System and method for evaluation of liver steatosis

Abstract

The present invention refers to a system and a method, based on Artificial Intelligence, which automate the estimation of hepatic steatosis on ultrasound images, calculating a quantitative index of hepatic steatosis called KLI (Kidney to Liver Index). The system comprises an ultrasound device connected with processing means designed to implement a method which analyzes the ultrasound images and obtains a quantitative index of said fatty liver disease (KLI - Kidney to Liver Index) through the estimation of the percentage of fat accumulated in the liver, and performs on the ultrasound image a semantic segmentation using an acyclic neural network based on the resnet50 model, from which, through subsequent processing, a degree of hepatic steatosis is defined.

Description

[0001] SYSTEM AND METHOD FOR EVALUATION OF LIVER STEATOSIS

[0002] The present invention refers to a system and a method, based on Arti ficial Intelligence , which automate the estimation of hepatic steatosis on ultrasound images , calculating a quantitative index of hepatic steatosis called KLI (Kidney to Liver Index ) , said calculation being performed through an estimate of both the degree of hepatic steatosis and the percentage of fat accumulated in the liver .

[0003] Hepatic steatosis is a pathology whose dif fusion is estimated between 25% and 44 % of the world population and is characteri zed by a growth trend . The use of non-invasive diagnostic techniques is increasingly required .

[0004] The diagnosis of liver diseases is frequently performed through ultrasound as this procedure is characteri zed by non-invasiveness , speed of analysis and calculation .

[0005] In particular, in the last two years , the main ultrasound manufacturers in the world have enriched their software tools useful for performing non- invasive assessments of liver health status . An advanced tool based on Arti ficial Intelligence that allows greater calculation speed, less dependence on the human operator with greater sensitivity / speci f icity, without the need for hardware modi fications of current machinery, can potentially af fect the entire world market of ultrasound, in any price range .

[0006] The current ultrasound techniques for evaluating steatosis are strongly linked to the operator performing the ultrasound and provide values with very high intra-operator and interoperator variability . Furthermore , cases of mild steatosis are di f ficult to diagnose .

[0007] In particular, there are currently no totally automatic methodologies for assessing hepatic steatosis based on the analysis of a single ultrasound proj ection image .

[0008] Although qualitative US (ultrasound) has limitations in the evaluation of fatty liver disease , it may be a good option in many clinical settings for an initial qualitative and quantitative evaluation of a fatty liver . The main advantages , compared to other modalities , are low cost , portability and lack of radiation exposure . Unlike other modalities , whose lack of portability limits their use , the portability of ultrasound scanners of fers a distinct advantage in clinical settings and in large epidemiological studies , which explains the growing importance of ultrasound scanners in the diagnosis of liver disease . Because no radiation is needed, US can be used for all patients , including children and pregnant women . Another important advantage of US is that they allow real-time visuali zation of the target structure and, due to their echogenicity, they can provide information on the presence of inflammation, fibrosis and lipid accumulation in the liver .

[0009] The main limitation of the US is the lack of a standardi zed protocol and the dependence on the examiner, which can lead to evaluation errors and thus interfere with the reproducibility of the results . For thi s reason, it is of great interest to establish a new index of fatty liver disease to assess not only the presence of fatty liver disease but also the degree of steatosis . To achieve this , a new Arti ficial Intelligence (A. I . ) -based technique with high sensitivity and validated with a reference standard quantitative technique is needed .

[0010] US patent application 2020 / 0367853 discloses a system based on I . A. for the qualitative assessment of the degree of steatosis (Normal , Mild, Medium, Severe ) . It is a system focused on the global evaluation of the degree of steatosis which, however, does not carry out quantitative evaluations .

[0011] US patent application 2022 / 0237798 describes an automatic system for calculating a hepato-renal index .

[0012] Patent application WO 2022 / 214410 describes an ultrasound machine equipped with tools that help a physician in determining the ROI (Region Of Interest ) useful for calculating the hepato-renal index . This is not a fully automatic evaluation and the same ratios / values or estimates presented in the present invention are not calculated .

[0013] Obj ect of the present invention is to at least partially overcome the drawbacks complained of through a method and a system, respectively in accordance with claims 1 and 11 , suitable for automating the estimation of hepatic steatosis starting from ultrasound images .

[0014] The system, adapted to automate the estimation of hepatic steatosis starting from ultrasound images , is of the type comprising an ultrasound machine , provided with a probe and connected to processing means designed to implement the following inventive method .

[0015] This method, based on Arti ficial Intelligence , is of the type that makes use of said ultrasound, equipped with said means suitable for analyzing the ultrasound images to obtain a quantitative index o f said hepatic steatosis (KLi - Kidney to Liver Index ) through the estimation of the percentage of fat accumulated in the liver . The method comprises the following steps : step 1 - acquisition of an " I" ultrasound image of the liver-kidney area and conversion of the same to make it suitable for subsequent treatments ; step 2 - from said ultrasound image " I" a graph cluster is created to prepare the image for the subsequent semantic segmentation; step 3 - a copy of the image " I" is created to which a smoothing filter AxB is applied, where A and B are generally equal to 3 , obtaining a new image " I fi" ; step 4 - a semantic segmentation is performed on said " I fi" image using an acycl ic neural network based on the resnet50 model; step 5 - from this semantic segmentation, two or more pixel areas are obtained which are marked as "Liver" and "Kidney" (Liver=L, Kidney=K) ; step 6 - said pixel areas are used to create pixel selection masks; step 7 - the masks thus obtained are applied to the image "I"; step 8 - if there is more than one area marked with the same name, only the largest area is kept, thus highlighting, on the original image "I", two sets of pixels: "L" (Liver) and "K" (Kidney) ; step 9 - each of said sets is subjected to "erosion", i.e. reduced along the entire perimeter using a square section of a predetermined number of pixels for liver and for kidney; step 10 - after the erosion, the uniqueness of each area is checked again and, in case the erosion has determined a number of areas greater than one for each organ, only the largest pixel area is kept, which it is labeled "K" for kidney and "L" for liver; step 11 - a centroid or barycenter is calculated for each set and, starting from each centroid, a predefined number "x" of ROIs (Region Of Interest) are obtained, called "Kx" and "Lx" , each ROI being a circular area whose radius is proportional to the length of the minor axis of the ellipse which has the same normali zed second central moments as the region; step 12 - for each ROI , a positioning on the image " I" is calculated; step 13 - for each ROI , a segmentation of the ROI itsel f is followed by applying a k-means algorithm to partition the pixels within said ROI ; step 14 - "n" Kx / Lx ratios of the average intensity value calculated in each segmented ROI are evaluated; step 15 - the value of KLI (Kidney to Liver Index ) is obtained as the average calculated on said "n" values ; step 16 - on the basis of the index obtained, the degree of hepatic steatosis is estimated according to the following scale : a . S O = Normal ; b . S I = Mild; c . S2 = Medium; d . S3 = Severe ; step 17 - the system, using a mathematical function, proposes an estimate of the percentage of fat accumulated in the liver .

[0016] I f the operator deems it appropriate , he can manually move one or more ROI s and view the result of the evaluation performed on the new moved ROI s .

[0017] Preferred embodiments and non-trivial variants of the present invention are the subj ect matter of the dependent claims .

[0018] It is understood that all attached claims form an integral part of the present description .

[0019] The present invention solves the cited drawbacks as it completely releases the result of the evaluation of the degree of steatosis from the operator ' s abilities by means of the definition of said quantitative index of hepatic steatosis KLi , said index being calculated by means of a standardi zed and repeatable procedure .

[0020] It will be immediately obvious that innumerable variations and modi fications can be made to what is described ( for example relating to the hardware components of the system and to the formulation of the algorithm with equivalent functionality) , without departing from the protection field of the invention, as appears from the attached claims .

[0021] The present invention will be better described by some preferred embodiments, provided by way of non-limiting example, with reference to the attached drawings in which:

[0022] - Figures 1 (a, b) show the block diagram of the system according to two possible configurations ;

[0023] - Figure 2 shows an ultrasound image "I" of a parasagittal and / or lateral scan containing the liver lobe and right kidney;

[0024] Figure 3 shows the scheme with which the comparisons between the different ROIs are made.

[0025] With reference to Figures 1 (a, b) , the system according to the invention comprises an ultrasound scanner (1) , equipped with a probe (2) and connected to processing means (3, 4) on which an algorithm / sof tware is installed which implements the method according to the invention.

[0026] According to a first embodiment of the invention, shown in Figure la, said algorithm / sof tware and its related method are running on a remote PC or server (3) in communication with the ultrasound machine (1) via a network connection.

[0027] According to a second embodiment of the invention, shown in FIG. lb, the algorithm / sof tware and its related method are running directly on the central unit (3) of the ultrasound machine (1) .

[0028] In both said embodiments, the image is acquired by means of said probe (2) directly connected to the echograph (1) .

[0029] Operation is as follows:

[0030] - the probe (2) acquires the ultrasound image signal and sends it to the ultrasound machine (1) ;

[0031] - the ultrasound (1) processes the signal from the probe (2) and generates a first image "I"; said processing means (3, 4) process the image and supply the result. Said result will be shown in the external processing unit (3) , connected to the ultrasound machine (1) via a network connection, or directly on the ultrasound machine (1) , depending on which are the processing means with which the ultrasound (1) is connected.

[0032] The method for obtaining the quantitative index of hepatic steatosis KLi (Kidney to Liver Index) will now be described.

[0033] Said processing means (3, 4) receive from the ultrasound machine (1) a rectangular image of a parasagittal and / or lateral ultrasound scan containing the lobe of the liver and the right kidney ( FIG . 2 ) .

[0034] The algorithm that underpins the software includes the following steps.

[0035] Step 1 - Acquisition of an "I" ultrasound image of the liver-kidney area and conversion of the same to make it suitable for subsequent treatments, by performing one or more of the following operations: a. extraction, if said "I" image is encapsulated (for example in DICOM) ; b. scaling to n x m pixels where n and m are predefined dimensions (e.g. 800x600) ; c. transformation to RGB, if it is in grayscale.

[0036] Step 2 - A graph cluster is created from said ultrasound image "I" to prepare the image for the subsequent semantic segmentation, said graph cluster being created by performing one or more of the following operations: a. a trapezoidal Region Of Interest (ROI) is positioned in the central area which falls approximately in correspondence with the ultrasound image to be analyzed, said ROI being considered as a foreground object which contains useful information for the algorithm; b. based on the proportions between the sides of the starting image "I", rectangular areas are drawn along the edges of said image "I", these areas being considered "background" which does not contain interesting information for the purposes of the algorithm; c. a Lazy Snapping algorithm is applied to said RO I; d. a binary mask is applied to said "I" image to extract the foreground object obtained by Lazy Snapping .

[0037] Step 3 - A copy of the "I" image is created to which a smoothing filter AxB (generally 3x3) is applied, obtaining a new "Ifi" image.

[0038] Step 4 - A semantic segmentation is performed on said "Ifi" image using an acyclic neural network based on the resnet50 model, said acyclic neural network resnet50 being modified as follows: a. the input layer is modified to receive n x m pixel images; b. the classification layer is redefined to manage 3 classes: Kidney, Liver, Other; c. the network is trained to segment kidney and liver on the ultrasound image "I" using a database created ad-hoc with images to which common "data augmentation" techniques have been applied. For the training of said neural network resnet50, the following training parameters are used: a. Momentum: 0.8 b. InitialLearningRate: 0.0094 c. LearningRate: 16 d . DropFa ctor: 0.5442 e. L2: 0.0077.

[0039] Step 5 - From this semantic segmentation two or more pixel areas are obtained which are marked as "Liver" and "Kidney" (Liver=L, Kidney=K) .

[0040] Step 6 - These pixel areas are used to create pixel selection masks. The mask is a map of pixels with dimensions equal to the starting image where each pixel position has a value of 1 if the pixel belongs to the area to be selected, 0 in other cases .

[0041] Step 7 - The masks thus obtained are applied to image "I" (image without smoothing filter applied) . Step 8 - If there is more than one area marked with the same name, only the largest area is kept, thus highlighting, on the original "I" image, two sets of pixels: "L" (Liver) and " K" (Kidney) .

[0042] Step 9 - Each of said sets is subjected to

[0043] "erosion", i.e. reduced along the entire perimeter using a square section of a pre-established number of pixels (for example 10) for liver and for kidney .

[0044] Step 10 - After the erosion, the uniqueness of each area is checked again and, in case the erosion has resulted in a number of areas greater than one for each organ, only the largest pixel area is kept, which is labeled "K" for kidney and "L" for liver. Step 11 - A centroid is calculated for each set and, starting from each centroid, a predefined number "x" (default setting of the software, for example 5) of all ROIs is obtained, called "Kx" and "Lx", each ROI being a circular area whose radius is proportional to the length of the minor axis of the ellipse which has the same second normalized central moments as the region. For example, in case of x = 5, these ROIs will be: KO, KI, K2, K3, K4, L0, LI, L2, L3, L4.

[0045] Step 12 - For each ROI, a positioning is calculated on the image "I" the positioning of said ROIs "Kx" and "Lx" being carried out according to the following procedure: a. a ROI ("Lx", "Kx") is positioned in its reference centroid (liver centroid for "Lx", kidney centroid for "Kx") ; b. said ROI ("Lx", "Kx") is then moved progressively along both axes (x,y) by a quantity "j" of pixels where j=i / 5 and "i" is the current iteration step of the displacement itself, the direction of the displacement being determined by the straight line joining the two centroids ("K" and "L") in the direction that goes from "K" to "L" and from "L" to "K", so that, when x = 0, the system proceeds along the directrix, for x > 0, the system moves the x-th ROI following a rotated directrix by:

[0046] - ±JI / 12 radians for KI, LI and K2, L2;

[0047] - ±JI / 5 radians for K3, L3 and K4, L4 and so on; c. the movement of the ROI ("Lx", "Kx") ends when it reaches (without going beyond) the edge of the area it belongs to, providing a "fan" of ROIs.

[0048] Step 13 - For each ROI, a segmentation of the ROI itself is performed by applying a k-means algorithm to partition the pixels within said ROI, said application of the k-means algorithm to partition the pixels within said ROI being carried out according to the following procedure: a. for the ROIs of liver "Lx", the algorithm is set by dividing the pixels inside the ROI into 4 partitions: the partition with greater average intensity is called 1 and the partition with lower average intensity is called 4, in the event that the intensity ratio between partition 1 and partition 2 is lower than the predefined threshold (e.g. 55%) , it will be assumed that said partition 1 represents structures of the liver that are not useful for the purposes of the algorithm (for example blood vessels) and said partition represents 4 dark / black areas lacking useful information for calculating the KLi index. b. the mean intensity value for said liver ROI "Lx" is taken as the weighted average of the mean intensities of the two clusters with the highest mean intensity; c. for the ROIs of kidney "Kx", the k-means are set to partition the pixels belonging to said ROIs "Kx" into 4 sets, assuming that the sets with the highest and lowest average intensity represent structures that are not useful for calculation purposes (for example renal calices or pyramids) ; d. the ratio between the intensities of the two intermediate clusters is evaluated: if the ratio is less than 0.8, the mean intensity value for the ROI of the kidney "Kx" is assumed to be equal to the mean intensity of the cluster with the highest mean intensity;

[0049] - if the ratio is greater than or equal to 0.8, the areas of the two clusters are compared; if the ratio is lower than 0.3, the intensity of the brightest cluster will be assumed as the average otherwise a weighted average of the average intensities of the two clusters will be used .

[0050] Step 14 - "n" Kx / Lx ratios of the average intensity value calculated in each segmented ROI are evaluated: KLv(n)= Kx / Lx with x = 0e4, where "x" is called the default number (setting of basis of the software, assumed equal to 5) .

[0051] The number of ROIs equals x where the ROIs are numbered from 0 to x-1.

[0052] Only the ratios between ROIs at close range are evaluated, according to the scale shown in Figure 3, which shows the criterion with which the comparisons between the various ROIs are made.

[0053] Each ROI "Kx" is compared only with the ROIs "Lx" which are closer in order to avoid comparing ROIs located at too different depths which could distort the evaluation.

[0054] The closest ROIs are compared with each other according to the scheme in Figure 3. The following 11 comparisons result: 1. KLv ( 1 ) = KQ / LQ ; 2. KLv (2)=K0 / Li;

[0055] 3. KLv (3)=K0 / L2;

[0056] 4. KLv (4)=Ki / L0;

[0057] 5. KLv (5) =KI / L2;

[0058] 6. KLv (6)=K2 / L0;

[0059] 7. KLv (7) =K2 / LI;

[0060] 8. KLv (8)=K3 / L2;

[0061] 9. KLv (9)=K3 / L4;

[0062] 10. KLv (10) =K4 / LI;

[0063] 11. KLv (11)=K4 / L3.

[0064] Step 15 - The value of KLi (Kidney to Liver Index) is obtained as the average calculated on said "n" values .

[0065] Step 16 - Based on the index obtained, according to the predefined cutoffs, the degree of hepatic steatosis is estimated according to the following scale, in which SO SI S2 S3 are the levels of hepatic steatosis that correspond to the severity levels of the steatosis: a. SO = Normal for l>KLi>0.8 b. SI = Slight for 0.8>KLi>0.7 c. S2 = Medium for 0.7>KLi>0.6 d. S3 = Severe for 0.6>KLi .

[0066] Step 17 - The system, using a predefined correspondence table, proposes an estimate of the percentage of fat accumulated in the liver. For example, for KLi between 0.85 and 1, it corresponds to a percentage of liver fat below 6%, between 0.75 and 0.85 liver fat between 6% and 11% and so on. Finally, if the operator deems it appropriate, he can manually move one or more ROIs and view the result of the evaluation performed on the new moved ROIs .

Claims

CLAIMS1 . Method based on Arti ficial Intelligence , suitable for automating an estimation of hepatic steatosis starting from ultrasound images , of a type that makes use of an ultrasound machine equipped with means suitable for analyzing said ultrasound images from which to obtain a quantitative index of said hepatic steatosis (KLi - Kidney to Liver Index ) through the estimation of the percentage of fat accumulated in the liver, the method including the following steps : step 1 - acquisition of an " I" ultrasound image of the liver-kidney area and conversion of the same to make it suitable for subsequent treatments ; step 2 - from said ultrasound image " I" a graph cluster is created to prepare the image for the subsequent semantic segmentation; step 3 - a copy of the " I" image is created to which a smoothing filter is applied, obtaining a new " I fi" image ; step 4 - a semantic segmentation is performed on said " I fi" image using an acycl ic neural network based on the resnet50 model ; step 5 - from this semantic segmentation two or more pixel areas are obtained, which are marked as"Liver" and "Kidney" ( Liver=L, Kidney=K) . step 6 - said pixel areas are used to create pixel selection masks ; step 7 - the masks thus obtained are applied to the image " I" ; step 8 - i f there is more than one area marked with the same name , only the largest area is kept , thus highlighting, on the original image " I" , two sets of pixels : "L" ( Liver - Liver ) and " K" (Kidney - Rene ) ; step 9 - each of said sets is subj ected to "erosion" , i . e . reduced along the entire perimeter using a square section of a predetermined number of pixels for the liver and for the kidney; step 10 - after the erosion the uniqueness of each area is checked again and, in case the erosion has determined a number of areas greater than one for each organ, only the largest pixel area is kept , which it is labeled "K" for kidney and "L" for liver ; step 11 - for each set a centroid or barycenter is calculated and, starting from each centroid, a predefined number "x" of ROI (Region Of Interest - Areas of Interest ) called "Kx" and "Lx" is obtained, each ROI being a circular area of radiusproportional to the length of the minor axis of the ellipse which has the same normali zed second central moments as the region; step 12 - for each ROI , a positioning on the image " I" is calculated; step 13 - for each ROI , a segmentation of the ROI itsel f is performed by applying a k-means algorithm to partition the pixels within said ROI ; step 14 - "n" Kx / Lx ratios of the average intensity value calculated in each segmented ROI are evaluated; step 15 - the value of KLI (Kidney to Liver Index ) is obtained as the average calculated on said "n" values ; step 16 - on the basis of the index obtained, the degree of hepatic steatosis is estimated according to the following scale : a . S O = Normal b . S I = Mild c . S2 = Medium d . S3 = Severe step 17 - the system, using a predefined correspondence table , proposes an estimate of the percentage of fat accumulated in the liver .2 . Method according to claim 1 , wherein one or moreof the following operations is performed on said ultrasound image "I": a. extraction, if said "I" image is encapsulated; b. scaling to n x m pixels where n and m are default sizes; c. transformation to RGB, if it is in grayscale.

3. Method according to claim 1, wherein said graph cluster is realized by performing one or more of the following operations: a. a trapezoidal Region Of Interest (ROI) is positioned in the central area which falls approximately in correspondence with the ultrasound image to be analyzed, said ROI being considered as a foreground object which contains useful information for the algorithm; b. based on the proportions between the sides of the starting image "I", rectangular areas are drawn along the edges of said image "I", these areas being considered "background" which does not contain interesting information for the purposes of the algorithm; c. a Lazy Snapping algorithm is applied to said ROI; d. a binary mask is applied to said "I" image to extract the foreground object obtained by LazySnapping .

4. Method according to claim 1, wherein said acyclic neural network based on the resnet50 model, is modified as follows: a. the input layer is modified to receive n x m pixel images; b. the classification layer is redefined to manage 3 classes: Kidney, Liver, Other; c. the network is trained to segment kidney and liver on the ultrasound image "I" using a database created ad-hoc with images to which "data augmentation" techniques have been applied.

5. Method according to claim 4, wherein, for the training of said neural network resnet50, the following training parameters are used: to. Momentum: 0.8 b. Initial Learning Rate: 0.0094 c. LearningRate : 16 d. Drop Factor: 0.5442 e. L2: 0.0077.

6. Method according to claim 1, wherein the positioning of said ROIs "Kx" and "Lx" is performed according to the following procedure: to. a ROI ("Lx", "Kx") is positioned in its reference centroid (liver centroid for "Lx", kidneycentroid for "Kx") ; b. said ROI ("Lx", "Kx") is then moved progressively along both axes (x,y) by a quantity "j" of pixels where j=i / 5 and "i" is the current iteration step of the displacement itself, the direction of the displacement being determined by the straight line joining the two centroids ("K" and "L") in the direction that goes from "K" to "L" and from "L" to "K", so such that when x = 0 the system proceeds along the directrix, for x > 0, the system moves the x-th ROI following a rotated directrix by:- ±JI / 12 radians for KI, LI and K2, L2;- ±JI / 5 radians for K3, L3 and K4, L4 and so on. c. The movement of the ROI ("Lx", "Kx") ends when it reaches (without going beyond) the edge of the area it belongs to, providing a "fan" of ROIs.

7. Method according to claim 1, wherein said application of the k-means algorithm to partition the pixels within said ROI is performed according to the following procedure: a. for the ROIs of liver "Lx", the algorithm is set by dividing the pixels inside the ROI into 4 partitions: the partition with greater average intensity is called 1 and the partition with loweraverage intensity is called 4 , in the event that the intensity ratio between partition 1 and partition 2 is lower than a predefined threshold, it will be assumed that said partition 1 represents structures of the liver which are not useful for the purposes of the algorithm and said partition 4 represents dark / black areas devoid of information useful for the calculation of the ' KLi index ; b . the mean intensity value for said liver ROI "Lx" is taken as the weighted average of the mean intensities of the two clusters with the highest mean intensity; c . for the "Kx" Kidney ROI s , the k-means is set to partition the pixels belonging to said "Kx" ROI s into 4 sets , assuming that the sets with the highest and lowest average intensity represent structures that are not useful for the purposes of the calculation; d . the ratio between the intens ities of the two intermediate clusters is evaluated :- i f the ratio is less than 0 . 8 , the mean intensity value for the ROI of the kidney "Kx" is assumed to be equal to the mean intensity of the cluster with the highest mean intensity;- i f the ratio is less than 0 . 8 , the mean intensityvalue for the ROI of the kidney "Kx" is assumed to be equal to the mean intensity of the cluster with the highest mean intensity;- if the ratio is greater than or equal to 0.8, the areas of the two clusters are compared;- if the ratio is lower than 0.3, the intensity of the brightest cluster will be assumed as the average; otherwise, a weighted average of the average intensities of the two clusters will be used .

8. Method according to claim 1, wherein said predefined number "x" is a basic setting of the software and is equal to 5.

9. Method according to claim 1, wherein said evaluation of the Kx / Lx ratios is performed only between closely spaced ROIs.

10. Method according to claim 1, wherein the operator manually moves one or more ROIs and displays the result of the evaluation performed on the newly moved ROIs (Kx, Lx) .

11. System based on Artificial Intelligence, designed for automating an estimate of hepatic steatosis starting from ultrasound images, of a type comprising an ultrasound machine (1) , equipped with a probe (2) and connected with processingmeans (3, 4) characterized in that said processing means (3, 4) are designed to implement the method according to any one of the preceding claims.

12. System according to claim 11, characterized in that said method is running on a remote PC or server (3) in communication with the ultrasound machine (1) via a network connection.

13. System according to claim 11, characterized in that said method is performed directly on the central unit (3) of said ultrasound machine (1) .

Images