Method for removing artifacts from echocardiography doppler video

Abstract

A computer-implemented denoising method for removing artifacts from Doppler type ultrasound videos comprises the steps of: acquiring a plurality (n) of frames constituting an ultrasound video; identifying Doppler-activated pixels; generating, for each frame belonging to said plurality of frames, a first set of image data so as to associate each pixel position of the corresponding frame with at least a first or second category so as to classify the pixels of the corresponding frame, and identifying, for each frame, Doppler-activated pixels; generating at least a succession sequence associated with a frame portion containing the identification data ordered according to the sequence of frames, so that adjacent identification data in the sequence refer to consecutive frames, wherein each identification data adopts a first value when the corresponding frame portion contains Doppler-activated pixels, or a second value different from the first value when the corresponding frame portion does not contain Doppler-activated pixels; in said succession sequence, calculating the length of each successive sub-sequence comprising consecutive identification data having the first value; automatically calculating a reference threshold value; eliminating Doppler pixels if the latter belong to a succession sequence lower than the threshold value.

Description

TECHNICAL FIELD

[0001] The present invention relates to a method for removing artifacts contained in echographic Doppler videos. BACKGROUND

[0002] Ultrasound imaging technology is commonly used to produce diagnostic images of internal features of a subject, such as human anatomy.

[0003] Ultrasound imaging is the technology that reproduces on a two-dimensional image the anatomical section of the human body corresponding to the scanning plane. The mechanism that transforms the information contained in the echo signals into a two-dimensional image is complex and consists of several steps. Some of these steps depend on the propagation of ultrasound in biological tissue, others depend on the device technology or how the operator manipulates them. Ultrasound is generated by transducers and is properly focused, they propagate in the tissue at an approximately constant speed, which varies slightly between tissues depending on the density or acoustic impedance. The contact of ultrasound with the various anatomical structures marks the beginning of various physical phenomena such as reflection, scattering, diffraction and absorption. The first two of the above phenomena generate echo signals that return to the transducers and are properly processed, producing the ultrasound image. Basically, the information contained in the raw signal is not sufficient to produce the ultrasound image, but it must be integrated with other basic information: the first is the depth of the signal source, which is determined as a function of the time elapsed between the emission of the beam and the return of the echo signal; the second is the direction of the signal along the lines that make up the scanning plane. In other words, in order to generate the ultrasound image, it is necessary to acquire these three information for each point of the scanning region.

[0004] Ultrasound imaging devices are complex instruments that can handle several imaging modalities. A simplified block diagram of an ultrasound imaging device includes:

[0005] - a pulse generator, which is responsible for generating the electrical signal;

[0006] - a piezoelectric transducer, which is driven to perform both the two duties of source and receiver. As a source, the transducer is responsible for sending the ultrasound acoustic signal generated by converting the electrical pulse into a wave. The pulse of acoustic waves that travels through the tissue is partially reflected by the material parts that have different acoustic impedance. Therefore, as a receiver, the transducer is responsible for detecting the reflected pulse and converting it into a radio frequency (RF) electrical signal;

[0007] - a TGC module, which acts on the received signal by amplifying it in a way proportional to the depth of the echo signal generation and the signal return time, thus compensating the gain in order to correctly represent the acoustic impedance differences that make up the surface of the discontinuity. In other words, the signals that arrive at the probe from deeper planes of the acoustic field are weaker than the echo signals that originate from reflectors with the same characteristics but located on shallower planes;

[0008] - a demodulator which converts the amplified RF signal into a single peak representing the signal of each reflector encountered by the ultrasound. This makes it possible to identify all the points at which a reflection has occurred and to quantify the echo amplitude;

[0009] - a scan converter which converts the time-continuous input signal, samples it at defined intervals and transforms it into a discrete numerical form which will be coded according to the format that can be represented on the monitor, according to the signal intensity with 16, 64, 256 grey levels, since the image is acquired along lines which make up the scan field, while the image is represented along horizontal lines on the monitor. This is achieved in two main macro-phases, in which, in the first phase, at the input, the scan converter performs the analog-digital conversion of the RF signal and stores the data in binary format on a storage matrix, and in the second phase, at the output, the scan converter performs the digital-analog conversion of the data stored on the storage matrix;

[0010] - a display / storage system which allows the trace to be viewed, usually by means of a monitor, and stored on a suitable storage medium for subsequent viewing.

[0011] A particular application of diagnostic ultrasound imaging uses the Doppler measurement to detect and display fluid flow velocity. This is a physical phenomenon whereby the frequency of an acoustic wave encountering a moving body undergoes a variation proportional to the speed with which the body moves itself.

[0012] The Doppler effect is therefore based on the measurement of the frequency variation between the incident beam and the reflected beam from a moving body, which in medical applications is in most cases represented by a red blood cell, and therefore by blood.

[0013] The Doppler effect is therefore the principle on which many ultrasound techniques are based, in which the movement in biological tissues is studied. The reflected echo frequency variation is related to the speed and direction of the reflecting body. The received echo signal is compared with a reference value in order to establish the fluid flow velocity through the area. The flow velocity can be determined by measuring the Doppler shift of the echo signal received from the structure under examination.

[0014] There are two basic Doppler acquisition systems:

[0015] - continuous wave (CW), which allows the flow pattern and movement to be studied;

[0016] - pulsed wave (PW), which also provides information about the distance between the reflecting surface and the transducer.

[0017] Color Doppler, by integrating flow information and real-time two-dimensional images, is based on the same principle, where, by convention, approaching flow has a red color, while leaving flow has a blue color. As known in the prior art, it is then possible to view the flow velocity by coloring, in which different shades and color intensities represent the flow rate and direction within a gray-scale image. If there is turbulence, for example at the bifurcation of a blood vessel, there will be an alternating pattern of blue and red patches. From this, it will be possible to distinguish the flow direction with respect to the probe; moreover, these systems help to evaluate the rate of flow itself and the laminar or turbulent state.

[0018] It is clear how important it is to obtain the most accurate results from the ultrasound machine, since this can be used for the correct diagnosis by the doctor.

[0019] This highlights two main problem categories of these systems, one related to the operation of the doctor or operator, and therefore to his experience with the ultrasound machine, which must be operated on the basis of the set parameters in order to obtain effective results, and the other related to the system itself.

[0020] The ultrasound machine parameters that can be set by the doctor or technician are, for example, the gain control and the ultrasound focus setting, in order to achieve the most uniform distribution of image brightness by amplifying the signal from deeper layers, maximizing the contrast and avoiding saturation. On the ultrasound machine, there is a command that allows you to act on the scale setting of the TGC used according to the personal working habits of the doctor or technician, to improve the representation of the deep layers.

[0021] This parameter directly acts on the signal background, understood as the signal-to-noise ratio, generated by the circuit in the production phase, the reception and processing of the ultrasound signal, which will be represented in the image as an incorrect representation of the flow signal in the vascular area, due to the echo structure characteristic of the offset that increases with excessive signal amplification or gain values, an example of incorrect representation is Figure 5eThe focus settings allow to modify the number of active foci, the shape and the thickness of the ultrasound beam, which when not properly set, affect the spatial resolution of the ultrasound image. Other settings that can be configured are the imaging frequency, where an increase in its value allows to optimize the resolution, while a decrease in its value allows to increase the beam penetration. Another parameter on which the physician or operator can act is that relating to the compression curve, which allows to change the correspondence law between the signal amplitude and the gray levels represented on the monitor. With a linear intensity / amplitude relationship, the gray levels are directly proportional to the differences in acoustic impedance of the tissues being examined. In some cases, this is not the best setting required for diagnosis, so the physician or operator is able to represent some acoustic impedance ranges with greater evidence than other acoustic impedance ranges that are less interesting, for example by compressing the higher intensity echoes and allowing the lower intensity echoes to be represented with a high number of gray levels. A sensitive parameter in clinical practice is the "pulse repetition rate" (or "flow sensitivity"), which controls the ability of the system to acquire slow flows. If the pulse frequency is too high, the sensitivity to slow flows decreases, resulting in a loss of the corresponding signal.

[0022] In the following description of the possible artifacts that can appear in the results of an ultrasound examination, other key categories belonging to the system or deriving from the limits of the technology will be highlighted, mainly due to the interaction between ultrasound and biological structures. The key problems of instrumental origin and / or problems related to the experience of the physician or technician in the use phase of the ultrasound machine and in the configuration phase are reflected on the results generated by the ultrasound machine itself and are generally identified with the name of artifacts. An artifact therefore refers to false or distorted information generated by the ultrasound machine or by the interaction of ultrasound with the tissues, which overlap the noise on the Doppler signal. In particular, the experience of the physician or technician in using the ultrasound machine directly leads to the variable presence of more or less quantities of artifacts, depending for example on the positioning of the probe, the movements made by the probe itself and the speed with which the probe performs such movements.

[0023] Given that, for example, in color Doppler ultrasound images, artifacts might be defined as solid-color pixels that do not accurately represent vascularization, artifacts that may be found in Doppler ultrasound imaging can be confusing or misleading in terms of flow information. As mentioned above, three main factors contribute to this problem: misconfiguration of equipment and insufficient signal acquisition due to human error, anatomical factors, and technical limitations. For example, incorrect gain settings, wall filter settings, or velocity scale settings can lead to the loss of clinically relevant information, such as the presence or absence of flow in the vessels, flow direction and velocity, or tracking distortion that may show conditions significantly different from actual physiological conditions. Motion artifacts are particularly relevant regarding artifacts dependent on inappropriate acquisition. These artifacts include, for example, artifacts caused by inappropriate acoustic wave angle settings and artifacts caused by overly rapid signal acquisition, which are the most common errors causing flash artifacts. Regarding problems caused by acoustic wave angle (i.e., the angle between the operator's hand and the probe), when it is greater than 60°, the amplitude of the spectral curve gradually decreases, making rate calculations increasingly unreliable. On the other hand, when the angle of action of the sound wave is close to 90°, no signal is recorded, although flow can produce low-amplitude signals. Below are the main artifacts found in color Doppler ultrasound according to previously defined categories.

[0024] Artifacts depending on misconfiguration

[0025] -Doppler gain setting error:

[0026] An appropriate gain setting is crucial for the accurate representation of flow characteristics. With gain settings that are too low, some relevant information may be lost, thus requiring frequent adjustment of the gain to maximize trajectory visualization. Conversely, higher gain degrades the envelope signal, corrupting its representation on screen, and the simulation may, for example, give a spectral broadening that gives the appearance of turbulent flow.

[0027] -Inappropriate angle setting:

[0028] For issues related to the acoustic wave action angle, correction parameters can be set on the ultrasonic machine; however, this correction is useless when the angle is completely incorrect. This type of artifact can also occur due to the use of a transducer with a very high frequency or a lack of gain adjustment.

[0029] -Inappropriate filter settings:

[0030] The filter phase is designed to remove low-frequency Doppler signals from echoes of slowly moving soft tissue, and the filter's cutoff frequency is operator-selectable. Setting it too high may result in the loss of diagnostically relevant speed information.

[0031] - Artifacts caused by spectral dispersion:

[0032] Spectral dispersion may occur due to excessive changes in system gain or grayscale sensitivity.

[0033] Anatomical artifacts

[0034] -Flash artifact:

[0035] This artifact occurs as a sudden burst of color extending into a more or less extended area of ​​the scan field. Color coding is entirely artificial and can be caused by excessively rapid transducer movement or by heart movement or arterial pulsation that causes slight movement of the reflective surface.

[0036] -Pseudo-flow artifacts:

[0037] Artifacts represent fluid movement that differs from that of blood. These types of artifacts can be observed as an over-amplification of Doppler signals, under-encoding of colors (which may result in signals appearing, for example, within cystic formations), continuous moving structures, or as a result of respiratory dynamics. They can also occur due to the presence of mirror artifacts or scintillation.

[0038] Artifacts dependent on technological limitations

[0039] - Unclear direction:

[0040] Directional ambiguity can occur when the ultrasound beam intersects a blood vessel at a 90° angle. Detected Doppler signals appear both above and below the spectral baseline. Furthermore, directional ambiguity is more pronounced and tracking less accurate at high-gain settings. When using a fan-shaped transducer to generate color Doppler images, flow perpendicular to the beam typically exists along small segments of the blood vessel parallel to the transducer surface; this problem becomes even more pronounced when using a linear probe.

[0041] -Side lobe artifacts:

[0042] An electron-focusing array transducer directs the main beam to the region of interest to be examined. However, due to the spacing of the array elements, weak side lobes may target areas unrelated to the main lobe. The exact location of these lobes depends on the transducer design itself. If these side lobes hit a highly reflective surface (such as bone), the echo returning to the transducer can be detected on the screen in conjunction with the main beam echo.

[0043] -Random noise artifacts:

[0044] In Doppler ultrasound machines, noise is proportional to gain, as in all circuits. Random noise, especially when the gain is set too high, manifests as flash artifacts (whenever there is reciprocating probe-tissue movement) or as the appearance / disappearance of colored areas.

[0045] - Blinking artifact:

[0046] This type of artifact is visible in highly reflective structures and manifests itself as a wavy, colored mosaic associated with the signal characteristics of the presence of background noise. Its phantom is strictly dependent on the ultrasound machine configuration and is generated by a narrow band of inherent electronic noise known as phase jitter.

[0047] In the prior art, several signal processing techniques are known to be used to filter out unwanted signals within ultrasound imaging systems, which can lead to one or more artifacts as previously described. However, these solutions, implemented as part of the image display generation process within the ultrasound machine and inserted upstream of image generation into the processing and / or filtering stages, exhibit more or less related self-failure effects, such as loss of instrument sensitivity and / or information relevant to diagnostic purposes.

[0048] Moreover, the elimination of potentially different types of artifacts depends on different combinations of ultrasound machine parameter configurations and on the available processing and filtering enhancements depending on the ultrasound machine model. Summary of the Invention

[0049] The object of this invention is to at least partially address the aforementioned drawbacks caused by instruments or human intervention by acting on the ultrasound video generated by an ultrasound machine, rather than on the signal information and / or by means of on-board filtering and / or processing blocks for image generation. This is achieved by performing a denoising step on the video generated by the ultrasound machine acting on a 2D representation. The analysis of the 2D representation aims to identify any artifacts present in the video in order to remove these artifacts, thereby minimizing the impact of suboptimal settings and / or operator experience and / or technical limitations, thus providing an artifact-free video or a video with significantly reduced artifacts.

[0050] Therefore, denoising involves analyzing Doppler ultrasound video through the steps of the following method, with the aim of identifying any potential artifacts within the video that may overlap with or be adjacent to the true signal, in order to remove them, thereby obtaining a video without such alterations that could potentially mislead the diagnostic process. Other terms included in this invention that are worth defining and helping to contextualize their use are:

[0051] -Doppler activation, i.e., a set of pixels that carries the Doppler signal and Doppler artifacts such as color pixels;

[0052] - Ultrasonic signal, that is, a set of pixels that carries a non-Doppler ultrasonic signal, such as grayscale pixels belonging to the outer region from which Doppler signals are collected;

[0053] -Doppler signal, that is, the set of pixels that carry the real Doppler signal, which thus indicate the presence of vascularization.

[0054] The computer-implemented noise reduction method of the present invention, starting from decomposing a video into individual frames using standard video segmentation techniques, includes the following steps:

[0055] - Obtain n frames from the ultrasound video;

[0056] - For each of the n frames, identify a first pixel category, which represents the first pixel of the nth frame containing the ultrasound signal; and a second pixel category, which represents the second pixel of the nth frame containing Doppler activation;

[0057] - Generate a first image data set for each of the plurality of frames to associate each pixel position of the corresponding frame with at least a first or second category in order to classify the pixels of the corresponding frame and identify pixels containing Doppler activation for each frame.

[0058] - Generate at least a continuous sequence associated with a frame portion containing identification data ordered according to the frame sequence, such that adjacent identification data refers to adjacent frames, wherein each identification data uses a first value when the corresponding frame portion contains Doppler activated pixels, or uses a second value different from the first value when the corresponding frame portion does not contain Doppler activated pixels.

[0059] - In the persistent sequence, the length of each persistent subsequence is calculated, and the persistent subsequence includes continuous identification data with a first value;

[0060] - Calculate the reference threshold based on the length distribution of the persistent subsequence and referencing complete identification data of n frames;

[0061] - For each of the n frames, identify the second image dataset (SEED) in order to associate the following information with the individual pixel locations in the corresponding frame:

[0062] o The pixel location belongs to the second category when frame n corresponds to a subsequence of length greater than the threshold in the persistent subsequence to represent a pixel with a true Doppler signal; or

[0063] o The pixel location in frame n corresponds to another subsequence in the persistent subsequence with a length shorter than the threshold, so as to represent information that the artifact belongs to the first category.

[0064] The method is based on the persistence of the Doppler signal in n frames and the identification data, which, although belonging to the first value indicating the presence of Doppler activation, is considered an artifact and persists relatively weakly in n frames, and the identification data, which is considered to belong to the first value indicating the presence of Doppler activation, persists more persistently in n frames.

[0065] The step of generating Doppler-activated temporal persistence sequences is applied to portions of each frame. According to one embodiment, such a region is a single pixel of the frame, and as many persistence sequences as there are pixels in a video format frame (i.e., the pixel matrix constituting a frame in, for example, 800×566 format) are generated. Each persistence sequence contains identification data for each frame of the video. According to another embodiment, the portion of the frame referred to by the persistence sequence is a connected component with Doppler-activated pixels.

[0066] In this second case, specifically, it is necessary to perform a search algorithm on the connected components including pixels of the first category on frame n and frame n+1 to define the frame portion, and wherein the step of generating a persistent sequence includes the following steps: comparing a first parameter of a first connected component of frame n with a second parameter of a second connected component of frame n+1 and associating the second connected component with a first persistent sequence of the first connected component, or generating a second persistent sequence of the second connected component based on the comparison step, so as to obtain tracking of the connected components between frame n and frame n+1, wherein the parameters are preferably centroids and / or parameters representing the overlap of the first and second connected components and / or parameters representing the shape similarity of the first and second connected components and / or parameters representing the size or dimension of the first and second connected components.

[0067] In this way, the connection components enable the tracking or tracing of the Doppler activation region, and improve processing accuracy in the presence of Doppler signal regions that change during Doppler data acquisition.

[0068] According to a preferred embodiment, when the frame portion is pixels, the method includes the following steps: performing a segmentation algorithm on at least one nth frame based on a second image dataset (SEED) associated with the nth frame to expand the second image dataset using additional pixels containing Doppler signals, the additional pixels having previously had pixels representing artifacts removed. Specifically, a connection component study is also performed on the second dataset for each frame with respect to pixels representing true Doppler signals, and for each frame, the connection components of the video frame and the connection components of the true Doppler signal pixels overlap, and if an overlap threshold of, for example, 90%, has been defined, the connection components of video frames that pass the overlap test are retained, and the pixels of Doppler-activated connection components of video frames that fail the overlap test and are therefore considered artifacts are disabled, for example, modified.

[0069] In particular, it was found that the second image dataset tended to underestimate the range of representation regions for pixels with genuine Doppler signals (i.e., pixels excluding those associated with artifacts). To increase the processing accuracy of the original frames, a segmentation algorithm was therefore applied to expand the second image dataset to append additional pixels whose Doppler signals had been previously eliminated, as shown in the previous paragraphs, which were considered to represent artifacts.

[0070] According to a preferred embodiment, each acquired frame can be represented by a three-dimensional matrix, wherein the first dimension represents the number of pixels on the vertical axis, the second dimension represents the number of pixels on the horizontal axis, and the last third dimension represents the number of channels (R, G, B). Thus, each pixel is described by a triplet of values.

[0071] According to an embodiment of the present invention, the identification of color pixels representing pixels containing Doppler signals in step 2 of the denoising method can be achieved by dividing the R, G, B values ​​of the pixels constituting the frame into clusters and selecting those pixels that do not fall into the clusters of the identification ultrasound machine monochrome scale.

[0072] According to another embodiment of the present invention, the identification of color pixels in step 2 of the denoising method can also be achieved by selecting pixels whose difference between the intensities of the R and G, G and B, or R and B channels is greater than or equal to a predefined threshold.

[0073] According to a preferred embodiment, the threshold is calculated as the 90th percentile of the distribution, because the aim is to obtain the minimum activation length value of the Doppler signal, such that activation sequence lengths greater than the threshold are assumed to correspond to the true signal, while activation sequence lengths less than the threshold are assumed to be artifacts.

[0074] According to another preferred embodiment, when the minimum activation length of the Doppler signal is derived by tracking the temporal persistence of the closed and separated Doppler regions (i.e., connected components) contained within the frame, the threshold is calculated as the 98th percentile.

[0075] According to another embodiment, the thresholds for the two embodiments of the frame portion are calculated as the sum of the average value and the logarithmic double standard deviation of the length of the representative Doppler activation sequence to compensate for distribution asymmetry.

[0076] According to a preferred embodiment, in the step of locating the real signal, the visible result in the denoised frame may include Doppler signal pixels with the values ​​of the color channels [R, G, B] of the original frame, while other pixels will be assumed to fall into the grayscale representation as the average value of the three channels.

[0077] According to a preferred embodiment, in the region growth step, a morphological erosion method is used, followed by a morphological dilation method, with a kernel diameter or edge of 5 pixels and a threshold of 10%.

[0078] According to a preferred embodiment, there is a step of displaying the nth frame, wherein pixels corresponding to the real Doppler signal are retained, and pixels corresponding to artifacts are modified, for example, to present a predefined grayscale value so that they are not displayed as Doppler-activated pixels. Attached Figure Description

[0079] Preferred embodiments of the invention will now be described with reference to the accompanying drawings for illustrative purposes only, in which:

[0080] Figure 1 An example frame is shown extracted from echo Doppler video;

[0081] Figure 2 An example is shown that identifies the colored pixels on the left and the corresponding image dataset (e.g., the binary mask on the right);

[0082] Figure 3 An example of a SEED on the left relative to the frame on the right is shown;

[0083] Figure 4 The four sub-steps of step 7 of the method described in this invention are shown, wherein, if starting from the top left and proceeding clockwise, one can see the connection component identification, the result of the dilation operation, the color pixel calculation, and the final result of the frame at the end of the execution of the steps forming the method described in this invention.

[0084] Figures 5a to 5e The left side shows an example of a frame extracted from an ultrasound video, and the right side shows the corresponding frame at the end of a denoising process for certain types of artifacts detectable in ultrasound video, operated by the method of the present invention.

[0085] Figure 6 A schematic overview of multiple frames in a time series is shown, where the corresponding image datasets are associated with the respective frames;

[0086] Figure 7 An example of the temporal variation of signal and noise between two consecutive frames of echo Doppler video is shown; and

[0087] Figure 8 A flowchart of the noise reduction process is shown. Detailed Implementation

[0088] Figure 1 An example of a single frame belonging to echo Doppler video is shown, in which the presence of a genuine Doppler signal in the upper left image region and the presence of artifacts in the dark region in the lower part of the image, which in a specific case represents a cyst, can be identified.

[0089] In the following description, the steps included in the method for analyzing ultrasound video of the present invention will be analyzed in further detail with reference to examples of preferred embodiments, the method being designed to remove multiple artifacts that may be located to the side of and / or overlap with the real signal.

[0090] Echo Doppler video input acquisition

[0091] In this step, ultrasound video is acquired in the form of frame groups, or the video can be directly acquired and unpacked by the algorithm itself. Analysis of each frame by the algorithm leads to an evaluation of the continuous temporal persistence of all frames. Each frame can be represented as a three-dimensional matrix, where the first dimension represents the number of pixels on the vertical axis, the second dimension represents the number of pixels on the horizontal axis, and the third dimension represents the number of color channels [R, G, B]. Therefore, the value of each pixel can be described by triplets of values ​​in the range of 0 to 255.

[0092] Pixel recognition including Doppler signals

[0093] To separate the real Doppler signal from the noise caused by artifacts, the entire Doppler activation (real + artifact / noise) needs to be isolated within each frame for all frames included in the echo Doppler video. This operation results in the identification of two datasets for each frame: the first dataset contains the Doppler activations (color pixels), and the second dataset contains the ultrasound signal (grayscale or monochrome).

[0094] The union of the two sets of frames identified in the ultrasound video can be represented by a matrix of size n frames × n pixels on the vertical axis × n pixels on the horizontal axis.

[0095] Specifically, in this step, color pixels, i.e. pixels containing Doppler activation, are identified for each frame, and an image dataset is subsequently generated, in which pixels containing Doppler activation (color pixels) have a value of 1, while other pixels (grayscale or monochrome pixels) have a value of 0.

[0096] Colored pixels can be identified by selecting those pixels where the modulus of the difference between the intensities of the R and G, G and B, or R and B channels is greater than or equal to a predefined threshold. According to a preferred embodiment, this threshold is identified as |RG|>=30V, |RB|>=30V, and |GB|>=30. This is due to computational speed requirements and because it is not necessary to assign known colors to the colored pixels carrying the Doppler signal in order to identify which dataset the pixel in the examination falls into. An alternative to identification based on the difference in channel intensities is a machine learning method.

[0097] According to this method, all pixels in a frame are divided into separate sets by color, and then merged into two subsets: the first subset includes all colors found in the Doppler spectrum (colored pixels), while the second subset includes the color set specific to grayscale ultrasound. These operations are based on prior knowledge of the specific hues displayed by the Doppler signal in a given video. This color label assignment for each pixel can be accomplished either through unsupervised clustering operations or by using a classifier / neural network trained to extract the dominant colors within the image.

[0098] In one embodiment, the method includes:

[0099] (1) Initialize unsupervised clustering methods such as k-means to the distribution of pixel values ​​of pixels [R, G, B];

[0100] (2) Once the pixels are grouped into clusters, the centroid is calculated. The centroid is defined as the center of the cluster [R, G, B] triplet - the most representative triplet of the cluster.

[0101] (3) Assign color labels to each centroid by extending to all pixels that make up a given cluster; this operation can be done by calculating the distance (ciede2000) between each centroid and a known color label in the color space and selecting the known label that is closer to the centroid as the label.

[0102] (4) Place all pixels with color labels different from grayscale values ​​(including white and black) in the same set and place pixels with grayscale values ​​in another set.

[0103] These methods require computational power, and in the case of unsupervised methods, prior knowledge of the chromaticity that the Doppler signal can represent in the video is also needed for tag creation. Therefore, there is a dependency on the characteristics of the colors that the Doppler signal can represent in the video. Consequently, there is a dependency on the characteristics and settings of the ultrasound machine used, and on the type of Doppler, specifically color Doppler rather than power Doppler. This method is preferred when, for example, color information needs to be preserved when analyzing artifacts, such as to be able to observe pixel filtering that indicates the direction of approach (typically red).

[0104] Figure 2 The image illustrates an example of a color pixel identification step using an intensity difference recognition method. The left side of the image shows the frame being examined, while the right side shows the corresponding elements belonging to the image dataset, where bright areas corresponding to the identified color pixels in the frame and dark areas of other pixels are visible.

[0105] Activation sequence recognition (second category)

[0106] The underlying assumption of the method of this invention is that Doppler artifacts and noise have temporal persistence, that is, the duration expressed in frames is shorter than the true signal of the given echo Doppler video being examined.

[0107] To quantify the duration of the Doppler signal, the “Doppler activation” length of each pixel is calculated through two operations.

[0108] According to the first embodiment, the input to this step is a matrix of a binary image of size n (frames) × n pixels on the vertical axis × n pixels on the horizontal axis, generated in the previous step. First, for each pixel in the matrix, an activation vector is filled in, defined as a vector of length n (equal to the number of frames in the video) and a base of [0, 1] (a value adopted by each pixel depending on whether it is colored). For each frame, each vector element is assigned the corresponding pixel value of that frame. These values ​​correspond to first type of information (i.e., 0 for pixels without Doppler activation) and second type of information (i.e., 1 or "not 0" for pixels with Doppler activation). Figure 6 ).

[0109] Subsequently, for each vector, the activation sequence length (1 or "non-zero" continuous sequence - second category) and the number of frames in which each sequence persists (or disappears) Doppler without interruption are calculated.

[0110] Based on the application of these operations, the result can be described as a matrix of size n pixels on the vertical axis × n pixels on the horizontal axis, where each matrix element is a persistent vector of varying length. The reason for the different persistent vector lengths is due to the number of activations / deactivations detected for each pixel. Each vector element will be a pair of values ​​(count, value), where the value refers to the sequence value (0, 1), and the count refers to the number of frames in the sequence.

[0111] For example, assuming the video is divided into 20 frames, and considering a pixel with coordinates x1, y1, its activation vector is defined as follows:

[0112] [0,0,0,1,1,0,0,0,0,0,1,1,1,1,1,0,0,1,1,1].

[0113] The corresponding duration (or persistence) vector will have a length of 6 (equal to the number of activation / deactivation sequences in the activation vector) and will consist of the following pairs (count, value):

[0114] [(3,0),(2,1),(5,0),(5,1),(2,0),(3,1)]

[0115] According to another preferred embodiment, the Doppler signal duration can be achieved by tracking the temporal duration of closed and separated Doppler regions within a frame (i.e., the connected components of the Doppler regions) rather than placing them on a single pixel.

[0116] Within a frame of Doppler ultrasound video, regions defined by grayscale (or monochrome) pixels and regions defined by color pixels can be identified within the region of interest (ROI) that is the object of the ultrasound scan. Grayscale (or monochrome) pixels are referred to as non-Doppler pixels, representing Doppler signals, while color pixels are referred to as Doppler pixels, representing Doppler signals. Doppler pixels can also be distinguished based on whether they represent true Doppler activation or artifacts. Doppler pixels representing true Doppler activation are called true Doppler pixels, and those representing artifacts are called artifact Doppler pixels.

[0117] Connectivity refers to the identification of different objects in an image, where each of the objects has the property of being formed by a set of pixels that satisfy the same adjacency relationship (called connectivity).

[0118] Connectivity components within a frame can be identified using algorithms that can recognize the characteristics of pixels within a frame, label them, and determine whether they should be grouped into a single set to be represented as objects with specific shapes. These algorithms allow for the differentiation of pixels belonging to the true Doppler pixel category from those belonging to the artifact Doppler pixel category, for example, based on criteria such as shape, centroid distance, or overlap. Once connectivity components in a given frame have been identified, this family of algorithms allows them to be tracked in subsequent frames in a manner that distinguishes objects belonging to the same connectivity component, even if they become distorted or fragmented in subsequent frames. Furthermore, if it allows the identification of these connectivity components, artifacts are filtered out when they are no longer visible in subsequent frames.

[0119] The fundamental necessity stems from different implementations of Doppler ultrasound video. In the case of static video, the probe does not move, and the region of interest and vessels always cover the same areas along the frame sequence constituting the video. In this case, temporal persistence can be determined pixel-by-pixel by considering the same row and column values ​​in the binary image matrix of all frames of the video. Conversely, in the case where the probe moves during the acquisition phase, the background is no longer static along the frame sequence constituting the video, and individual pixels do not maintain a fixed correspondence between consecutive frames. In this case, temporal persistence can be performed by considering the connected components present along the frame sequence constituting the video.

[0120] Once the entire Doppler activation has been isolated (considering real and artifact / noise signals), the connected components are identified in each frame through two main steps, as described in the previous paragraph, “Pixel Identification Containing Doppler Signals.”

[0121] In the first initialization phase, all connected components (objects) located in the first non-empty binary mask associated with the corresponding frame of the video are identified.

[0122] For this mask, the centroid of the rectangular bounding box or the centroid of the connecting component itself is calculated for each identified component, and the vector V0 associated with the component is initialized with the following information:

[0123] - Unique identifier;

[0124] - The centroid of the rectangular bounding box or the connecting component itself;

[0125] - The coordinates of the rectangle that defines the component within the current mask;

[0126] - A vector V1 of size equal to the total length of the frame. When an object exists in the relevant frame, this vector will contain an indication of the coordinates (rows and columns) occupied by the object. In this initialization phase, the coordinates (rows and columns) occupied by the object in the current frame are stored as the first element.

[0127] - Boolean information B indicating whether the object is still being tracked. v0 It is initialized to mark the object as a value being tracked. Examples of values ​​are T, 1, Y, S, etc.

[0128] - Indicates the number of frames in which an object is no longer visible after its first appearance (I). v0 ;

[0129] - A vector V2 whose size is equal to the total length of the frames that make up the video, where each element, when set, indicates the presence of an object in the relevant frame.

[0130] refer to Figure 7 Assuming the image on the left represents the first non-empty binary mask, the initialization phase will detect the existence of four components within the mask: S(Tn), N1(Tn), N2(Tn), and N3(Tn). For each of these components, the initialization vector is denoted as vector V. s(Tn) V N1(Tn) V N2(Tn) V N3(Tn) It contains the information described above.

[0131] For each subsequent binary mask identified during the initialization phase, it should be determined whether an object existing in the current mask Tn+1 represents a new object, rather than being tracked as a moving object identified in a previous mask Tn. To achieve this, the following will be required:

[0132] - Locate all objects in the current mask Tn+1;

[0133] - For each object belonging to mask Tn+1, determine the centroid and coordinates of the rectangle that defines the components within mask Tn+1;

[0134] - For all possible object pairs between Tn and Tn+1, calculate the distance between centroids and the overlap between coordinates.

[0135] Once the above values ​​have been calculated, OT located in mask Tn+1 is determined if at least one of the following conditions is met. n+1 The object will be identified as an object in the previous mask Tn:

[0136] - There exists an object OT n Its relation to (OT) n OT n+1 The overlap value is at least equal to the predetermined threshold Ts;

[0137] - There exists an object OT n Its relation to (OT) n OT n+1 The distance between the coordinates of ) is at most equal to the threshold Td.

[0138] Additional comparison criteria can be used to determine whether an object in a given mask can be identified as the same object as in a previous mask, such as shape features and similarity to overlapping object masks. According to an embodiment of the algorithm, object OT... n With object OT n+1 The difference in pixel size is less than a given threshold Tm.

[0139] If the object is OT n+1 If an object is identified as the tracked object in the previous mask, then the object OT is used. n+1 Use the characteristics of [the system] to update the relevant initialization vectors:

[0140] - By adding objects containing OT n+1 Update vector V1 with the new element that occupies the coordinates (row and column) in the current frame;

[0141] - Update vector V2 by setting the element values ​​corresponding to the inspected frame.

[0142] If multiple objects in the current mask Tn+1 are identified as tracked objects from the previous mask, then update the V1 and V2 vectors for each object using the corresponding characteristics of those objects.

[0143] In object OT n+1 In the case of an object being identified as a non-tracked object in a previous mask, the new object is tracked by defining a new Vo vector in the same way as described during the initialization phase.

[0144] Objects that no longer have a match in binary mask Tn+1 are also updated as follows:

[0145] - Specifies the boolean property B for whether to track objects. v0 Set to indicate that the value of this object should no longer be tracked. Examples of values ​​are F, 0, N, etc.

[0146] - A gradual increase in the number of frames represented after an object is no longer being tracked. v0 Increase by 1. For a given number of frames, set an allowable threshold so that objects are tracked even if they disappear for a certain number of frames.

[0147] According to one embodiment, the permissible threshold is defined as considering the possibility that an object might disappear in a given number of frames and then reappear in subsequent frames. This threshold allows the event to be processed such that a temporarily missing object is treated as a single tracked object, rather than two separate objects. According to one embodiment, the threshold is 0, so the event of an object disappearing and reappearing throughout the frame sequence is treated as two separate objects.

[0148] Figure 7 An example of the temporal variation of signal and noise between two consecutive frames is shown.

[0149] Based on the previous description of the identification of connected components, and assuming that the binary mask Tn is the first non-empty mask, then:

[0150] - During the initialization phase, four vectors V01, V02, V03, and V04 will be defined and associated with objects S(Tn), N1(Tn), N2(Tn), and N3(Tn) respectively, and set to values ​​with their corresponding properties as described above;

[0151] - In the subsequent mask Tn+1, the signal degrades, resulting in the appearance of new objects [S1(Tn+1), S2(Tn+1), S3(Tn+1)] and the disappearance of others [N2(Tn), N3(Tn)] compared to the situation existing in Tn;

[0152] -At the end of tracking subsequent mask Tn+1:

[0153] Objects S1(Tn+1), S2(Tn+1), and S3(Tn+1) will be tracked as connected components of object S(Tn), and their properties will be updated on the vector V01 initialized at time T.

[0154] Object N1(Tn+1) should be tracked as the connected component of object N1(Tn), and its properties will be updated on the vector V02 initialized at time T.

[0155] Objects N2(Tn) and N3(Tn) that no longer have a match in mask Tn+1 will be updated to no longer be tracked.

[0156] Calculation of reference threshold for activation sequence length distribution

[0157] The purpose of this step is to obtain the minimum Doppler activation length value, such that activations longer than a threshold are considered to be genuine Doppler signals, while activations shorter than the threshold are associated with artifacts. This step begins with the n-pixel dimension set of the duration (or persistence) vector calculated in the previous step. However, in this step, only the activation length value will be considered, and the deactivation length will be ignored.

[0158] Continuing with the example defined in the previous step, we find a continuous vector relative to pixels x1, y1 along the 20 frames that make up the ultrasound video:

[0159] [(3,0),(2,1),(5,0),(5,1),(2,0),(3,1)].

[0160] Only the activation length will be considered.

[0161] [(2,1),(5,1),(3,1)],

[0162] And from these, the set of durations can be defined as:

[0163] [2,5,3].

[0164] This process is performed on all pixels that make up the frame. The reference threshold is calculated as the 90th percentile value of a normal distribution or (mean + 1.282 * standard deviation) over the distribution of all activation durations of all pixels.

[0165] When the minimum Doppler signal activation length is obtained by tracking the temporal duration of closed and separated Doppler regions within a frame, the threshold is calculated as the 98th percentile.

[0166] Real signal SEED positioning

[0167] Once the threshold is defined, the true signal coordinates, or Doppler signal seeds, must be located for each frame.

[0168] For each pixel, consider the vector of the previously calculated activation sequence and compare it with the threshold, as follows:

[0169] - Sequences with a length greater than or equal to the threshold are kept "on" (keeping the pixel value = 1 in the activation vector sequence);

[0170] - Shorter activation sequences are "disconnected" (by setting pixel values ​​to 0 for those activation sequences, even though "connection" does not reach the threshold considered to be a true Doppler signal).

[0171] Continuing with the example defined in the previous step, and assuming the previous step produced a threshold of 4, it can be seen that in the vector of activation lengths of pixels x1 and y1:

[0172] [(3,0),(2,1),(5,0),(5,1),(2,0),(3,1)]

[0173] Only sequences with a duration exceeding a threshold:

[0174] (5,1).

[0175] Therefore, the initial calculation was as follows

[0176] [0,0,0,1,1,0,0,0,0,0,1,1,1,1,1,0,0,1,1,1]

[0177] The activation vector will become at the end of this step.

[0178] [0,0,0,0,0,0,0,0,0,0,1,1,1,1,1,1,0,0,0,0,0],

[0179] As stated above.

[0180] At the end of this step, a matrix of size n frames × n pixels on the vertical axis × n pixels on the horizontal axis will be generated, where for each frame, each pixel will have a value of 1 if the relative activation sequence is at least equal to the threshold, and 0 otherwise.

[0181] Real signal positioning

[0182] This step considers a matrix of size n frames × n pixels on the vertical axis × n pixels on the horizontal axis, defined in the pixel identification step containing Doppler signals. For each frame in the matrix, we use the image dataset (e.g., a binary mask) defined in the previous step to isolate the real Doppler signal in the corresponding original video frame (to obtain a denoised frame). For each frame, the [R, G, B] values ​​of the pixels belonging to the original frame are preserved as the real signal by considering the pixel coordinates with a value of 1 in the image dataset obtained in the previous step and taking the corresponding triplet of the [R, G, B] channel values ​​for each coordinate in the original frame, or by multiplying the binary mask by the original frame.

[0183] A possible denoised frame view is that the Doppler signal pixels retain the [R, G, B] channel values ​​of the original frame and are therefore in color, while the other pixels take the average of the three [R, G, B] channel pixels in the original frame and are therefore grayscale.

[0184] Figure 3 On the left ( Figure 3 a) shows an example of a signal SEED for a given frame, which is obtained as an active sequence (second category pixels) that retains only pixels with a length at least equal to a threshold. Figure 3 On the right ( Figure 3 b) shows the application of SEED. Figure 1 The frame is generated from the original frame in the image. You can see the colored pixels corresponding to the real signal, while the artifacts have been masked with grayscale.

[0185] Segmentation and expansion

[0186] The previous real signal localization step returns pixel coordinates containing the real Doppler signal. However, the returned real signal area may not be correctly estimated, but rather underestimated. This also leads to the removal of pixels belonging to the real signal, for example, as a result of removing artifacts that overlap with the signal. Other examples that may involve the removal of pixels belonging to the signal might be due to operator probe movement, which causes peripheral pixels of blood vessels to become thinner in subsequent frames, or due to pulsation of the signal itself.

[0187] To accurately identify the edges of the signal regions identified during seed localization, starting from each seed (such as the signal region / coordinates identified at the end of the previous step), the concatenated Doppler activation components from the original video of the overlapping seeds are attached to the signal region / coordinates. For example, this operation is performed on each frame according to the following steps of the first segmentation algorithm:

[0188] a. Select the corresponding active image dataset, where a value of 1 represents a colored pixel, and a value of 0 represents other pixels (the result obtained in step 2 of this method). It should be noted that from this image dataset, the pixels corresponding to artifacts have not yet been "broken" ( Figure 2 b).

[0189] b. Apply a morphological erosion operation to the image dataset to separate as many of the major connected components of the true signal as possible from those associated with artifacts. In this embodiment, a circular kernel with a diameter of 5 pixels is used to apply the erosion ( Figure 4 a).

[0190] c. In the image dataset at point b, identify and label the connected components, i.e., the macroscopic components formed by n pixels greater than 1 that are adjacent and connected. Figure 4 a—connecting components with their colors) have connecting components with relative colors.

[0191] d. Selection based on image dataset Figure 4 The connection component of 'a' retains only the activation sequence of pixels with a length at least equal to the threshold in this image dataset (second category pixels— Figure 3 a). For example, based on and Figure 3 a's data and Figure 4 The data overlapping with e only retains the brightest area indicated by the arrow, because it is related to... Figure 3 The percentage of pixels whose active pixels overlap with those of a is above a predetermined threshold.

[0192] e. For pixels based on the second category ( Figure 3 a) For each selected connected component, apply a morphological dilation operation to the same extent as the previous erosion application (as a reverse operation to recover the erosion). In this embodiment ( Figure 4 b) The extension is applied using a circular core with a diameter of 5 pixels.

[0193] When an extended image needs to be displayed on the screen, from Figure 4 b. Begin performing the following steps:

[0194] For each extended connection component, count how many pixels (based on the pixels containing the real signal from the previous step) it contains that are filtered and survive.

[0195] If the number of colored pixels is greater than or equal to a given threshold, then in the new denoised frame, the color is expanded across the entire component, thereby assigning the corresponding [R, G, B] values ​​from the original frame to the pixels. In this embodiment, the optimal threshold is set to 10%.

[0196] According to an alternative embodiment, Figure 3 The region of a is expanded as follows (not shown):

[0197] Will Figure 3 Image data in a (second category) and related original frames ( Figure 1 Overlapping is used to define the seed;

[0198] Based on seeds, such as region growing, segmentation algorithms are applied to the original frame to identify... Figure 3 'a' represents the largest region of the seed.

[0199] Even with this alternative approach, a larger regional representation of the Doppler signal in the original frame can be obtained because artifacts, especially motion artifacts, have already been removed, and regions with underestimated Doppler spread have been identified. Figure 3 a.

[0200] Finally, Figure 5 illustrates some embodiments of the method described in this invention for removing artifacts from frames of ultrasound video by placing the original frame alongside its denoised frame.

[0201] The left side shows the frame before denoising was applied, and the right side shows the frame after denoising was applied. Note that color pixels considered "artifacts" have been replaced with grayscale pixels. These five videos were selected because they contain different types of artifacts, they show different anatomical objects within them, and they have different characteristics. In their original size, each frame is 800×566 pixels. In summary, color pixels were identified as those pixels that showed an intensity difference of at least 30 points between the R, G, and B channels. The selected activation threshold was 90% of the total activation length of each pixel in the video.

[0202] Figure 5a Example of removing flash artifacts. The video consists of 113 frames recorded at 57fps; the algorithm has set an acceptance threshold of 17 consecutive active frames.

[0203] Figure 5b Example of removing flash artifacts. The video consists of 70 frames recorded at 13fps; the algorithm has set an acceptance threshold of 11 consecutive active frames.

[0204] Figure 5c Example of removing flash artifacts. The video consists of 138 frames recorded at 57fps; the algorithm has set an acceptance threshold of 16 consecutive active frames.

[0205] Figure 5d Example of removing artifacts from a fake stream. The video consists of 499 frames recorded at 57fps; the algorithm has set an acceptance threshold of 19 consecutive active frames.

[0206] Figure 5eExample of random and blur artifact removal. The video consists of 229 frames recorded at 57fps; the algorithm has been set with an acceptance threshold of 14 consecutive active frames.

[0207] According to this embodiment, n frames with their corresponding pixels overlap, and are corresponding to the j-th pixel at the p×q position (height×width), because each frame has the same size, and is highly stable, for example, based on machine-executed ultrasound video.

[0208] Figure 8 A flowchart of a Doppler video artifact removal method is shown, in which the above sequence of steps can be seen, and it allows for obtaining videos with no artifacts or significantly reduced artifacts.

Claims

1. A computer-implemented denoising method for removing artifacts from Doppler ultrasound video, comprising the following steps: - Obtain the n frames that make up the ultrasound video; - For each of the n frames, a first pixel category and a second pixel category are identified, where the first pixel category represents the first pixel of the nth frame containing the ultrasound signal; The second pixel category represents the second pixel in the nth frame that contains Doppler activation; - For each of the n frames, generate a first image data set to associate each pixel position of the corresponding frame with at least a first pixel category or a second pixel category in order to classify the pixels of the corresponding frame and identify the pixels containing Doppler activation for each frame; - Generate at least a persistent sequence associated with a frame portion containing identification data ordered according to a frame sequence of n frames, such that adjacent identification data in the persistent sequence refer to consecutive frames, wherein each identification data uses a first value when the corresponding frame portion contains a Doppler-activated pixel, or uses a second value different from the first value when the corresponding frame portion does not contain a Doppler-activated pixel. - In the persistent sequence associated with the frame portion, the length of each persistent subsequence is calculated, the persistent subsequence comprising continuous identification data having the first value; - Calculate the reference threshold based on the length distribution of the persistent subsequence and referencing all the identification data of the n frames; - For each of the n frames, identify a second image dataset in order to associate the following information with each pixel location in the corresponding frame: - When frame n corresponds to a subsequence of the sustained subsequence with a length greater than the threshold, the information that the pixel location belongs to the second pixel category is used to represent a pixel with a true Doppler signal; or - When frame n corresponds to another subsequence of the persistent subsequence with a length shorter than the threshold, the information that the pixel location belongs to the first pixel category is used to indicate artifacts. - Modify the pixels corresponding to the artifacts over n frames. Wherein, the frame portion refers to the pixels of the frame.

2. The noise reduction method according to claim 1, wherein, The generation step includes generating multiple persistent sequences, each persistent sequence being used for each pixel of multiple frame formats.

3. The noise reduction method according to claim 1, comprising the following steps: An algorithm is executed to search for connected components including pixels of the first pixel category on frames n and n+1 to define the portion of the frame, wherein the step of generating a persistent sequence includes the following steps: comparing a first parameter of a first connected component of frame n with a second parameter of a second connected component of frame n+1 and associating the second connected component with a first persistent sequence of the first connected component, or generating a second persistent sequence of the second connected component based on the comparison step, so as to obtain tracking of the connected components between frame n and frame n+1, wherein the first parameter and the second parameter are preferably centroids and / or parameters representing the overlap of the first connected component and the second connected component and / or parameters representing the shape similarity of the first connected component and the second connected component and / or parameters representing the size or dimension of the first connected component and the second connected component, wherein the connected components are the identification of different objects present in the image.

4. The noise reduction method according to claim 2, wherein, The step of identifying the second image dataset is performed based on a search algorithm for connected components to identify regions representing the true Doppler signal, and includes the following steps: performing a segmentation algorithm based on a search for connected components of pixels of the second pixel category on at least nth frame of the ultrasound video; comparing the connected components of the second pixel category with the connected components representing the true Doppler signal based on an overlap criterion; and deactivating the connected components of the second pixel category that do not satisfy the overlap criterion to obtain a region growing effect of the connected components representing the true Doppler signal.

5. The noise reduction method according to claim 1, wherein, The step of identifying the second pixel category is performed according to multiple subcategories, one subcategory for each Doppler activation color present in the frame.

6. The noise reduction method according to claim 1, wherein, The step of identifying the second pixel category pixels is performed through clustering.

7. The noise reduction method according to claim 1, wherein, The step of identifying the second pixel category pixel is performed by selecting via the intensity differences of the R, G, and B channels.

8. The noise reduction method according to claim 2, wherein, The threshold calculation step is performed by calculating the 90th percentile of the length distribution of the subsequences of continuous identification data with the first representative Doppler activation value, or by summing the average length of the subsequences of continuous identification data with the first representative Doppler activation value to twice the standard deviation.

9. The noise reduction method according to claim 3, wherein, The step of calculating the threshold is performed by calculating the 98th percentile of the distribution of activation lengths or by summing the average length of the subsequences with the first representative value of Doppler activation based on continuous identification data to twice the standard deviation.

10. The noise reduction method according to claim 1, wherein, Thresholding is performed by color-labeling the individual pixels produced by the previous steps and using a classifier trained to extract the dominant colors found in the image.

11. The noise reduction method according to claim 1, wherein, The denoised frame has the same R, G, B channel values ​​as the original frame when represented in color, and has an average value when represented in grayscale.

12. The noise reduction method according to claim 4, wherein, The region growth is performed using a morphological erosion and expansion method with a kernel of 5 pixels in diameter or side length and a threshold of 10%.

13. The noise reduction method according to claim 1, wherein, The frame can be represented as a three-dimensional matrix.

14. The method according to any one of the preceding claims, comprising the step of displaying an nth frame, wherein pixels corresponding to the true Doppler signal are maintained and pixels corresponding to artifacts are modified in the nth frame.

Images