. In this case, the test wave is a square wave.

[0240] The cost function is based on Fig.17 The difference between the test curve 144 in the experimental area and the frame-by-frame probability value curve 142 is given by:

[0241]

[0242] where s is the frame-by-frame object presence estimate (probability value) according to the frame index, and where L is the number of total number frames captured in the dataset (the number of frames captured during a complete pullback through the vessel).

[0243] The process consists in identifying the values ​​of d and p that minimize the cost C, ie identifying the region (contiguous set of frames) having a minimum value for the cost function.

[0244] The values ​​of d and p that minimize C are chosen as the estimated distal and proximal object edge locations.

[0245] exist Fig.17 In the example set of probability values ​​of Fig.18 The values ​​of d and p that result in the minimum cost function are shown in . Fig.18The resulting minimum cost trial region 145 for consecutive frames is shown. This region 75 is thus provided by the algorithm as the inferred region occupied by the intravascular object.

[0246] Direct calculation of C(d, p) involves computing three sum terms for each combination of d and p. This is somewhat computationally inefficient.

[0247] A computationally faster solution can be considered. This solution is based on the initial calculation of two cumulative cost functions. The first function N(i) describes the cumulative cost of the "no object" assumption between frame 0 and frame i,

[0248]

[0249] The value of N effectively describes the Fig.19 The area 148 below the probability curve is shown in shaded area.

[0250] The second function S(i) describes the cumulative cost of the “object presence” hypothesis between frame 0 and frame i,

[0251]

[0252] The value of S effectively describes the Fig. 20 The area 150 below the probability curve is shown in shaded area.

[0253] Using the precomputed cumulative cost functions N(i) and S(i), each of the three summation terms in expression (3) above can now be computed more efficiently as follows:

[0254] C(d,p)=[N(d-1)]+[S(p)-S(d)]+[N(L-1)-N(p+1)].(6)

[0255] exist Fig.21 , the associated regions for each of the three terms in equations (3) and (6) are depicted, with shaded area 154 corresponding to the first term, area 156 corresponding to the second term, and area 158 corresponding to the third term.

[0256] Using precomputed cumulative cost functions, N(i) and S(i) have been shown to increase the speed of the object region position estimation algorithm by a factor of approximately 300.

[0257] The increased speed can be understood from the fact that this solution pre-computes the cumulative cost functions for the two hypotheses (bracket presence and no bracket presence) according to the frame index, resulting in two functions N(i) and S(i). S(i) and N(i) can be understood as simple 1D arrays. Each cumulative cost function therefore takes only one parameter, rather than multiple parameters, resulting in increased speed.

[0258] Specifically, it can be seen in Expression (6) that each of the three sum terms (in Expression (3)) is replaced by an evaluation using N(i) and S(i).

[0259] According to some examples, the method is configured to test all possible combinations of d and p, with an additional constraint of pd>M, where M is a predefined minimum number of frames that an object is expected to span. Increasing values ​​of M reduce the number of trial regions tested and thus increase algorithm speed, but at the expense of loss of detection capability for objects smaller than M frames.

[0260] According to some examples, the method can be applied to detect the position of a stent (e.g., stent edge) within the data. However, it is also possible to detect any other object, such as the area of ​​consecutive frames occupied by a graft, guide sheath, or wire. These represent only a limited set of examples.

[0261] According to one or more examples, an additional constraint may be imposed in the form of an assumption that the near edge of the object is always located at the last frame in the received data set. Thus, here, p=L-1. This greatly reduces the complexity of the cost function calculation and also reduces the number of cost functions that are required to be searched.

[0262] Using this assumption is particularly useful for, for example, detecting the edge of a sheath within a blood vessel. This is because, in practice, most IVUS pullback scans end at the location of the introducer sheath. As the introducer sheath thus tends to appear at the end of the pullback, the last frame of the pullback usually contains the introducer sheath. Therefore, it is a reasonable assumption to fix the proximal edge p to be equal to the index of the last frame, where appropriate.

[0263] The inner diameter of the guide sheath tends to be smaller than the lumen of a healthy vessel. Automatic searching for the smallest lumen region may therefore mistakenly identify a frame containing a guide sheath as a diseased region of the imaged vessel. A narrow guide sheath is then mistaken for a stenosis. Thus, the ability to automatically identify the presence of a guide sheath is valuable for avoiding this error.

[0264] According to one or more examples, the algorithm can be configured to test for the probability that the object is present in the received IVUS data, which spans the entire set of frames in the data set. Thus, this allows the algorithm to successfully handle situations where the object is longer than the total length of the set of frames in the data set, i.e., L <M。

[0265] In one or more examples, the algorithm can be configured to allow detection of more than one object within a set of frames, such as two or more objects that are separated from each other.

[0266] According to an example of another aspect of the present invention, an ultrasound data processor for analyzing ultrasound data captured by an IVUS device is provided. The processor is adapted to perform the following steps:

[0267] A plurality of frames of ultrasound data captured by an IVUS device are received, each frame including data for a plurality of radial scan lines, each radial scan line corresponding to a different rotation angle ( ) about a longitudinal axis (A-A') of the device. ) received acoustic signal;

[0268] processing ultrasound data for each radial scan line in each frame to reduce the data for each line to a single representative data value for the line;

[0269] deriving a set of probability values ​​corresponding to the probability of existence of the object in the artificial blood vessel at least in each frame based on the set of representative data values;

[0270] A region within the ultrasound data occupied by an intravascular object is determined based on the probability value.

[0271] The implementation options and details for each of the above steps implemented by the processor may be understood and explained based on the explanation and description provided above for the method aspects of the present invention.

[0272] Any of the examples, options or embodiment features or details described above in relation to the method aspect of the present invention may be applied or combined or incorporated into the present processor aspect of the present invention with appropriate modifications.

[0273] The processors discussed above can be implemented in various ways, using software and / or hardware, in order to perform the various functions required. The processor typically employs one or more microprocessors that can be programmed using software (e.g., microcode) to perform the required functions. The processor can be implemented as a combination of dedicated hardware to perform some functions and as one or more programmed microprocessors and associated circuits to perform other functions.

[0274] Examples of circuits that may be employed in various embodiments of the present disclosure include, but are not limited to, conventional microprocessors, application specific integrated circuits (ASICs), and field programmable gate arrays (FPGAs).

[0275] In various implementations, the processor may be associated with one or more storage media such as volatile and non-volatile computer memory (e.g., RAM, PROM, EPROM, and EEPROM). The storage media may be encoded with one or more programs that, when executed on one or more processors and / or controllers, perform the required functions. The various storage media may be fixed within the processor or controller or may be transportable so that one or more programs stored thereon can be loaded into the processor.

[0276] According to an example of a further aspect of the present invention, an ultrasound system is provided.

[0277] The ultrasound system includes an intravascular ultrasound (IVUS) device for capturing ultrasound data within a vessel lumen along a plurality of different radial scan lines; and

[0278] The ultrasound data processor according to any of the examples or embodiments listed above and described below or according to any of the claims of the present application is operably coupled to an intravascular ultrasound device and is configured to receive ultrasound data captured by the IVUS device.

[0279] Those skilled in the art can understand and implement variations to the disclosed embodiments when practicing the claimed invention by studying the drawings, the disclosure and the appended claims. In the claims, the word "comprising" does not exclude other elements or steps, and the expression "one" or "an" does not exclude multiple. A single processor or other unit can implement the functions of several items cited in the claims. The only fact is that certain measures cited in the common different claims do not indicate that the combination of these measures cannot be fully utilized. If a computer program is discussed above, it can be stored / distributed on a suitable medium, such as an optical storage medium or solid-state medium provided together with other hardware or as part of other hardware, but can also be distributed in other forms, such as via the Internet or other wired or wireless telecommunications systems. If the expression "suitable for" is used in the claims or the specification, it is noted that the expression "suitable for" is intended to be equivalent to the expression "configured to". Any figure mark in the claims should not be constructed to limit the scope of protection.

Claims

1. An ultrasound data analysis method (30) for analyzing ultrasound data captured by an intravascular ultrasound (IVUS) device, the method comprising: receiving (32) a plurality of frames of ultrasound data captured by an IVUS device, each frame comprising data for a plurality of radial scan lines, each radial scan line corresponding to a different rotation angle along a longitudinal axis (A-A') about the device Received acoustic signal; processing (34) the ultrasound data for each radial scan line in each frame so as to reduce the data for each line to a single representative data value for that line; deriving (36) a set of probability values ​​corresponding to the probability of the presence of the object within the artificial blood vessel at least in each frame based on the set of representative data values; A region within the ultrasound data occupied by an intravascular object is determined (38) based on the probability value.

2. The method (30) according to claim 1, wherein: Determining (38) the region occupied by the intravascular object includes identifying a contiguous subset of frames of the data occupied by the object, and / or identifying a contiguous subset of radial lines occupied by the object.

3. The method (30) according to claim 1 or 2, wherein: Determining (38) the region occupied by the intravascular object includes identifying a contiguous subset of frames for which the derived probability values ​​are all above a predefined threshold.

4. The method (30) according to claim 1 or 2, wherein: Determining (38) the area occupied by the intravascular object includes detecting edges of the intravascular object.

5. The method (30) according to claim 1 or 2, wherein: Determining (38) the area within the data occupied by the intravascular object includes testing a plurality of trial areas within the data, the testing including calculating, for each trial area, the difference between a set of probability values ​​for the trial area and an equivalent set of probability values ​​representing the object precisely occupying the area.

6. The method (30) according to claim 5, wherein: The test comprises calculating a cost function comprising, as at least one additional term, the summation of a set of difference values ​​according to claim 5 for the entire set of frames throughout the trial area.

7. The method (30) of claim 6, wherein: The cost function comprises, as one or more further additional terms, the sum of the probability values ​​for consecutive sets of frames on one or both sides of the trial area.

8. The method (30) according to any one of claims 1, 2, 6 and 7, wherein: The representative data value represents the maximum intensity value of the corresponding radial line.

9. The method (30) according to claim 8, wherein: The method also includes determining a set of index values ​​representing the location of each maximum intensity value along each radial line.

10. The method (30) according to any one of claims 1, 2, 6, 7 and 9, wherein: The method comprises generating a curve or graph representing the data values, for example a curve or graph representing the values ​​plotted against frame numbers and / or radial line numbers.

11. The method (30) according to any one of claims 1, 2, 6, 7 and 9, wherein: A set of probability values ​​is derived (36) based on the set of representative data values ​​including one or more characteristic pattern features for detecting the presence of an object within the set of representative data values.

12. The method (30) according to any one of claims 1, 2, 6, 7 and 9, wherein: Deriving (36) a set of probability values ​​from the set of representative data values ​​comprises using a classifier algorithm, and preferably wherein the classifier algorithm comprises a machine learning algorithm such as a neural network.

13. The method (30) according to any one of claims 1, 2, 6, 7 and 9, wherein: Each frame corresponds to a different longitudinal position along the lumen.

14. An ultrasound data processor for analyzing ultrasound data captured by an IVUS device, the processor being adapted to: receiving a plurality of frames of ultrasound data captured by the IVUS device, each frame comprising data for a plurality of radial scan lines, each radial scan line corresponding to a different rotation angle along a longitudinal axis (A-A') about the device; Received acoustic signal; processing the ultrasound data for each radial scan line in each frame to reduce the data for each line to a single representative data value for the line; deriving a set of probability values ​​corresponding to the probability of the presence of the object within the artificial blood vessel at least in each frame from the set of representative data values; A region within the ultrasound data occupied by an intravascular object is determined based on the probability value.

15. An ultrasound system comprising: An intravascular ultrasound (IVUS) device (12) for capturing ultrasound data within a blood vessel lumen along a plurality of different radial scan lines; as well as The ultrasound data processor of claim 14, operatively coupled to the intravascular ultrasound device and configured to receive ultrasound data captured by the IVUS device.

Images