Systems, methods and devices for identifying heterogeneous liver fat

Abstract

An ultrasound imaging system collects images of the liver. The liver is segmented from the images. Parameters such as an image homogeneity map, an intensity probability chart, and / or a speckle size diagram are extracted from the liver portion of the images. These parameters are used to determine whether fatty liver deposits are diffuse or heterogeneous. In some embodiments, heterogeneous regions are excluded from the calculation of liver fat quantification measures. In some embodiments, the heterogeneous regions are displayed, allowing a user to select a region of interest that excludes the heterogeneous regions to calculate liver fat quantification measures.

Description

[Technical field]

[0001] FIELD OF THE DISCLOSURE The present disclosure relates to imaging systems and methods for assessing liver fat with ultrasound imaging, and in particular to imaging systems and methods for identifying regions of heterogeneous liver fat. [Background technology]

[0002] The prevalence of hepatic steatosis (fat content of 5% or more of the total liver) is increasing worldwide due to obesity and insulin resistance, which are risk factors for nonalcoholic fatty liver disease (NAFLD). Detection and quantification of liver fat is clinically important. For example, in the context of liver transplantation, successful liver transplantation requires that both donor and recipient have less than 10% liver fat. Some companies have developed and implemented liver fat quantification tools using either ultrasonic attenuation coefficient, ultrasonic elasticity point measurement, or two-dimensional (2D) or three-dimensional (3D) ultrasonic elastography imaging. To increase the accuracy and reliability of the measurement, the region of interest (ROI) for liver fat measurement is set to a rectangular or circular sector of M cm (height) × N cm (width) to increase the signal-to-noise ratio (SNR) from the relatively large measurement area, which is often a rectangular or circular sector. Typically, M is usually set to 8 cm, and N varies from 2 to 6 cm depending on the algorithm used for fat quantification. Current algorithms assume a diffuse (e.g., homogeneous) fat distribution in the liver. However, an estimated 10% of chronic liver disease patients have heterogeneous hepatic fatty infiltration. Fat quantification algorithms may provide inaccurate liver fat measurements for these patients. Therefore, for liver fat quantification, it is important to determine whether a chronic liver disease patient has diffuse or heterogeneous liver fat distribution in the ROI during the measurement workflow. Summary of the Invention

[0003] SUMMARY OF THE DISCLOSURE Systems, methods, and devices for identifying heterogeneous liver fat regions on ultrasound images are disclosed.

[0004] A series of 2D liver ultrasound images are collected, and liver regions are segmented from the 2D liver ultrasound images. After segmentation, liver vessels are detected and excluded from the segmented liver regions. One or more parameters are extracted from the segmented liver regions, such as image uniformity, distribution of pixel intensity, etc. The parameters are organized, such as in an image uniformity map, an intensity probability chart, and / or a speckle size diagram. In some embodiments, the parameters are provided to a classification model, which may be an artificial intelligence model (such as a deep learning model). The ROI is classified by the model into, for example, one of the following two cases: a first case indicating that the liver fat distribution in the ROI is uniform, or a second case indicating that the fat distribution in the ROI is non-uniform. The determined classification is reported as a result to a user of the ultrasound imaging system. In some embodiments, the segmented liver ultrasound images are provided to the user. The systems, methods, and devices disclosed herein provide an automatic or semi-automatic pre-processing tool in the liver fat quantification tool to facilitate obtaining accurate fat quantification measurements.

[0005] According to at least one embodiment of the present disclosure, an ultrasound imaging system provides a classification of hepatic fat deposition, the system including a non-transitory computer readable medium encoded with instructions, the non-transitory computer readable medium storing a sequence of images, and at least one processor in communication with the non-transitory computer readable medium executing the instructions, the instructions when executed cause the at least one processor to segment a liver region from at least one image of the sequence of images, extract at least one parameter from the liver region, classify hepatic fat deposition in the liver region as diffuse or heterogeneous based at least in part on the at least one parameter, and provide a result based at least in part on the classification of hepatic fat deposition.

[0006] In accordance with at least one embodiment of the present disclosure, a method for providing a classification of hepatic fat deposition includes receiving a series of images; segmenting, using at least one processor, a liver region from at least one image of the series of images; extracting, using the at least one processor, at least one parameter from the liver region; classifying, using the at least one processor, hepatic fat deposition in the liver region as diffuse or heterogeneous based at least in part on the at least one parameter; and providing a result based at least in part on the classification of hepatic fat deposition. [Brief description of the drawings]

[0007] [Figure 1] FIG. 1 is a block diagram of an ultrasound system in accordance with the principles of the present disclosure. [Diagram 2] FIG. 2 is a flow diagram of a method for classifying liver fat in accordance with the principles of the present disclosure. [Diagram 3] FIG. 3 illustrates a block diagram of a neural network training and deployment process in accordance with the principles of the present disclosure. [Figure 4] FIG. 4 shows an example of a two-dimensional image of a liver portion with homogenous fat distribution. [Diagram 5] FIG. 5 shows an example of a two-dimensional image of a liver portion having a non-uniform fat distribution. [Figure 6] FIG. 6 illustrates an example intensity probability distribution for a homogeneous fatty liver according to the principles of the present disclosure. [Figure 7] FIG. 6 illustrates an example intensity probability distribution of a heterogeneous fatty liver according to the principles of the present disclosure. [Figure 8] FIG. 8 shows example autocorrelation functions for homogeneous and heterogeneous fatty liver according to the principles of the present disclosure. [Figure 9] FIG. 9 is a table showing example parameters extracted from a two-dimensional image of the liver. [Figure 10] FIG. 10 is a flow chart of a method according to the principles of the present disclosure. [Figure 11]FIG. 11 is a block diagram illustrating an example processor 1100 in accordance with the principles of the present disclosure. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0008] The following description of specific embodiments is exemplary in nature and is not intended to limit the invention or its application or uses. In the following detailed description of embodiments of the present system and method, reference is made to the accompanying drawings, which form a part of this specification and which, by way of illustration, show specific embodiments in which the described system and method may be practiced. These embodiments have been described in sufficient detail to enable those skilled in the art to practice the disclosed apparatus, systems, and methods, and it is to be understood that other embodiments may be utilized and structural and logical changes may be made without departing from the spirit and scope of the present disclosure. Moreover, for the purposes of clarity, detailed descriptions of specific features will not be discussed if they would be apparent to those skilled in the art so as not to obscure the description of the present apparatus, systems, and methods. Therefore, the following detailed description is not to be taken in a limiting sense, and the scope of the present system is defined only by the appended claims.

[0009] Existing fat quantification algorithms typically provide acceptable estimates for liver fat quantification of diffuse / homogeneous liver fat deposition due to the relatively large regions used for averaging purposes to increase the signal-to-noise ratio (SNR). Liver fat quantification measurements (also referred to simply as measurements) are typically repeated multiple times on a selected ultrasound image by moving the location of the ROI or changing the size of the ROI. If the standard deviation (SD) of these measurements is significantly smaller than the mean value, the measurements are considered acceptable. If the SD value is large, the measurements are considered unacceptable. Large SD values ​​may be due to the operator (e.g., not placing the ultrasound probe at the proper transmission angle) or the patient (e.g., not holding their breath during image acquisition). However, in other cases, large SD values ​​may be due to heterogeneous fat deposition in the liver. It is difficult for an operator to determine how to determine whether the patient under examination has homogeneous or heterogeneous liver fat distribution in a particular region. Although a visual inspection before the measurement can be performed by an operator, the success depends on the operator's experience, and visual perception may be unreliable due to the inability of human vision to distinguish between gray scales. Therefore, a pre-processing step that can automatically determine the type of liver fat distribution (e.g., homogeneous or heterogeneous) of a patient in advance may be desirable.

[0010] By automatically determining whether a region of a patient's liver contains heterogeneous fat deposits, an operator may select a suitable ROI for liver fat quantification measurement (e.g., a region of homogeneous fat distribution) or may select a different algorithm for liver fat quantification measurement. In some applications, this may improve the reliability of the liver fat quantification measurement. In some applications, pre-identifying a suitable region for placing an ROI may reduce operator workload by allowing reliable measurements to be obtained in less time. In some applications, the presence and / or prevalence of heterogeneous liver fat distribution may itself be a diagnostic indicator of liver health.

[0011] The embodiments of the present disclosure include systems and methods for calculating various image parameters of a 2D liver ultrasound image. Examples of image parameters include, but are not limited to, a homogeneity map, speckle size, e.g., a probability distribution of image pixel intensities. The computer image parameters are provided to a machine learning model. The machine learning model is trained to determine whether one or more regions of the liver contain homogeneous (e.g., diffuse) or heterogeneous liver fat distribution. The ultrasound imaging system provides a decision indicator to a user of the ultrasound imaging system. For example, the ultrasound imaging system provides a visual or audio warning when a user selects an ROI containing heterogeneous liver fat distribution for liver fat quantification measurement. This prompts the user to select another region of the liver as an ROI for measurement. In another embodiment, the ultrasound imaging system provides an overlay on the 2D liver ultrasound image indicating regions containing homogeneous fat distribution and regions containing heterogeneous fat distribution. Thus, the user can see which regions of the 2D liver ultrasound image are suitable for selection as ROIs for liver fat quantification measurement.

[0012] Optionally, in some embodiments, the systems and methods include techniques for allowing a user to verify that the liver has been properly segmented in the 2D ultrasound image. Optionally, in some embodiments, the systems and methods include providing information on the 2D image (e.g., providing on a display the values ​​of the calculated image parameters, the shape of the segmented liver).

[0013] FIG. 1 illustrates a block diagram of an ultrasound imaging system 100 constructed in accordance with the principles of the present disclosure. The ultrasound imaging system 100 according to the present disclosure includes a transducer array 114 that may be included in an ultrasound probe 112 (e.g., an external probe or an internal probe). The transducer array 114 transmits ultrasound signals (e.g., beams, waves) and receives echoes in response to the ultrasound signals. A variety of transducer arrays may also be used, e.g., linear arrays, curved arrays, or phased arrays. For example, the transducer array 114 includes a two-dimensional array (as shown) of transducer elements that can be scanned in both elevation and azimuth dimensions for 2D and / or 3D imaging. As is commonly known, the axial direction is the direction perpendicular to the surface of the array (for curved arrays, the axial direction is fanned out), the azimuth direction is generally defined by the longitudinal dimension of the array, and the elevation direction is transverse to the azimuth direction.

[0014] In some embodiments, the transducer array 114 is mounted within the ultrasound probe 112 and is coupled to a microbeamformer 116 that controls the transmission and reception of signals by the transducer elements in the array 114. In some embodiments, the microbeamformer 116 controls the transmission and reception of signals by the active elements in the array 114 (e.g., an active subset of the elements of the array that define an active aperture at any given time).

[0015] In some embodiments, the microbeamformer 116 is coupled to a transmit / receive (T / R) switch 118, for example, by a probe cable or wirelessly. The T / R switch 118 switches between transmit and receive and protects the main beamformer 122 from high energy transmit signals. In some embodiments, for example in a portable ultrasound system, the T / R switch 118 and other elements in the system may be included in the ultrasound probe 112 rather than in an ultrasound system base that houses the image processing electronics. The ultrasound system base typically includes software and hardware components (e.g., processing circuitry 150 and user interface 124) that contain circuitry for signal processing and image data generation as well as executable instructions for providing a user interface.

[0016] The transmission of ultrasound signals from the transducer array 114 under the control of the microbeamformer 116 is directed by a transmit controller 120, which is coupled to the T / R switch 118 and the main beamformer 122. The transmit controller 120 controls the direction in which the beam is steered. The beam can be steered straight ahead from the transducer array 114 (orthogonal to the transducer array) or at a different angle to obtain a wider field of view. The transmit controller 120 is also coupled to a user interface 124 to receive input from a user manipulating user controls. The user interface 124 includes one or more input devices, such as a control panel 152. The control panel can include one or more mechanical controls (e.g., buttons, encoders), touch-sensitive controls (e.g., track pads, touch screens, etc.), and / or other known input devices.

[0017] In some embodiments, the partially beamformed signals generated by the microbeamformer 116 are coupled to a main beamformer 122, where the partially beamformed signals from individual patches of transducer elements are combined into a fully beamformed signal. In some embodiments, the microbeamformer 116 is omitted and the transducer array 114 is under the control of a main beamformer 122, which performs beamforming of all signals. In embodiments with and without the microbeamformer 116, the beamformed signals of the main beamformer 122 are coupled to processing circuitry 150. Processing circuitry 150 includes one or more processors (e.g., signal processor 126, B-mode processor 128, Doppler processor 160, and one or more image generation and processing components 168) that generate ultrasound images from the beamformed signals (e.g., beamformed RF data).

[0018] The signal processor 126 processes the received beamformed RF data in various ways, such as bandpass filtering, decimation, separation of I and Q components, and separation of harmonic signals. The signal processor 126 also performs additional signal enhancements, such as speckle reduction, signal combining, and noise removal. The processed signals (also referred to as I and Q components or IQ signals) are coupled to additional downstream signal processing circuits for image generation. The IQ signals are coupled to multiple signal paths in the system. Each signal path is associated with a particular arrangement of signal processing components suitable for generating different types of image data (e.g., B-mode image data, Doppler image data). For example, the system includes a B-mode signal path 158 that couples a signal from the signal processor 126 to a B-mode processor 128 to generate B-mode image data.

[0019] The B-mode processor can employ amplitude detection for imaging of structures within the body. Signals generated by the B-mode processor 128 are coupled to a scan converter 130 and / or a multi-planar reformatter 132. The scan converter 130 arranges the echo signals in the spatial relationship they were received in the desired image format. For example, the scan converter 130 arranges the echo signals in a two-dimensional (2D) sector-type format, or a pyramidal or other shaped three-dimensional (3D) format. As described, for example, in U.S. Pat. No. 6,443,896 (Detmer), the multi-planar reformatter 132 can convert echoes received from points in a common plane within a volumetric region of the body into an ultrasound image of that plane (e.g., a B-mode image). In some embodiments, the scan converter 130 and the multi-planar reformatter 132 are implemented as one or more processors.

[0020] The volume renderer 134 generates an image (also called a projection, render, or rendering) of the 3D dataset as viewed from a given reference point, as described, for example, in U.S. Patent No. 6,530,885 (Entrekin et al.). In some embodiments, the volume renderer 134 is implemented as one or more processors. The volume renderer 134 generates renders, such as positive or negative renderings, by any known or future known technique, such as surface rendering and maximum intensity rendering.

[0021] In some embodiments, the system includes a Doppler signal path 162 that couples the output from the signal processor 126 to a Doppler processor 160. The Doppler processor 160 estimates the Doppler shift and generates Doppler image data, including color data that is overlaid on B-mode (i.e., grayscale) image data for display. The Doppler processor 160 removes unwanted signals (i.e., noise or clutter associated with non-moving tissue), for example, using a wall filter. The Doppler processor 160 also estimates velocity and power according to known techniques. For example, the Doppler processor includes a Doppler estimator, such as an autocorrelator, where the velocity (Doppler frequency, spectral Doppler) estimate is based on the argument of the lag 1 autocorrelation function, and the Doppler power estimate is based on the magnitude of the lag 0 autocorrelation function. Motion is also estimated by known phase domain (e.g., parametric frequency estimators such as MUSIC, ESPRIT, etc.) or time domain (e.g., cross-correlation) signal processing techniques. The velocity and / or power estimates are then mapped to a desired range of display colors according to a color map. The color data (also referred to as Doppler image data) is coupled to a scan converter 130, which converts the Doppler image data to a desired image format and overlays it on a B-mode image of the tissue structure to form a color Doppler image or a power Doppler image.

[0022] Output from the scan converter 130, multiplanar reformatter 132, and / or volume renderer 134 are coupled to an image processor 136 for further enhancement, buffering, and temporary storage before being displayed on an image display 138. A graphics processor 140 generates graphic overlays that are displayed with the images. These graphic overlays may include standard identifying information such as the patient name, date and time of the image, imaging parameters, etc. To this end, the graphics processor 140 receives input from a user interface 124, such as a typed patient name or other annotations. The user interface 124 is also coupled to the multiplanar reformatter 132 for selection and control of the display of multiple multiplanar reformatted (MPR) images.

[0023] System 100 includes local memory 142. Local memory 142 may be implemented as any suitable non-transitory computer-readable medium (e.g., flash drive, disk drive). Local memory 142 stores data generated by system 100, such as ultrasound images, executable instructions, image parameters, training data sets, or any other information required for operation of system 100. In some embodiments, local memory 1242 includes multiple memories that may be of the same or different types. For example, local memory 142 may include dynamic random access memory (DRAM) and flash memory.

[0024] As previously discussed, the system 100 includes a user interface 124. The user interface 124 includes a display 138 and a control panel 152. The display 138 includes a display device implemented using various known display technologies, such as LCD, LED, OLED, or plasma display technologies. In some embodiments, the display 138 includes multiple displays. The control panel 152 receives user input (exam type, imaging parameters, etc.). The control panel 152 includes one or more hard controls (e.g., buttons, knobs, dials, encoders, mouse, trackball, etc.). In some embodiments, the control panel 152 may additionally or alternatively include soft controls (e.g., GUI control elements or simply GUI controls) provided to a touch-sensitive display. In some embodiments, the display 138 is a touch-sensitive display that includes one or more soft controls of the control panel 152.

[0025] In some embodiments, the various components shown in FIG. 1 are combined. For example, the multiplanar reformatter 132 and the volume renderer 134 are implemented as a single processor. In some embodiments, the various components shown in FIG. 1 are implemented as separate components. For example, the signal processor 126 is implemented as a separate signal processor for each imaging mode (e.g., B-mode, Doppler). In another example, the image processor 136 is implemented as separate processors for different tasks and / or for parallel processing of the same task. In some embodiments, one or more of the various processors shown in FIG. 1 are implemented by general-purpose processors and / or microprocessors that perform designated tasks. In some examples, the processor is configured to provide instructions for a task from a non-transitory computer-readable medium (e.g., from the local memory 142). The instructions are then executed by the processor. In some embodiments, one or more of the various processors are implemented as application-specific circuits. In some embodiments, one or more of the various processors (e.g., the image processor 136) are implemented with one or more graphical processing units (GPUs).

[0026] Figure 2 is a flow diagram of a method for classifying liver fat in accordance with the principles of the present disclosure. The method illustrated in flow diagram 200 is performed, at least in part, by an ultrasound imaging system, such as ultrasound imaging system 100 shown in Figure 1. Although flow diagram 200 is described with reference to ultrasound imaging system 100, this is for illustrative purposes only and the method illustrated in flow diagram 200 is not limited to being performed by ultrasound imaging system 100.

[0027] An imaging system, such as ultrasound imaging system 100, acquires and records (e.g., stores) a series of 2D images of the liver, as shown in step 202. For example, the 2D images are acquired with ultrasound probe 112. The 2D images may be stored in a non-transitory computer-readable medium, such as local memory 142. In some embodiments, the 2D images are of the same location of the liver acquired over time. In some embodiments, the 2D images are of different locations of the liver. That is, the 2D images may correspond to imaging planes located in different parts of the liver. The imaging planes may or may not overlap. The series of 2D images may cover a portion of the liver or the entire liver. Alternatively, only a single 2D image of the liver may be acquired.

[0028] As shown in step 204, an image processor, such as image processor 136, segments regions of one or more 2D images that contain liver tissue (e.g., liver regions) from the 2D images. Any segmentation method can be used, such as edge detection, gradient analysis, texture analysis, watershed algorithms, machine learning models, etc. Once the liver regions are segmented from the 2D images, the image processor analyzes the liver regions of each 2D image to detect liver vessels and other anatomical features (e.g., gallbladder) that should be excluded, as shown in step 206. Any suitable detection (e.g., feature extraction) method can be used, such as Frangi model, Hessian-based algorithms, Hough-forest-based detectors, machine learning models, etc. Pixels corresponding to detected liver vessels and / or other anatomical features are excluded from further analysis of the liver regions in the 2D images.

[0029] As shown in step 208, the image processor extracts (e.g., determines, calculates) one or more parameters from the liver regions of each 2D image (excluding regions corresponding to detected liver vessels and / or other anatomical features). Example parameters include speckle size distribution (e.g., characteristic speckle size), pixel intensity (e.g., probability) distribution, and image uniformity. In some embodiments, the parameters are extracted by analyzing one or more pixels within the liver regions of each 2D image. Exemplary techniques for extracting the parameters are disclosed herein.

[0030] The parameters are organized in various ways, such as an image uniformity map, an intensity probability chart, and / or a speckle size diagram. The parameters are provided to one or more machine learning models, such as a machine learning classification model shown at 210. In some embodiments, the machine learning models are implemented by an image processor. In some embodiments, the image processor executes instructions stored on a computer-readable medium to implement the machine learning models. The image processor implements any one or more deep learning, AI algorithms, and / or multiple neural networks (collectively, machine learning models). In some embodiments, the image processor implements one or more of a deep neural network (DNN), a convolutional neural network (CNN), a recurrent neural network (RNN), an autoencoder neural network, and / or a single-shot detector, etc., to identify and classify liver fat as homogeneous (e.g., diffuse) or heterogeneous. The machine learning models are implemented with hardware components (e.g., neurons of a neural network are represented by physical components) and / or software components (e.g., neurons and pathways are implemented with software applications). Neural networks implemented by the present disclosure use various topologies and learning algorithms to train the neural network to generate desired outputs. For example, software-based neural networks are implemented using a processor (e.g., a single-core or multi-core CPU, a single GPU or a GPU cluster, or multiple processors arranged for parallel processing) that executes instructions. The instructions are stored on a computer-readable medium and, when executed, cause the processor to execute the trained algorithm. In some examples, the image processor implements the machine learning model in combination with other image processing methods (e.g., segmentation, histogram analysis).

[0031] As shown in step 212, the results, such as the classification determined at 210 by the machine learning model, are reported. In some embodiments, the report is provided on a display, such as display 138. In some embodiments, the results are text indicating whether heterogeneous fat is present in the liver. In some embodiments, the results are provided graphically. For example, an overlay is generated and displayed on one or more of the 2D images. The overlay may include pixels corresponding to pixels of the liver region in the 2D images. Pixels of the regions determined to have diffuse liver fat have one or more display properties (e.g., a particular intensity or hue), while pixels of the regions determined to have heterogeneous liver fat have different display properties (e.g., different intensity or hue). Other reporting techniques (e.g., auditory notification by a speaker) may be used. In some embodiments, based on the reporting of the results, the user is guided to select an ROI for liver fat quantification measurement that does not include regions with heterogeneous fat deposits. In some embodiments, the user may select another algorithm or technique for obtaining liver fat quantification measurements adapted to heterogeneous fat deposits.

[0032] Optionally, in some embodiments, rather than analyzing or reporting results for the entire liver region, user input can be received via a user interface, such as user interface 124. The user input indicates an ROI within the liver region of one or more of the 2D images. The parameter extraction, classification, and / or reporting is limited to pixels within the ROI. In these embodiments, the reporting includes alerting the user when the selected ROI is determined to contain heterogeneous fat. The alert can be a graphical alert (e.g., a color change in the ROI or a box representing the ROI), a text alert (e.g., an on-screen alert message), and / or an alert via other media (e.g., an auditory alarm from a speaker, a vibration from a motor in a control panel). In some embodiments, the user can be prompted to select another ROI for obtaining liver fat quantification measurements.

[0033] Optionally, in some embodiments, rather than reporting a liver fat classification from the machine learning model to the user, the result is used by the image processor to calculate a liver fat quantification measurement. For example, the liver part containing heterogeneous liver fat indicated by the machine learning model is excluded from the liver region, similar to excluding liver blood vessels and / or other anatomical features. The remaining liver region is used to calculate a liver fat quantification measurement. For example, the image processor applies an algorithm that assumes a diffuse / uniform liver fat distribution to calculate the liver fat quantification measurement. In these embodiments, the result reported to the user is a liver fat quantification measurement (e.g., as text on a display), rather than a classification.

[0034] Optionally, in some embodiments, the segmented images generated in step 204 and / or step 206 are provided to a user on a display, allowing the user to verify that the segmentation operation was performed properly. In some embodiments, if the user is not satisfied with the segmentation performed by the image processor, the user can adjust the segmentation boundaries and / or other parameters (such as the seed points).

[0035] The technique described in flow diagram 200 is applied to RF data (such as data received before or after beamforming by beamformer 122) or minimally processed data (such as data received from signal processor 126) with reference to a 2D image (such as an image provided to the image processor from scan converter 130 and / or multiplanar reformatter 132).

[0036] Although the technique of flow diagram 200 has been described with reference to an image processor, in some embodiments, other processors may perform some or all of the technique. For example, a graphics processor, such as graphics processor 140, generates an overlay that is displayed on one or more 2D images. In some embodiments, the image processor may include one or more processors that perform tasks in parallel and / or are designated for different tasks. For example, one processor may segment liver regions from a series of 2D images, while another processor implements a machine learning model.

[0037] In various examples, one or more machine learning models used to classify hepatic fat deposition are trained using any of a variety of currently known or later developed learning techniques to obtain a trained model (e.g., a hardware-based system of trained algorithms or nodes), such as a neural network, that analyzes input data in the form of ultrasound images, parameters, measurements, and / or statistics. In some embodiments, the machine learning model is trained statically. That is, the machine learning model is trained using a data set, introduced to the imaging system, and implemented by one or more processors, such as image processor 136. In some examples, the machine learning model is trained dynamically. In these examples, the machine learning model is trained with an initial data set and introduced to the imaging system. However, the machine learning model continues to be trained and modified based on ultrasound images acquired by the system after the machine learning model is introduced to the system and implemented by the image processor.

[0038] FIG. 3 illustrates a block diagram of a neural network training and deployment process according to the principles of the present disclosure. The process illustrated in FIG. 3 is used to train a machine learning model implemented by a medical imaging system, such as the machine learning model illustrated in FIG. 2. The left side of FIG. 3, i.e., phase 1, illustrates training of the machine learning model. To train the machine learning model, a training set including multiple instances of input arrays and output classifications is presented to the machine learning model's training algorithm (e.g., the AlexNet training algorithm described in "ImageNet Classification with Deep Convolutional Neural Networks" by Krizhevsky, A., Sutskever, I., and Hinton, GE, NIPS 2012, or a descendant thereof). For example, the training set includes one or more parameters extracted from ultrasound images (e.g., speckle size, probability distribution, etc.) and corresponding classifications (e.g., homogeneous or heterogeneous liver fat, etc.).

[0039] Training involves the selection of a starting architecture 312 and the preparation of training data 314. The starting architecture 312 may be a blank architecture (e.g., an architecture with defined layers and node arrangements but no previously trained weights) or a partially trained model such as an Inception network that will later be further customized for classification of ultrasound data. The starting architecture 312 (e.g., blank weights) and training data 314 are provided to a training engine 310 (such as an ADAM optimizer) to train the model. Once a sufficient number of iterations have been performed (e.g., the model is performing consistently within acceptable error), the model 320 is said to be trained and ready for deployment. This is shown in the center of FIG. 3, i.e., phase 2.

[0040] On the right side of FIG. 3, i.e., in Phase 3, the trained model 320 is applied (via the inference engine 330) for the analysis of new data 332, which is data that was not presented to the model during the initial training (in Phase 1). For example, the new data 332 includes unknown data, such as live ultrasound images (e.g., liver images) collected during a patient scan, and / or parameters extracted from the unknown images. The trained model 320 implemented via the engine 330 is used to classify the unknown images according to the training of the model 320 to provide an output 334 (e.g., a determination of whether a pixel or group of pixels contains a homogeneous or heterogeneous fat distribution). The output 334 is used by the system for subsequent processes 340 (e.g., generating an overlay to display on the ultrasound image, calculating fat quantification measurements). Optionally, if the trained model 320 is dynamically trained (e.g., as described with reference to FIG. 1), the trained model 320 continues to train after implementation, as indicated by field training 338. In some embodiments, the new data 332 is used for field training of the trained model 320 .

[0041] In an embodiment where the trained model 320 is used to implement a neural network executed by a processor such as the image processor 136, the starting architecture is a convolutional neural network or a deep convolutional neural network starting architecture. This starting architecture is trained to determine whether a pixel or a region containing multiple pixels contains heterogeneous fat deposits or a homogeneous (e.g., diffuse) fat distribution. The training data 314 may include multiple (hundreds, often thousands or more) annotated / labeled parameters (also referred to as training data). It is understood that the training data need not be extracted from a complete image (e.g., representing the entire field of view or volume of the ultrasound probe) generated by the imaging system, but may include a patch or portion of the image, for example, a portion that includes an organ of interest or a portion of an organ of interest.

[0042] 4 shows example two-dimensional images of a liver portion having homogeneous fat distribution. Image 400 is an image of a portion of a liver that includes homogeneous (e.g., diffuse) fat distribution. The ROI is indicated by box 402. Image 404 is a "zoomed-in" view of the ROI of box 402. Image 406 is an image of a portion of a liver that includes homogeneous fat distribution. The ROI is indicated by box 408. Image 410 is a "zoomed-in" view of the ROI of box 408. Both images 400 and 402 are 2D ultrasound images acquired by an ultrasound imaging system.

[0043] 5 shows example two-dimensional images of a liver portion having non-uniform fat distribution. Image 500 is an image of a portion of a liver including non-uniform fat distribution. The ROI is indicated by box 502. Image 504 is a "zoomed-in" view of the ROI of box 502. Image 506 is an image of a portion of a liver including non-uniform fat distribution. The ROI is indicated by box 508. Image 510 is a "zoomed-in" view of the ROI of box 508. Both images 500 and 502 are 2D ultrasound images acquired by an ultrasound imaging system.

[0044] As described herein, parameters are extracted from the liver portion of an ultrasound image, such as images 400, 406, 500, and / or 506. The extracted parameters are then provided to a machine learning model to determine whether the liver portion contains homogeneous fat and / or heterogeneous fat. Example parameters include speckle size distribution (e.g., characteristic speckle size), pixel intensity and / or its distribution (e.g., probability distribution), and image uniformity, which are used to generate an image uniformity map, an intensity probability chart, and / or a speckle size distribution diagram.

[0045] The homogeneity map is calculated for the liver portion of the 2D image or a subset of the liver portion (e.g., the ROI of box 502 in image 500). In some embodiments, the image homogeneity map is calculated by a sliding window method. In image (i,j), where i is the number of pixels equal to the width of the image and j is the number of pixels equal to the length of the image, a window (e.g., 15×15 pixels) is translated across the image in both axial and lateral directions to cover the entire image or the entire liver region of the image. The mean (m) of the intensity values ​​of the pixels in the window and the standard deviation (SD) of the intensity values ​​of the pixels are calculated for each location of the window. The homogeneity point value is the standard deviation divided by the mean (SD / m). In some embodiments, the number of calculated homogeneity point values ​​is equal to the number of pixels in the image or the number of pixels in the liver region. In some embodiments, further parameters are extracted from the homogeneity map, such as the homogeneity index (HI). HI is the average of the uniformity map point values, or a subset of the uniformity map point values ​​(eg, a ROI within the uniformity map).

[0046] The machine learning model uses the homogeneity map and / or HI to classify (e.g., distinguish) between diffuse and heterogeneous fat distribution in the liver. For example, for the ROI shown in image 404, the HI is 9%, and for the ROI shown in image 410, the HI is 11%. In contrast, the HI for the ROI shown in image 504 is 39.8%, and for the ROI in image 510, the HI is 26.23%. Thus, the HI for heterogeneous liver fat is higher than the HI for diffuse liver fat.

[0047] In some embodiments, for the liver portion of the 2D image or a subset within the liver portion (e.g., the ROI of box 402 in image 400), one or more probability distributions of the grayscale intensity values ​​of the pixels of the image are calculated. In some embodiments, the probability distribution is calculated by plotting a histogram of the intensity values ​​of the pixels of the liver region and fitting a curve to the histogram to find a probability density function. In some embodiments, the probability density function is normalized. In some embodiments, additional parameters are extracted from the probability density function, for example, a number of peaks and / or a width of the distribution are extracted.

[0048] FIG. 6 shows an example of a homogeneous fatty liver intensity probability distribution, and FIG. 7 shows an example of a heterogeneous (e.g., inhomogeneous) fatty liver intensity probability distribution according to the principles of the present disclosure. Plots 600, 602, 604, and 606 show histograms 608, 610, 612, and 614 of pixel intensities of an image of a homogeneous fatty liver. To find the distribution of intensities, curves 618, 620, 622, and 624 are fitted to each histogram 608, 610, 612, and 614. That is, the distribution defined by curves 618, 620, 622, and 624 shows the probability of a pixel in the image having a given intensity. Similarly, plots 700, 702, 704, and 706 show histograms 708, 710, 712, and 714 of pixel intensities of an image of a heterogeneous fatty liver. To find the intensity distribution of the pixels, curves 718, 720, 722, and 724 are fitted to the histograms. The probability distributions shown in plots 600, 602, 604, and 606 of FIG. 6 are normalized distributions, while the probability distributions shown in plots 700, 702, 704, and 706 of FIG. 7 are non-normalized distributions, some with multiple peaks. Furthermore, the distribution width of the heterogeneous fatty liver in FIG. 7 is wider than that of the diffuse fatty liver in FIG. 6. Thus, in some embodiments, the machine learning model uses the probability distribution (e.g., the curve fitted to the histogram of pixel intensities) instead of or in addition to the heterogeneity map to classify the diffuse and heterogeneous fat distributions in the liver.

[0049] In some embodiments, the speckle size map is extracted from a liver portion of the 2D image or a subset within the liver portion (e.g., a ROI). In some embodiments, the speckle size map is generated by taking an autocorrelation of the liver portion of the image. FIG. 8 illustrates example autocorrelation functions for homogeneous and heterogeneous fatty liver in accordance with the principles of the present disclosure. Plot 800 illustrates the autocorrelation function of a liver portion of an ultrasound image of a liver with homogeneous fat distribution. Plot 802 illustrates the autocorrelation function of a liver portion of an ultrasound image of a liver with heterogeneous fat distribution. In some embodiments, the characteristic speckle size is calculated by measuring the height and / or width of the peak of the autocorrelation function. In some embodiments, the characteristic speckle size is calculated by taking the ratio of the height and width. The size of speckle 804 in plot 800 is smaller than speckle 806 in plot 802. Thus, the machine learning model uses speckle size distributions, such as speckle size distributions determined by taking the autocorrelation function, instead of or in addition to image inhomogeneity maps and / or intensity probability charts to classify diffuse and heterogeneous fat distribution in the liver.

[0050] The speckle size map is system dependent in some applications. In some embodiments, the speckle size map varies depending on the type of ultrasound probe used and / or image acquisition settings. Thus, in some embodiments, the autocorrelation function and / or the speckle size distribution derived therefrom is normalized before being provided to the machine learning model. In some embodiments, the speckle size distribution is normalized based on speckle measurements from a standardized reference phantom. In some embodiments, the speckle size distribution is normalized based on the point spread function of the ultrasound imaging system.

[0051] Although image heterogeneity maps, intensity probability distribution charts, and speckle size diagrams are shown as examples, other parameters may be extracted from the 2D ultrasound images to provide to the machine learning model for classifying diffuse and heterogeneous liver fat. Additionally, although the machine learning model is described as classifying liver fat as homogeneous or heterogeneous, in some embodiments, the machine learning model may perform more detailed classification. FIG. 9 is a table showing example parameters extracted from a 2D image of the liver. As shown in FIG. 9, in the column labeled "description," there are various types of homogeneous and heterogeneous liver fat deposits. Thus, in some embodiments, the machine learning model uses parameters extracted from the ultrasound images to classify fat deposits as homogeneous or heterogeneous, as well as specific subtypes of homogeneous or heterogeneous liver fat.

[0052] 10 is a flow chart of a method according to the principles of the present disclosure. In some embodiments, the method 1000 is performed in whole or in part by an ultrasound imaging system, such as the ultrasound imaging system 100 shown in FIG 1. In some embodiments, the method 1000 includes some or all of the techniques shown in FIG 2.

[0053] As shown in step 1002, one or more processors, such as image processor 136, receive the series of images. As shown in step 1004, the processor segments a liver region from at least one image of the series of images. Optionally, in some embodiments, the segmented liver region is provided on a display for review by a user. If desired, the user can modify the liver segmentation.

[0054] As shown in step 1006, the one or more processors extract at least one parameter from the liver region. In some embodiments, the one or more processors identify at least one liver blood vessel or other anatomical feature in the liver region and exclude the identified liver blood vessel or other anatomical feature from the liver region before extracting the at least one parameter. In some embodiments, the at least one parameter includes an inhomogeneity map, a probability distribution, and / or a speckle size.

[0055] In some embodiments, extracting the heterogeneity map includes translating a window across the liver region in both axial and lateral directions, calculating the average value of the intensity values ​​of the pixels in the window for each translation of the window, calculating the standard deviation of the intensity values ​​of the pixels in the window for each translation of the window, and dividing the standard deviation by the average value to calculate a uniformity point value for each translation of the window. Optionally, a uniformity index (HI) may be extracted from the map. The HI is the average value of the uniformity points. In some embodiments, extracting the intensity probability chart includes plotting the intensity values ​​of the pixels of the liver region in a histogram and fitting a curve to the histogram. Optionally, the width of the probability distribution may be calculated from the curve. In some embodiments, extracting the speckle size diagram includes calculating an autocorrelation function of the liver region.

[0056] As shown in step 1008, the at least one processor classifies hepatic fat deposition in the liver region as diffuse or heterogeneous based at least in part on the at least one parameter. In some embodiments, the classification is performed by a machine learning model, such as a neural network, implemented by the at least one processor. In some embodiments, the method 1000 further includes training the machine learning model to classify hepatic fat deposition by providing a training dataset that includes parameters extracted from liver regions of the images labeled as either diffuse or heterogeneous.

[0057] As shown in step 1010, at least one processor and / or other components of the imaging system (e.g., display 138, user interface 124, etc.) provide a result based at least in part on the classification of liver fat deposition. In some embodiments, the result is provided to the user by displaying a graphic overlay over the image or a liver region of the image. The graphic overlay includes a plurality of pixels corresponding to a plurality of pixels of the liver region. A property of a pixel of the plurality of pixels of the graphic overlay is based at least in part on the classification.

[0058] Figure 11 is a block diagram illustrating an example processor 1100 in accordance with the principles of the present disclosure. Processor 1100 may be used to implement one or more of the processors and / or controllers described herein, such as image processor 136 shown in Figure 1 and / or any other processor or controller shown in Figure 1. Processor 1100 may be any suitable processor type, including, but not limited to, a microprocessor, a microcontroller, a digital signal processor (DSP), a field programmable array (FPGA) (wherein the FPGA is programmed to form a processor), a graphics processing unit (GPU), an application specific circuit (ASIC) (wherein the ASIC is designed to form a processor), or a combination thereof.

[0059] The processor 1100 may include one or more cores 1102. The cores 1102 may include one or more arithmetic logic units (ALUs) 1104. In some embodiments, the cores 1102 include a floating point logic unit (FPLU) 1106 and / or a digital signal processing unit (DSPU) 1108 in addition to or instead of the ALUs 1104.

[0060] The processor 1100 may include one or more registers 1112 communicatively coupled to the core 1102. The registers 1112 may be implemented using dedicated logic gate circuits (e.g., flip-flops) and / or any memory technology. In some embodiments, the registers 1112 are implemented using static memory. The registers may provide data, instructions, and addresses to the core 1102.

[0061] In some embodiments, the processor 1100 includes one or more levels of cache memory 1110 communicatively coupled to the cores 1102. The cache memory 1110 can provide computer readable instructions to the cores 1102 for execution. The cache memory 1110 can provide data for processing by the cores 1102. In some embodiments, the computer readable instructions may be provided to the cache memory 1110 by a local memory (e.g., a local memory connected to an external bus 1116). The cache memory 1110 can be implemented with any suitable cache memory type, e.g., metal-oxide-semiconductor (MOS) memory, such as, for example, static random access memory (SRAM), dynamic random access memory (DRAM), and / or any other suitable memory technique.

[0062] The processor 1100 includes a controller 1114. The controller 1114 controls input to the processor 1100 from other processors and / or components included in the system (e.g., the control panel 152 and the scan converter 130 shown in FIG. 1) and / or output from the processor 1100 to other processors and / or components included in the system (e.g., the display 138 and the volume renderer 134 shown in FIG. 1). The controller 1114 controls data paths within the ALU 1104, the FPLU 1106, and / or the DSPU 1108. The controller 1114 is implemented as one or more state machines, data paths, and / or dedicated control logic. The gates of the controller 1114 are implemented as stand-alone gates, FPGAs, ASICs, or any other suitable technology.

[0063] The registers 1112 and cache memory 1110 may communicate with the controller 1114 and the core 1102 via internal connections 1120A, 1120B, 1120C, and 1120D. The internal connections may be implemented as a bus, a multiplexer, a crossbar switch, and / or any other suitable connection technology.

[0064] Input and output for the processor 1100 may be provided via a bus 1116. The bus 1116 may include one or more conductive lines. The bus 1116 may be communicatively coupled to one or more components of the processor 1100, such as, for example, the controller 1114, the cache memory 1110, and / or the registers 1112. The bus 1116 may be coupled to one or more components of the system, such as the display 138 and the control panel 152 discussed above.

[0065] The bus 1116 may be coupled to one or more external memories. The external memory may include a read only memory (ROM) 1132. The ROM 1132 may be a masked ROM, an electronically programmable read only memory (EPROM), or any other suitable technology. The external memory may include a random access memory (RAM) 1133. The RAM 1133 may be a static RAM, a battery backed up static RAM, a dynamic RAM (DRAM), or any other suitable technology. The external memory may include an electrically erasable programmable read only memory (EEPROM) 1135. The external memory may include a flash memory 1134. The external memory may include a magnetic storage device such as a disk 1136. In some embodiments, the external memory is included within a system such as the ultrasound imaging system 100 shown in FIG. 1, for example, as local memory 142.

[0066] The systems, methods, and devices disclosed herein mitigate the effects of heterogeneous fatty liver deposits that interfere with the liver fat quantification measurement calculation algorithm by determining which regions of the liver contain heterogeneous fatty liver deposits. In some embodiments, these regions are excluded from the calculation of the fat quantification measurement. In some embodiments, the user is alerted to the regions of heterogeneous liver fat, which allows the user to select to place an ROI for calculating the liver fat quantification measurement in an area without heterogeneous liver fat. Excluding regions of heterogeneous fat distribution provides a more consistent and reliable measurement for liver fat quantification in some applications. In some applications, providing the user with information about where the heterogeneous liver fat deposits are located reduces the time required for the user to obtain the liver fat quantification measurement.

[0067] Although the examples described herein refer to an ultrasound imaging system, analysis of the images may also be performed by another computing system that stores the ultrasound images for review after the ultrasound examination, such as a Picture Archiving and Communication System (PACS). The PACS may be communicatively coupled to the ultrasound imaging system by a wired or wireless connection. For example, the PACS receives ultrasound images acquired by the imaging system via an Ethernet cable and / or Wi-Fi. Additionally, while the examples described herein describe processing of ultrasound image data, it is understood that the principles of the present disclosure are not limited to ultrasound, but may be applied to image data from other modalities, such as magnetic resonance imaging and computed tomography.

[0068] In various embodiments in which the components, systems, and / or methods are implemented using a programmable device, such as a computer-based system or programmable logic, it should be understood that the aforementioned systems and methods can be implemented using any of a variety of known or later developed programming languages, such as "C", "C++", "C#", "Java", "Python", and the like. Accordingly, various storage media, such as magnetic computer disks, optical disks, electronic memory, and the like, can be provided that can contain information capable of instructing a device, such as a computer, to implement the above-described systems and / or methods. When the appropriate device is able to access the information and programs contained therein, the storage media can provide the information and programs to the device, thereby enabling the device to perform the functions of the systems and / or methods described herein. For example, when a computer is provided with a computer disk containing appropriate material, such as source files, object files, executable files, and the like, the computer can receive the information, appropriately configure itself, and perform the functions of the various systems and methods described in the figures and flow charts above to perform the above-described functions. That is, the computer can receive various portions of information related to the various elements of the above-described systems and / or methods from the disk, and can implement and coordinate the functions of the individual systems and / or methods described above.

[0069] In view of the present disclosure, it should be noted that the various methods and devices described herein can be implemented in hardware, software, and firmware. Moreover, the various methods and parameters are included by way of example only, and not by way of limitation. In view of the present disclosure, one skilled in the art can implement the present teachings to determine his or her own methods and the equipment necessary to effect those methods while remaining within the scope of the present invention. One or more functions of the processors described herein can be incorporated into a few or a single processor unit (e.g., a CPU), and can be implemented using an application specific integrated circuit (ASIC) or general purpose processing circuitry programmed to respond to executable instructions to perform the functions described herein.

[0070] Although the present system has been described with particular reference to an ultrasound imaging system, it is envisioned that the present system can be extended to other medical imaging systems in which one or more images are systematically acquired. Additionally, the present system also includes one or more programs that can be used with conventional imaging systems, thereby enabling the conventional imaging systems to provide the features and advantages of the present system. Certain additional advantages and features of the present disclosure will be apparent to those of skill in the art after reviewing the present disclosure, or will be experienced by those who employ the novel systems and methods of the present disclosure. Another advantage of the present systems and methods is that conventional medical imaging systems can be easily upgraded to incorporate the features and advantages of the present systems, devices, and methods.

[0071] Of course, it is to be understood that any one of the examples, embodiments, or processes described herein can be combined with one or more other examples, embodiments, and / or processes, or separated and / or implemented among separate devices or device parts, in accordance with the present systems, devices, and methods.

[0072] Finally, the above discussion is merely illustrative of the present system, and should not be construed as limiting the appended claims to any particular embodiment or group of embodiments. Thus, while the present system has been described in particular detail with reference to exemplary embodiments, it should also be understood that those skilled in the art may devise numerous modifications and alternative embodiments without departing from the broader intended spirit and scope of the present system, as set forth in the following claims. Accordingly, the specification and drawings are to be regarded as illustrative, and are not intended to limit the scope of the appended claims.

Claims

1. 1. An ultrasound imaging system for providing a classification of hepatic fatty deposits, the ultrasound imaging system comprising: a non-transitory computer readable medium encoded with instructions, the non-transitory computer readable medium storing a sequence of ultrasound images; at least one processor in communication with the non-transitory computer-readable medium and for executing the instructions; Including, The instructions, when executed, cause the at least one processor to: Segmenting a liver region from at least one image of the series of images; extracting at least one parameter from the liver region; classifying hepatic fat deposition in the liver region as diffuse or heterogeneous based at least in part on the at least one parameter; providing a result based at least in part on said hepatic steatosis classification; The ultrasound imaging system includes:

2. 2. The ultrasound imaging system of claim 1, wherein the at least one processor further identifies at least one hepatic blood vessel or other anatomical feature in the liver region and excludes the identified hepatic blood vessel or other anatomical feature from the liver region prior to extraction of the at least one parameter.

3. 2. The ultrasound imaging system of claim 1, further comprising a display, and wherein the results include a graphic overlay for display over the at least one image on the display, the graphic overlay including a plurality of pixels corresponding to a plurality of pixels of the liver region, and a property of a pixel of the plurality of pixels of the graphic overlay is based at least in part on the classification.

4. 2. The ultrasound imaging system of claim 1, further comprising a display and a user interface, the user interface receiving user input indicating a region of interest (ROI) within the liver region, and the at least one processor further providing display information for a warning if the ROI contains the hepatic fat deposits classified as heterogeneous.

5. The ultrasound imaging system of claim 1 , wherein the at least one processor implements a machine learning model to analyze the at least one parameter and generate the classification of the liver fatty deposits.

6. The ultrasound imaging system of claim 5 , wherein the machine learning model comprises a neural network.

7. The ultrasound imaging system of claim 1 , wherein the at least one parameter includes at least one of an inhomogeneity map, a distribution of pixel intensity, or a speckle size.

8. The ultrasound imaging system of claim 1 , wherein the at least one processor is further configured to calculate a fat quantification measurement based at least in part on the classification of the liver fat deposition.

9. The ultrasound imaging system of claim 8 , wherein the fat quantification measurements are based on the liver region excluding the hepatic fat deposits classified as heterogeneous.

10. 1. A method for providing a classification of hepatic fat deposition, comprising: receiving a series of ultrasound images; segmenting, with at least one processor, a liver region from at least one image of the series of images; extracting, with the at least one processor, at least one parameter from the liver region; classifying, with the at least one processor, hepatic fat deposition in the liver region as diffuse or heterogeneous based at least in part on the at least one parameter; providing a result based at least in part on said liver fat classification; A method comprising:

11. identifying at least one hepatic vasculature or other anatomical feature of the liver region; excluding the identified liver vessels or other anatomical features from the liver region prior to extracting the at least one parameter; The method of claim 10 further comprising:

12. The method of claim 10 , wherein the at least one parameter comprises at least one of an inhomogeneity map, a probability distribution of pixel intensities, or a speckle size.

13. The step of extracting the inhomogeneity map comprises: translating a window across the liver region in both axial and lateral directions; for each translation of said window, calculating an average of the intensity values ​​of pixels within said window; calculating, for each translation of the window, a standard deviation of the intensity values ​​of the pixels within the window; calculating a uniformity point value for each translation of the window by dividing the standard deviation by the mean value; The method of claim 12, comprising:

14. The method of claim 13 , further comprising the step of extracting a uniformity index, said uniformity index comprising an average value of said uniformity point values ​​for said respective translations of said window.

15. The step of extracting a probability distribution comprises: plotting the intensity values ​​of the pixels of the liver region in a histogram; fitting a curve to the histogram; The method of claim 12, comprising:

16. The method of claim 15 , further comprising calculating a width of the probability distribution based at least in part on the curve.

17. The method of claim 12 , wherein extracting the speckle size comprises calculating an autocorrelation function of the liver region.

18. The method of claim 10 , wherein the classifying step is performed by a machine learning model.

19. 20. The method of claim 18, further comprising training the machine learning model to classify the hepatic fatty deposits by providing a training dataset comprising parameters extracted from liver regions of images labeled as either diffuse or heterogeneous.

20. 11. The method of claim 10, wherein the step of providing a result comprises displaying a graphic overlay including a plurality of pixels corresponding to a plurality of pixels of the liver region, wherein a property of a pixel of the plurality of pixels of the graphic overlay is based at least in part on the classification.

Images