Cyst or solid lesion istotype estimation method based on an ultrasound image to support tomoral diagnosis

Abstract

The computer-implemented method according to the present invention provides two different classifications starting from an ultrasound image i.e. a binary benign / malignant classification and a histotype classification based on a segmentation of morphological parameters. Furthermore, a disambiguation algorithm is performed when the two classifications are inconsistent.

Description

CYST OR SOLID LESION ISTOTYPE ESTIMATION METHOD BASED ON AN ULTRASOUND IMAGE TO SUPPORT TOMORAL DIAGNOSIS.DESCRIPTIONTECHNICAL FIELDThe present invention refers to a method of estimating a histotype of cysts and solid lesions starting from an ultrasound image for the aid of medical staff in order to increase the repeatability of medical diagnoses and reduce procedure management times and costs tumor diagnostics.STATE OF ARTThe extraction of parameters useful for the diagnosis of bladder cancer and cysts, such as ovarian, renal or pancreatic cysts, is thorough but the numerous parameters obtained still require relatively important human intervention to be put into relation with each other. Sometimes, the large amount of information can cause a situation in which data potentially indicative of a pathology are present together with many others which tend to make the relevant data less recognizable. This poses a risk regarding the reliability of the diagnosis which thus becomes very dependent on the human factor e.g. from the experience of the professional who evaluates the data.SCOPES AND SUMMARY OF THE INVENTIONThe scope of the present invention is to provide a selection of data extracted from ultrasound images in order to increase the reliability of the diagnosis and facilitate interpretation by medical personnel by drawing the latter's attention to the most significant parameters of the ultrasound images. Furthermore, based on this selection, an estimate of one or more histotypes is provided.The scope of the present invention is achieved by a computer implemented method according to claim 1. According to the invention, medical staff has a support tool to increase the accuracy of histotype evaluation and to provide output data understandable by personnel and preferably visualized together e.g. via a graphic user interface to supporta decision in case of conflicting output data. In fact, with reference to the benignity and malignancy of the solid lesion or cyst, the output information is performed via two parallel modalities i.e. through a model trained to classify the histotype based on the morphological parameters, preferably based on exclusively morphologic parameters, and through another model trained to output the benign / malignant estimate based on images with solid lesions or cysts already labeled as benign or malignant. It is in fact known that each histotype is uniquely associated, with the consensus of the medical community, with either benignity or malignancy. Since the two output pieces of information i.e. the histotype and the benign / malignant indication without the histotype, coming from parallel approaches to each other, in case of concordance, the overall data provided to the medical staff received a first level of verification. Furthermore, the classification from the two models is formally different, one is binary benign / malignant, the other one comprises each tumor subtype i.e. starting from four to forty, depending on the tumor. Even otherwise, the use of disambiguation provides both an explicit trace of the various classification steps and of the verification performed. It should also be noted that the output of the second algorithm could be inconclusive, for example because either the histotype or benign / malignant has a confidence percentage, e.g. returned by the corresponding ML algorithm such as the case of SVM, below a predefined threshold. Neural networks and ML probabilistic algorithms are examples of algorithms that provide both a classification and a confidence probability or score. Instead, a SVM algorithm does not provide a confidence probability and the latter shall be calculated based on the SVM algorithm.Disambiguation is achievable e.g. by searching in an archive of clinical cases already classified or as histotype e.g. through histochemical examination and / or as benign malignant (without histotype), those with the plurality of morphological parameters input to the deep learning algorithm for estimating the histotype, and presenting the histotypes and / or the benign / malignant class of the cases clinical with equal values or within a similar range of morphological parameters.Otherwise, it is possible to run an unsupervised machine learning algorithm such as masked autoencoder which processes in a training step a plurality of ultrasound images already classified as histotype and / or benign / malignant and, after training, this unsupervised algorithm produces an estimation of the histotype based on the similarity of the ROI input to the algorithm for estimating the benign / malignant classification (without histotype).As an alternative, according to a further embodiment, a fresh ROI is provided. Indeed, the former ROI may not identify the lesion in the best possible way. A fresh ROI may change the benign / malignant estimation without histotype and / or the histotype estimation of both ML algorithms in series.Input parameters for the second machine learning algorithms may be provided by one or more ultrasound imaging algorithms or by manual input from an operator.Advantageously, ascites is added to the morphological parameters i.e. the collection of liquid in the abdominal cavity, and preferably the data representing the patient's ascites to which the ultrasound image refers is entered manually via a special window displayed via an 1 / O user interface such as a screen and a keyboard.For example, the identification of contours is based on the identification of connected components.According to a preferred embodiment, the method includes the step of identifying the ROI via a further machine learning algorithm, preferably deep learning, via frame and / or ultrasound clip processing.According to a preferred embodiment, the benign / malignant category, the histotype group, the first quantitative parameter and the second parameter are reproduced visibly to a user on the image outside the ROI.In this way, the data is presented in a synoptic manner for the medical staff, who is therefore in the best conditions to provide a diagnosis.BRIEF DESCRIPTION OF THE DRAWINGSPreferred embodiments of the present invention will be described below, purely byway of example, with reference to the attached drawings, in which:- Fig. 1 shows a first embodiment of the method of the present invention;- Figs. 2-6 show successive steps of a segmentation of an ultrasound frame performed according to an embodiment of the present invention;- Fig. 7-10 show subsequent steps to obtain the ROI based on an algorithm according to the present invention; and- Fig. 11 shows ultrasound images of solid cysts having solid and serous fractions, serous only cysts and solid only cysts.DETAILED DESCRIPTION OF THE INVENTIONFig. 1 shows a first embodiment of the method of the present invention. This embodiment comprises a 300 Segmentation step and a 400 Identification step of an ultrasound frame either single or from an ultrasound clip containing a putative cyst and / or a putative solid lesion of the ovary, kidney, pancreas or bladder.Segmentation 300 has the objective of delimiting and identifying one or more image objects e.g. areas of pixels, representative of an organ, such as an ovary, a bladder, a pancreas, a kidney, and cysts or solid lesions on / in such organs. Segmentation 300 also provides quantitative morphological parameters based on the processing of radiomic parameters, i.e. on the texture and geometric information obtainable by analyzing the various gray levels of the pixels of the frame and relating these gray levels.Examples of parameters representing the consistency of an area of the cyst or solid lesion are based on frame texture analysis and allow to estimate whether an object e.g. a frame area refers to a serous, mucinous, solid, vascularized, nonvascularized, mixed content area i.e. combination of serum, mucin and solid part. The identification of shadow cones, being an ultrasound artefact generated by the presence of a solid component, is also part of these parameters.Examples of quantitative morphological parameters are one or more of: the regularity of a closed border of the serous area of the cyst, the regularity of an external border of a solid lesion, presence and number of loculi within the cyst, centroid of the cyst or lesion solid or loculi, characteristic area and / or dimensions of the loculi, presence andnumber of solid excrescences and / or papillae, ratio between the area occupied by the solid and serous components of the cyst contents, presence and area of non-solid hyperechoic regions (e.g. mucin or blood).Step 300 identifying the contour of an object and classifying the edge and the object can be carried out using different methods.This step delimits the entire intended object e.g. the cyst within the ROI rather than just portions of interest of the object, characterized by a uniform appearance within them.The input of this step is an ultrasound image, for example one or more frames of an ultrasound clip, while the output is defined in the form of an outline of one or more objects, coordinates of each pixel included within the object. These in turn can be represented in the form of binary masks or masks of probability or value of a given parameter or a set of parameters, such as for example the range of values that can be assumed by the features automatically extracted from the trained neural network (value of the weights assumed by neurons according to a specific pattern for each learned label); the range of values that can be assumed by one / more radiomics features and associated with a specific learned label.A first example of a processing algorithm involves the use of deep neural networks capable of performing semantic segmentation. Examples of architectures are: UNet, Mask- R CNN, GCN. These networks typically have a two-step structure. A first encoder step, which can be created using a deep convolutional network, where the learning step takes place and the networks learn to recognize the object to be segmented. The second step is decoder, where the parameters learned in the previous step are applied to the image to delimit the portion whose features correspond to the learned features. The output is generally an image of the same size as the image to be segmented where, however, each pixel is labeled as belonging or not to one or more objects to be segmented. This method does not require indications on which features and / or parameters to select, but these are implicitly extracted from the encoder.These methods involve a training step based on a set of images / frames containing the object / objects to be segmented. The set of images is provided as input to the network,as are the labels to be "learned", applied manually or by other automatic segmentation systems, typically binary masks of the same dimension of the input image, where pixels corresponding to object / objects have a non-zero value and non-corresponding pixels have a zero value. Segmentation accuracy is measured by comparing non-zero pixels in the input and output masks, through metrics that reward corresponding values, and penalize discordant values, for example IOU, Dice coefficient. In this case, to learn the algorithm, images labeled by qualified medical personnel are used.Figure 11 shows examples of ultrasound images of cysts having solid and serous fractions, serous only cysts and solid only cysts. Referring to figure 11, the method suggested for the identification of the region of interest and of its solid and serous fractions comprises the following steps:Capturing of the input medical imagePreprocessing of the image through normalizing and resizing techniquesUsing a U-net architecture for semantic segmentation, characterized by an encoder to capture the hierarchic features of the image and a decoder to rebuild the segmentation mapIdentifying and delimiting of solid lesions through a pixel classification process Resizing of the image to obtain the original dimensionThe table shows experimental results of the method, providing a quantitative valuation of loU and Dice indexes, and precision and recall. Such metrics quantify the precision of segmentation comparing the intersection area with the area of combination between said segmentation and the ground of truth (loU, Dice) and estimating the error via pixels classified as false positive and negative (precision and recall). Results are shown for the ROI segmentation (first column) and separately for the segmentation of the solid fraction (second column) and serous fraction (third column).The algorithm was trained by 917 images and 412 clips from 559 clinical cases.A second example of known technique involves the use of segmentation algorithms based on the morphology of the image and / or on the distribution of pixel values in the image. Examples of this group are the watershed method, binarization methods and the following method for detecting the contours of cysts with mixed contents i.e. serum, mucin and solid areas.Figure 2 shows an ultrasound image of a rescaled mucinous cyst e.g. at 300x300 pixels, in which in particular the mucinous part extends substantially over the entire area of the cyst. A mucinous area, as well as an area with blood effusions or relating to a fluid of a gelatinous or viscous consistency generating high scattering, has a characteristic decrease in contrast and, to be identified, the following steps are carried out.In order to increase the homogeneity or uniformity of the texture that defines the region of interest to be segmented, a smoothing operation must be performed. This homogeneity or uniformity can be defined as the distribution of the local standard deviation of the pixel value. Note that the reference parameter is the local standard deviation, i.e. calculated on a limited portion of adjacent pixels at a time, not the total one, calculated on the entire image, which could be very high and not vary even after smoothing, e.g because of the presence of an edge, although it does not carry information about the dissimilarity of adjacent pixels.A higher measurement of local standard deviation indicates a greater variation in the value of the pixels under examination, and consequently a greater dissimilarity of the texture, due to the scattering effect caused by the density of the liquid. A lower measure of local standard deviation will indicate less difference and therefore greater uniformity.The application of a filter and the parameters that regulate the behavior of the aforementioned filter are, according to the invention, selected with the aim of obtaining asignificant reduction (in the current examples a halving) of the average and maximum measurement of local standard deviation calculated on the image original and filtered.For example, in the case of applying a frequency filter (figures 2b - 2c) the following is applied:1. a Fourier transform of the image in fig. 2a2. a low-pass filter to image data resulting from the application of the transform, whose filter width (hamming window) is preferably selected as the parameter that significantly reduces e.g. the local standard deviation of the region of interest (the area to be segmented) of the image in figure 2a is 50% . In the case of cystic ultrasound images this value stands at approximately % of the input size (typically varies between 50 and 100 pixels - images measuring 300 x 300 pixels)3. Inverse Fourier transform.It should be noted that for description purposes the images are illustrated with each of the transformations but, in use, the various phases are performed on sets of image data, each set being able to be converted and displayed if necessary in one of the images attached to the description.Alternatively, it is possible to process the ultrasound image in frequency via wavelet filtering, anisotropic diffusion or Gaussian blur.For example, in the case of anisotropic diffusion, three main parameters are important: kappa, controls diffusion (smoothing) as a function of the image gradient. At low values of this parameter, very small gradients are sufficient to block diffusion (typically it varies between 20 and 100 pixels - in this specific case the value of 90 was used). High values extend the diffusion even beyond high gradients (e.g. near high contrast contours); gamma: controls the diffusion speed at each iteration (in our case = 0.25); and iterations: number of repetitions of the algorithm on the same image, usually between 20 and 100 (in the specific case positive tests were performed with 30 iterations) depending on the non-homogeneity of the starting image (few iterations for homogeneous images, higher for images with high scattering).Subsequently, the image is binarized e.g. isodata or via the Otsu algorithm, withthreshold identification by a two-component Gaussian mixture model or generalized histogram thresholding.In particular, with an isodata binarization, an algorithm automatically selects a threshold as the intermediate value between the average of the pixel values below the threshold and the pixel mean over the threshold.With reference to Otsu binarization, an algorithm automatically selects a threshold as the value that minimizes the variance between pixel values within a class and maximizes the difference between classes.In this way, two sub-sets of data are obtained respectively relating to areas considered representative of a low intensity or pixel value (or closer to the seed / inside the box) or a high intensity or pixel value (or far from the seed / external to the box).The ultrasound image in fig. 2a is also processed to obtain information regarding homogeneity via an entropic filter (fig. 3c). In particular, the information on homogeneity is represented by the Shannon entropy.To the image in fig. 3c a binarization algorithm is applied e.g. hysteresis thresholding. The latter carries out a binarization on a variable threshold set as a parameter and medium-high entropy threshold values, descriptive of the mucinous / blood effusion component. In particular, this binarization algorithm attributes the pixels of the input image to one class or another based on two selected thresholds. For each pixel, this is compared to the upper threshold. If the value is higher than the threshold, the pixel becomes white and the surrounding pixels are compared with the lower threshold value, if they exceed it they are also white. This algorithm allows the identification of more uniform connected components and is more resistant to noise and local variations than binarization methods such as Otsu. In our example, the threshold describing the characteristic entropy distribution for the mucinous areas is between 0.6 and 0.8 considering the local entropy of each pixel normalized between 0 and 1.Alternatively, the image in fig. 2a is processed via a Li binarization algorithm. In this case, the binarization algorithm receives a grayscale image as input, calculates a threshold based on cross entropy, so that the pixels with a value higher than the threshold belong toa first class ("figure"), those lower than a second class ("background"). The threshold value is chosen in order to minimize the cross entropy between the distribution of the pixels of the figure and its average, and the distribution of the pixels of the background and their average.In this step, a third and fourth sub-set of image data are therefore generated, one relating to the area considered mucinous or highly scattering, for example by assigning the value 1, and the other categorized as background or not belonging to the third sub-set for example by assigning the value 0. It should be noted that an area with high scattering e.g. due to the presence of mucin it is always part of the cyst and never of a tissue area. Therefore, the third sub-data set belongs to the same category as the first sub-data set e.g. these are areas of the cyst and not the background. This approach segments the various data sub-sets, with their connected components, i.e. with reference to binarized images, connected regions with area greater than 1 pixel composed of connected pixels, i.e. adjacent pixels that share the same value (or around the value, if they are colored or grayscale images).Preferably, the image in Fig. 2c is also segmented to verify the outermost contours of the cyst since frequency filter processing is applied to decrease the relevance of the mucinous / blood effusion components which could be, without this step, erroneously classified as tissue areas. The integration of the two approaches, with the integration of fragmented but well-defined segmented regions extracted from the original filtered image, returns a correct approximation of the cystic area. For example, in the case of a region growing algorithm, starting from the seed, a tolerance of 50 is set. The areas or connected components thus identified are subsequently segmented on the basis of the distance from the seed and grouped to obtain a fifth representative subset of image data of a distance of each area lower than a maximum threshold and a sixth sub-set of image data representing a distance of each area higher than said threshold. In this way, the fifth sub-set belongs to the first category of image data, e.g. the cyst area, and the sixth sub-set is related to the second category of image data, e.g. background.Subsequently, processing is performed on the basis of the sub-sets of image data forwhich each value assigned after binarization e.g. 0 or 1, a meaning such as 'background area' or 'cyst area' has been applied via segmentation, and each sub-dataset includes at least the pixel value and its position. When it comes to sub-sets obtained by binarization, the value is e.g. 0 or 1 for the fifth and sixth data sub-set, areas are aggregated and a tag 0 or 1 is applied to each area within a subset. For example, value 1 refers to low intensity or seed close (light color in figure 6) areas in all image data sub-sets and 0 value refers to high intensity or high distance from the seed (dark color in figure 6).For example, sub-sets of image data are combined based on the position of each pixel and the position is found to be congruent for the three groups of data since the starting image e.g. fig. 2a, it's always the same.A result image can be generated such that, for each pixel position in the result image, the pixel value is the average of the values of the three pixels each belonging to the first and second data sets, and the segmented data set (fig. 5b and labeled). When averaging is applied, the resulting image is displayed following the application of a new threshold to determine whether the pixel is from the first or second category.Otherwise, it is possible to generate the result image by applying the mode of the three values of each pixel for a given position: each pixel acquires the value of the mode of the three values obtained from each of the three elaborations of figures 6a-c.It is also possible to use correlation: cross correlation of binary masks in pairs, and averaging of correlated images. This last method returns a clear localization of the cystic area, but, in addition to computational power, requires the selection of a threshold on the correlation value to select the range of pixels to be included in the final segmentation.After identifying the resulting image data as above (fig. 6d) an area of cyst, e.g. yellow area or with reference value 1, it is then possible to calculate the area, the centroid and other morphological parameters relating to the regularity of the edges of the binary mask.According to an implementation variant, the local entropy calculation is applied to the first data set, e.g. to the result of the 'smoothing' phase in order to increase the effectiveness of identifying any areas of high scattering inside the cyst.These algorithms do not require a training step, although the choice of parameters is made based on the features of the dataset to which the segmentation is applied. Assuming a set of test data, on which the area to be segmented has been delimited manually or by other automatic segmentation systems, the segmentation accuracy is measured by comparing the non-zero pixels in the original mask and in the output mask, through metrics that reward corresponding values, and penalize discordant values, for example: IOU, Dice coefficient.A third implementation example involves the use of algorithms based on the extraction of radiomic and textural features / parameters.These methods, unlike methods based on deep neural networks, require the calculation and subsequent classification of a set of characteristics, for example shape, texture, pixel value, etc., of the object to be segmented which are selected in the training step. Using this method generally involves the following steps:- calculating selected features from an entire object or portions of images;- dividing the calculated values between those belonging to the object or portion of the image with a certain identity and those with different identities;- training a classifier, such as machine learning, neural network, etc., to associate the calculated values and the identities of the object / portion;- sending the new image and the assignment of an identity for each portion of the image as input to the trained classifier. Portions with the same identity will be assigned the same label and can be represented through, for example, binary or coordinate masks. Assuming a set of test data, on which the area to be segmented has been previously delimited manually or by other automatic segmentation systems, the segmentation accuracy is measured by comparing the non-zero pixels in the original mask and in the output mask, through metrics that reward corresponding values, and penalize discordant values (e.g. IOU, Dice coefficient). A fourth example of known technique involves the use of the previously described algorithms as a combination or subset.Through the above, it is possible to identify: a. the presence and, if so, the number of papillaeb. the presence and, if so, the number of niches c. an outline of the cyst and the maximum size of the cyst or solid lesion, with the maximum size of any solid component of the cyst d. irregularity or regularity of the edge, preferably in a binary mannerFor example, with reference to c) the input is the entire ROI to identify the entire mass and only the portion containing solid area for the corresponding maximum dimension. The maximum size of the mass is for example calculated as the major axis of the ellipse that circumscribes the area under examination.For the solid component(s), all connected components are considered (for example if there are regions of separate solid content within the mass) and the region with the largest dimension is considered as the output.With reference to regularity / irregularity, irregular margins are defined as those that have convexities or complex shapes that deviate from the circular one.The input of this step is given by the contours of the ROI of the mass, for the evaluation of the external walls; the contours of the protruding solid components (whether papillae or not) for the evaluation of the internal walls.Regularity can be calculated as:Presence or absence of papillae; if present, the margins are irregular and / orBy applying an acceptance threshold to geometric parameters such as at least one or more of circularity / fractality / shape index / roughness etc.With reference to the parameters representing the consistency of the solid lesion or cyst, some examples are the following.With reference to the cystic content, and to that on mucin for the identification of areas with cystic content, the segmentation can also be carried out with the aid of deep learning. In particular, the cystic (non-solid) content is categorized into some categories (pure or anechoic, mucinous, hemorrhagic). The input of this phase is the portion of the ROI classified as non-solid. The process can be conducted: through the extraction of textural characteristics and subsequent classification through SVM-type algorithms or through the use of deep learning for classificationPreferably, in addition to the above, a further algorithm identifies any shadow cones emitted by a solid component in the mass. The process involves: extracting candidate cones, i.e. sequences of radial pixels, starting from the origin of the ultrasound, which are all anechoic less than a tolerance threshold selecting cones that intersect solid components and are therefore potentially generated by them.An example of how this can be achieved is reported in 'An automatic geometrical and statistical method to detect acoustic shadows in intraoperative ultrasound brain images' by Pierre Hellier, Pierrick Coupe, Xavier Morandi, D Louis Collins in Med Image Anal. 2010 Apr;14(2):195-204. doi: 10.1016 / j.media.2009.10.007. Epub 2009 Nov 17.Preferably, in addition to the above, the quantitative estimate of the vascularity of the mass is calculated using a specific algorithm.For example, the calculation is carried out as: ratio or level between vascularity and solid area of the ROI; i.e. ratio between colored pixels (doppler) previously not identified as artefacts i.e. denoise and portions of ROI with solid content. Only solid components can be vascularized. Depending on the presence or absence of a predefined threshold, the level of vascularization can also be binary e.g. present / absent or high / low. An example of processing an ultrasound Doppler image to identify Doppler pixels not related to artifacts is described in WO-A1-2022130257 in the name of the same applicant.This ratio is then classified with an SVM / logistic regression type model which associates ratio and vascularization class (e.g. slightly / very vascularized...). Alternatively, deep learning can be used for implicit ultrasound frame classification.Preferably, in addition to the above, a proportion of solid component is also calculated. For example, the input is:The entire ultrasound imageThe ROI extracted at the cyst or solid lesion.With the aim of identifying portions of solid tissue and cystic content (liquid with various characteristics e.g. anechoic, mucinous, with hemorrhagic content)The process can be conducted by extracting textural features from the ultrasound image and subsequent classification through SVM-type algorithms or through the use of deep learning (e.g. UNet)The proportion of solid area is given by the ratio of solid area to the entire ROI of the mass.For example, when starting from an ultrasound clip there are one or more ultrasound images from which the previous eight parameters, including morphological and representative of the consistency of the cyst or solid lesion, are extracted either through the previous algorithms or through human intervention, preferably combined with ascites, a machine learning algorithm, preferably deep learning, is trained with a plurality of already classified images e.g. by medical personnel with the corresponding histotype i.e. the istologic classification or cellular characterization as identified via an histologic test of the solid lesion or cyst of the soundwave image. For example, ovary cyst hystotypes are indicated in references 1:17 of the bibliographic section.Examples of the machine learning algorithms that can be used are: logistic regression, linear regression, decision trees, support vector machine, Bayesian regression, perceptron, naive bayes in the case of supervised approaches; clustering algorithms such as nearest neighbours, k means, Gaussian mixture in the case of unsupervised approaches; and / or combinations thereof. The training of supervised models uses datasets equipped with labels indicating the histotype for each input data, in this case affixed by expert doctors. Through iterations, the algorithm selects the range of defining parameter values that maximize learning, i.e. the model's ability to assign the correct label to a given set of input data. The learning defines the function underlying the model that best describes and separates, in the case of conflicting labels, the input data. The parameters that define the model are for example the slope and intercept of the line in the case of linear regression, the polynomial in the case of the support vector machine, the depth or the number of branches generated in the case of a decision tree. It is also possible to define regularization terms, i.e. terms added to the function that defines the model itself, and which assign a penalty to incorrect answers during learning. An example of regularization is L2 forlogistic regression. For some models such as support vector machine it is possible to define the function underlying the model, which can be linear, cubic, or an n-dimensional polynomial.Unsupervised model training uses datasets without labels. The purpose of learning is to define a space (cluster) to which each input data belongs, in order to maximize the proximity of data belonging to the same space, and maximize the distance of different spaces. An example of a parameter that defines many clustering algorithms is the number of clusters allowed at the output, i.e. the number of categories implicitly detected and into which the input data can be divided. Following this training, when the machine learning algorithm provides its classification estimate, it also provides a percentage confidence of that estimate. When the estimation result is displayed on an I / O user interface such as a screen and keyboard, it is possible to display only the histotype with the highest confidence percentage or a group of two or more histotypes whose confidence percentage is progressively decreasing with a sum overall less than 100%. It has in fact been verified that the eight parameters indicated above, preferably in addition to ascites, allow estimates of all histotypes of solid lesions or cysts to be provided with good precision. It should be noted that, except in rare cases in which the histological examination presents mixed biological tissues, each histotype corresponds to one and only one indication between a benign mass or a malignant mass.A preferable method to identify the histotype through Random Forest comprises the following steps:Receiving some or all morphologic parameters extracted from previous algorithms or human interventionProcessing and transforming previously selected parameters via proprietary parameter engineering techniques applied to said parameters to improve the discrimination of the classification model.The Random Forest algorithm is used to classify elements in target classes (histotype), by using multiple decision trees acting on a bundle of features previously processed via the proprietary engineering process to maximize the separation amongclasses and decrease the classification error.The present invention further includes an Identification step 400 to preferably confirm the presence of the cyst or solid lesion and estimate a benign / malignant category of the cyst or solid lesion within the ultrasound frame. For example, one or more of the frames described above may have produced contour and content parameters that suggest a cancerous lesion. Phase 400 performs a frame analysis in order to estimate a benign or malignant category to which the lesion belongs.The identification step 400 can be carried out using different methods and manufacturing techniques. For example, the input is an ultrasound image e.g. a frame, an ultrasound video, which outputs a benign or malignant category of the cyst or solid lesion and, preferably, the probability associated with the category to which it belongs.A first example of an implementation suitable for the purpose involves the use of deep neural networks for classification of the entire ultrasound image e.g. of the frame.The architecture generally involves the use of a convolutional network (CNN). The input is a set of images representing objects, e.g. cysts or solid lesions, where each image is associated with a label indicating a category or categories to which the object belongs. The purpose of the learning phase is to teach the algorithm the association between an image and the appropriate category, through the identification of parameters automatically extracted by the network from the training dataset, and which best represent a given category.The training of the neural network occurs by dividing the available dataset into three subsets: a training subset, a validation subset, a test subset. Typically, the training subset consists of 60% of the data, while the validation and test subsets consist of the remaining 20% and 20%. Each subset shares the same properties, for example in terms of frequency of occurrence of a given label, origin of the data, characteristics of the data (size, appearance). Training consists of the network learning a specific response pattern in association with the presentation of an input belonging to a given category. For effective classification, different categories will be associated with separate response patterns. The response pattern is made up of the weights, i.e. the connections between the individualneurons that make up the network. In a preferred version of the training, the network parameters are set as follows: the learning rate (i.e. the coefficient of variation of the network weights at each iteration) is between 0.001 and 0.1; the optimization algorithm (i.e. the approximation operation of the network weight pattern that maximizes learning) is ADAM; the number of epochs (i.e. iterations in which the network is exposed to the training dataset and the weights are modified) is chosen adaptively through an early stopping algorithm, i.e. the iterations are interrupted, with a tolerance factor, when the Prediction error on the training dataset decreases, but that on the validation and test dataset increases (a phenomenon called overfitting). The network parameters are selected iteratively through observation of the performances obtained on the validation set. Once the selection process is completed, training is measured on the test dataset and performance evaluated. Learning is considered completed when performances are greater than or equal to an established threshold (for example 90% of correct answers) and stable (i.e. with variations lower than a given threshold, for example plus or minus 3 percentage points) through reiterations of the process following changes in the combination of subportions of the dataset.In a preferred version of the invention, the dataset is subjected to "augmentation" operations, that is, each piece of data is randomly subjected to geometric operations (rotations, translations, scale) and appearance variations (contrasts and luminance), for the purpose to increase their number and limit the effect on learning of non-replicable random variables (for example, all the data collected in a medical center have very low contrast values).The output is a subset of the relevant images, together with the probability that the object belongs to the benign / malignant category with reference to a tumor pathology. The learning accuracy is calculated on a second set of images / frames not present in the training dataset, with statistics and characteristics comparable to the training dataset, comparing the number of correctly categorized images, missed ones, and false positives and obtaining accuracy, sensitivity, and specificity scores.A second implementation example involves the use of algorithms based on the extraction of features, radiomic and textural parameters. These methods, unlike methods based on deep neural networks, require the calculation and subsequent classification of a set of characteristics, such as shape, texture, pixel value, etc., of the object to be categorized which are selected by the experimenter in phase of training. One use of this method involves: (1) the calculation of the features selected from an entire object / portions of images, (2) division between the calculated values between those belonging to the object / portion of image with a certain identity and those with others identity, (3) training of a classifier (machine learning e.g. neural network, logistic regression, linear regression, decision trees, support vector machine, Bayesian regression, perceptron, naive bayes in the case of supervised approaches; clustering algorithms such as nearest neighbors , k means, Gaussian mixture in the case of unsupervised approaches; and / or combinations of these) to associate calculated values and identity of the object / portion, (4) administration of a new image as input to the trained classifier and assignment of an identity to the image. The learning accuracy is calculated on a second set of images e.g. frames not present in the training dataset, with statistics and characteristics comparable to the training dataset, comparing the number of correctly categorized images, those missed, and false positive ones, and obtaining scores for accuracy, sensitivity and specificity.A third example involves the use of the previously described algorithms as a combination or subset.Also in this case, the machine learning algorithm 400 returns the estimate of the benign or malignant mass indication with a percentage of confidence in the estimate.Still considering Fig.l, the outputs generated by phases 300 and 400 are compared with each other with reference to the common parameter i.e. benign or malignant mass and / or with reference to the confidence percentages of the estimate.When mutual confirmation emerges from the comparison e.g. the most probable histotype in which the confidence percentage is higher than a predefined threshold is benign (or malignant) and this is confirmed by the output of the 400 algorithm also in this case with a confidence percentage higher than a predefined threshold (the predefinedthresholds for the algorithm 300 and 400 can be the same or different, preferably the threshold of the algorithm 300 is lower than that of the algorithm 400), the automatic analysis is considered concluded and the result is displayed on the screen.However, when the comparison reveals an inconsistency between the benign / malignant information or one or both confidence percentages are lower than the corresponding predefined thresholds, a disambiguation algorithm 800 is performed.This algorithm can be performed in numerous ways and most preferably provide the estimate of the benign or malignant mass indication (without histotype) and / or the histotype via a different machine learning algorithm e.g. unsupervised if the algorithm 300 or 400 is supervised or vice versa, or on the basis of parameters additional to those of the algorithm 300, which can be entered manually via a special window opened automatically in the case of starting the disambiguation algorithm 800, or on the basis of data archives e.g. clinical and / or literature, in which at least the parameters of the algorithm 300 possibly integrated by the others inserted as inputs of the disambiguation algorithm 800, allow to extract, for example through known type indexing of the data in the archives, the histo types and / or the benign / malignant indication (without histotype) with the incidence of classification compared to the total of records identified with the same values of the input parameters to the algorithm 300 or with values that differ from those input to the algorithm 'algorithm 300 within corresponding pre-defined intervals.When the disambiguation algorithm 800 includes an unsupervised machine learning algorithm, e.g. transformes or masked autoencoder, the same inputs as the supervised 300 or 400 algorithms are provided as input and the unsupervised algorithm provides for example one or more percentages of occurrences of the histotypes or of banigno / malignant (without histotype) within the cluster a which the input(s) are categorized as closest in the multidimensional space generated during unsupervised training.Alternatively or in combination with the foregoing, the disambiguation algorithm 800 includes the step of automatically opening a window for entering the input data of the algorithm 300, through which the user reads the inputs to the algorithm 300 and canmodify one or more of them, thus causing a new execution of the algorithm 300. It is in fact possible that one or more input data of the algorithm 300 can be classified by the medical personnel in a different way from the result of the algorithms that they have extracted from the the ultrasound image the estimate of each input parameter of the algorithm 300.According to a preferred embodiment, the method of the present invention comprises a navigation step 600 (not shown in Figure 1) to produce a region of interest (ROI) containing a cyst or solid lesion in the ultrasound image. This phase may involve receiving manually generated data, e.g. the medical staff independently indicates the ROI, or, preferably, an algorithm is set up to process the ultrasound image and automatically generate the ROI. This last embodiment is particularly useful when the input is an ultrasound clip: it is in fact possible that some frames do not include cysts or solid lesions.By region of interest, or ROI, we mean a portion of the ultrasound image delimited by a visual indicator within which the cyst or solid lesion is located on which to carry out detection, measurement and / or data analysis. The indicator can be of different shapes such as polygonal, circular, elliptical; the most commonly adopted shape is the rectangular one. Preferably, the navigation phase 600 also returns an indication of the absence of an ROI if an area containing a cyst or a solid lesion cannot be identified.The ROI reduces observational and computational efforts to a specific area, leaving out the remaining part of the ultrasound image and increasing processing speed.The navigation step 600 has an input represented by an ultrasound image, e.g. a frame or from an ultrasound video and the returned output includes:- a subset of the ultrasound images on each of which the delimitation of the region of interest is applied;- a delimitation using a bounding box of a sub-area of the image or frame within which the ROI is contained;- a localization of the ROI, by localizing the centroid of the region.A first example of a phase performed automatically to obtain a subset of ultrasound images containing an ROI is through the use of deep neural networks to carry out aclassification.The architecture is called a convolutional network (CNN), and is trained on a set of images / frames that contain or do not contain a region of interest. The learning accuracy is calculated on the basis of a second set of images / frames that was not used in the training phase, but with statistics and characteristics comparable to the training dataset, as previously indicated in the segmentation phase 300. Comparing the number of regions of interest identified, those not identified and false positives, accuracy, sensitivity and specificity scores are obtained.A second example of a phase performed automatically to obtain a delimitation using bounding boxes of the area occupied by the ROI is through the use of object detection neural networks.The training in this case requires the input to be a set of images with the corresponding bounding box coordinates around the object to be identified. These networks analyze sub-portions of the image and, for each, extract and classify n-degree characteristics automatically extracted from the neural network, they are outputs of the mathematical operations of the various neurons of the network, and of the combination operations of the output of the neurons that take place to each layer of the network starting from the input, inputs which are the ordered pixel values of the image. These are compared with those contained in the subportion indicated as the label of the image being trained. The algorithm learns to recognize the combination of characteristics that maximizes the probability of identifying the desired object and consequently its position. Also in this case the learning accuracy is calculated on a second set of images / frames that was not used in the training phase, but with statistics and characteristics comparable to the dataset used in the training phase. The boxes placed by the algorithm are compared in terms of overlap and size with those placed manually. The metrics most commonly used to evaluate the goodness of the boxes are the intersection over union (IOU) and the mean average precision applied to the bounding boxes, comparing the real area and position (provided by the user) and predicted (by the network) of the bounding box.A third example of a phase performed automatically to obtain a binarized map if theknown parameters are subjected to thresholding, bounding box, or the centroid of the isolated portion and is obtained using segmentation algorithms based on the morphology of the image and / or on the distribution of the pixel values in the image.If the object of interest has known properties, such as for a cyst: the presence of serum, the presence of highly echogenic edges, the presence of hyperechoic areas, these are searched for within the image and the portions of the image that the closer these properties become, they are delimited. If the object is not present in the image, the algorithm does not identify portions that correspond to the selection criteria. After identifying the image area as above, a magnification factor is applied, for example 1.5, to increase the probability that the area thus defined includes all the morphological characteristics and the corresponding pixels of the cyst or solid lesion.According to an embodiment of the present invention, the disambiguation algorithm 800 comprises the step of modifying the ROI e.g. modifying the size and / or position so that the subsequent histotype estimation algorithm based on morphological parameters receives a new input. The modification of the ROI can be either manual e.g. a window is opened and the user selects the new ROI being preferably the previous ROI indicated on the image so that the user can indicate a new ROI different from the previous ROI; or automatic e.g. YOLO or sliding box.According to a preferred embodiment, the method further comprises the step of applying 700 to the processed image, i.e. the image initially processed in the Identification phase 600, the results of the processing with graphics readable by medical personnel when e.g. displayed on an ultrasound device screen or connected to an ultrasound machine, preferably outside the ROI and / or in an area of the image labeled during the segmentation step 300 as background.According to a preferred embodiment, when the results of the Segmentation 300 and Navigation 600 phases are obtained in parallel, the method comprises a congruence check step, preferably to check whether ROI and one or more objects or contours recognized in segmentation step 300 are at least partially overlapped. Also this information, preferably, is provided to the medical staff.Figures 7-10 show an embodiment of a mixed algorithm to implement the navigation phase 600. For example, the entropy of the image is calculated using an entropy filter (figure 7) in order to make the areas whose gray levels are homogeneous. Indicatively, this makes both the serous areas and the surrounding tissues more visible, when actually visually homogeneous. Alternatively it is possible to apply filters to Hessian matrices.Subsequently, the crop of the entropic image is subjected to binarization using specific thresholds and then segmented in order to extract three different segmentations. Two of these (figures 7a, 7b) represent an approximation of the serous area with respective different but both low thresholds for identifying the areas with high contrast and low entropy, the remaining one instead the surrounding tissues (figure 8c) through the application of a high threshold, i.e. to identify areas of high entropy. The two approximations of the serous area, therefore inside the cyst, have slightly different visual characteristics and are useful for validating the results. In particular, the two internal segmentations differ in that the respective thresholds adopted for binarization have similar but not identical values.Specifically, taking the two segmentations from the inside, the centroids of the dark connected components are calculated, i.e. the closed areas representing the serous zones of an image (figure 9a) and verify that they are contained within the connected components of the other segmentation with low threshold. When this and other conditions are verified, e.g. area of the connected component above a predetermined and settable threshold, distant connected component, e.g. outside of a predetermined and settable limit distance from the edges of the ultrasound, and a predetermined and settable shape factor based on the relationship between a major axis and a minor axis of the connected component approximated with an ellipse, then the components are considered connected as actually representing a serous area (figure 9b).In this way, any number of serous zones can be localized, the maximum number of which is defined via a specific modifiable parameter; consequently the method of the present invention is capable of identifying multilocular cysts and localizing the individual locules. The identification of the loculi with the corresponding position is subsequentlyused to define the extension on the ROI image, i.e. the ROI must contain all the niches.The segmentation from the outside, i.e. the one that identifies the contours of the cyst, must be compared with the serous areas actually identified and selected in the previous phase. Preferably, only contours sufficiently close to the serum are selected. To calculate the distance between the surrounding tissues and the serous area, the Euclidean distances between the centroids of the serous areas (figure 9b) and those of the individual segmented contours (figure 10a) are considered: if this distance is less than a predefined and settable threshold the contour under examination is associated with the area and therefore selected (figure 10b). The default threshold can be fixed or adaptive, i.e. be expressed as a function of the size of the serous zone, for example of one or both of the semi axes. In this way, each niche will be associated with its contours.Once both the serous areas and their contours have been extracted, we proceed to identify the ROI e.g. a box is positioned around this area, which has as its side the maximum length and maximum height covered by the mask, with a possible scale factor to include any undetected portions.BIBLIOGRAPHY[1] D. Timmerman et al. “ESGO / ISUOG / IOTA / ESGE Consensus Statement on preoperative diagnosis of ovarian tumors". En. In: Ultrasound in Obstetrics& Gynecology 58.1 (lug. 2021), pp. 148-168. Issn: 0960-7692, 1469-0705. Doi:10.1002 / uog.23635. url: https: / / onlinelibrary.wiley.com / doi / 10.1002 / uog.23635.[2] Francesca Moro et al. “Ultrasound evaluation of ovarian masses and assessment of the extension of ovarian malignancy". En. In: The British Journal of Radiology 94.1125 (set. 2021), p. 20201375. Issn: 0007-1285, 1748-880X. doi:10.1259 / bjr.20201375. url: https: / / www.birpublications.org / doi / 10.1259 / bjr.20201375.[3] A. C. Testa et al. “Imaging in gynecological disease (1): ultrasound features of metastases in the ovaries differ depending on the origin of the primary tumor". En. In: Ultrasound in Obstetrics and Gynecology 29.5 (mag. 2007), pp. 505-511. Issn:09607692, 14690705. Doi: 10.1002 / uog.4020. url: https: / / onlinelibrary.wiley.com / doi / 10.1002 / uog.4020.[4] V. N. Demidov et al. “Imaging of gynecological disease (2): clinical and ultrasound characteristics of Sertoli cell tumors, Sertoli-Leydig cell tumors andLeydig cell tumors". En. In: Ultrasound in Obstetrics and Gynecology 31.1(gen. 2008), pp. 85-91. Issn: 09607692, 14690705. Doi: 10.1002 / uog.5227. url: https: / / onlinelibrary.wiley.com / doi / 10.1002 / uog.5227.[5] C. Van Holsbeke et al. “Imaging of gynecological disease (3): clinical and ultrasound characteristics of granulosa cell tumors of the ovary". En. In: Ultrasound in Obstetrics and Gynecology 31.4 (apr. 2008), pp. 450-456. Issn: 09607692, 14690705. Doi: 10.1002 / uog.5279. url: https: / / onlinelibrary .wiley. com / doi / 10.1002 / uog.5279.[6] L. Savelli et al. “Imaging of gynecological disease (4): clinical and ultrasound characteristics of struma ovarii". En. In: Ultrasound in Obstetrics and Gynecology 32.2 (ago. 2008), pp. 210-219. Issn: 09607692, 14690705. Doi:10.1002 / uog.5396. url: https: / / onlinelibrary.wiley.com / doi / 10.1002 / uog.5396.[7] D. Paladini et al. “Imaging in gynecological disease (5): clinical and ultrasound characteristics in fibroma and fibrothecoma of the ovary". En. In: Ultrasound in Obstetrics and Gynecology 34.2 (ago. 2009), pp. 188-195. Issn: 09607692, 14690705. Doi: 10.1002 / uog.6394. url: https: / / onlinelibrary.wiley.com / doi / 10.1002 / uog.6394.[8] S. Guerriero et al. “Imaging of gynecological disease (6): clinical and ultrasound characteristics of ovarian dysgerminoma". En. In: Ultrasound in Obstetrics & Gynecology 37.5 (mag. 2011), pp. 596-602. Issn: 09607692. Doi:10.1002 / uog.8958. url: https: / / onlinelibrary.wiley.com / doi / 10.1002 / uog.8958.[9] I. Dierickx et al. “Imaging in gynecological disease (7): clinical and ultrasound features of Brenner tumors of the ovary: Brenner tumors of the ovary". En. In: Ultrasound in Obstetrics & Gynecology 40.6 (die. 2012), pp. 706-713. Issn: 09607692. Doi: 10.1002 / uog.lll49. url: https: / / onlinelibrary.wiley.com / doi / 10.1002 / uog.11149.

[0010] D. Franchi et al. “Imaging in gynecological disease (8): ultrasound characteristics of recurrent borderline ovarian tumors: Borderline ovarian tumor recurrence". En. In: Ultrasound in Obstetrics & Gynecology 41.4 (apr. 2013), pp.452- 458. Issn: 09607692. Doi: 10 . 1002 / uog . 12276. url: https : / / onlinelibrary.wiley.com / doi / 10.1002 / uog.12276.

[0011] F. Moro et al. “Imaging in gynecological disease (11): clinical and ultrasound features of mucinous ovarian tumors: Clinical and US characteristics of mucinous ovarian tumors". En. In: Ultrasound in Obstetrics & Gynecology 50.2 (ago. 2017), pp. 261-270. Issn: 09607692. Doi: 10.1002 / uog.17222. url: https: / / onlinelibrary.wiley.com / doi / 10.1002 / uog.17222.

[0012] F. Moro et al. “Imaging in gynecological disease (12): clinical and ultrasound features of invasive and non-invasive malignant serous ovarian tumors". En.In: Ultrasound in Obstetrics & Gynecology 50.6 (die. 2017), pp. 788-799. Issn: 09607692. Doi: 10.1002 / uog.l7414. url: https: / / onlinelibrary .wiley. com / doi / 10.1002 / uog,17414.

[0013] F. Moro et al. “Imaging in gynecological disease (13): clinical and ultrasound characteristics of endometrioid ovarian cancer: Endometrioid ovarian cancer". En. In: Ultrasound in Obstetrics & Gynecology 52.4 (oft 2018), pp. 535- 543. Issn: 09607692. Doi: 10.1002 / uog.l9026. url: https: / / onlinelibrary.wiley.com / doi / 10.1002 / uog.19026.

[0014] F. Pozzati et al. “Imaging in gynecological disease (14): clinical and ultra- sound characteristics of ovarian clear cell carcinoma". En. In: Ultrasound in Obstetrics & Gynecology 52.6 (die. 2018), pp. 792-800. Issn: 0960-7692, 1469- 0705. Doi: 10.1002 / uog.l9171. url: https: / / onlinelibrary wiley.com / doi / 10.1002 / uog,19171.

[0015] B. A. Virgilio et al. “Imaging in gynecological disease (16): clinical and ultra- sound characteristics of serous cystadeno fibromas in adnexa". En. In: Ultra- sound in Obstetrics & Gynecology 54.6 (die. 2019), pp. 823-830. Issn: 0960- 7692, 1469-0705. Doi: 10.1002 / uog.20277. url: https: / / onlinelibrary. wiley.com / doi / 10.1002 / uog.20277.s

[0016] P. Anfelter et al. “Imaging in gynecological disease (17): ultrasound features of malignant ovarian yolk sac tumors (endodermal sinus tumors)". En. In: Ultra- sound in Obstetrics & Gynecology 56.2 (ago. 2020), pp. 276-284. Issn: 0960- 7692, 1469-0705. Doi: 10.1002 / uog.22002. url: https: / / onlinelibrary. Wiley.com / doi / 10.1002 / uog.22002.F. Moro et al. “Imaging in gynecological disease (22): clinical and ultrasound characteristics of ovarian embryonal carcinomas, non-gestational choriocarci- nomas and malignant mixed germ cell tumors". En. In: Ultrasound in Obste- tries & Gynecology 57.6 (giu. 2021), pp. 987-994. Issn: 0960-7692, 1469-0705. Doi: 10.1002 / uog.23529. url: https: / / onlinelibrary.wiley.com / doi / 10.1002 / uog.23529.

Claims

CLAIMS1. Computer implemented method of processing an ultrasound image representative of a solid lesion of the ovary, kidney, pancreas, bladder or cyst e.g. ovarian, renal, hepatic, pancreatic, bladder, gallbladder lesion object of a subsequent medical diagnosis, including the steps of:- identifying a ROI on the image representative of a region of interest containing a solid lesion or cyst and background- applying a first machine learning algorithm (400), preferably deep learning algorithm, tothe region of interest to classify the ROI as benign or malignant without the hystotype- receiving a plurality of morphological parameters and / or representative of the consistency of the solid lesion or cyst in the ROI, preferably selected from one or more: number of papillae; number of loculi; level of vascularity; contents of the cyst or lesion between liquid (e.g. anechoic, mucin, hemorrhage) and solid; maximum size of the cyst or lesion; maximum size of the solid portion; proportion of the solid part to the area of the cyst; shadow cone; irregularity or regularity- estimating at least one histotype of the cyst or solid lesion in the image on the basis of a second machine learning algorithm (300) receiving as input the plurality of morphological parameters and preferably also receiving data relating to ascites- comparing a benignity label o malignancy label of the histotype to the estimate of the step of applying when the label and the estimate are discordant and the histotype and / or the estimate have confidence percentages generated on the basis of or by the corresponding machine learning algorithm below a corresponding predefined threshold, e.g. the output is not conclusive, executing a disambiguation algorithm (800) receiving as input said plurality of morphological parameters or the ROI to provide a further benign / malignant estimation without historype or receiving a further ROI.

2. The method according to claim 1, wherein the disambiguation algorithm (800) includes the step of automatically opening a dialog box on a user interface to receive fromthe user a new value of at least one of the plurality of morphological parameters and / or representative of the consistency and of re-estimating the histotype.

3. Method according to one of claims 1 or 2, wherein the disambiguation algorithm (800) estimates the ROI as benign or malignant without the histotype on the basis of an algorithm that receives as input the ROI or the plurality of morphological parameters and / or parameters representative of consistency and is different from the first and second machine learning algorithms, preferably an unsupervised machine learning algorithm when the first or second machine learning algorithm is supervised or vice versa.

4. Method according to any of the previous claims, wherein the disambiguation algorithm comprises the step of connecting to an indexed data archive and of providing as output at least the histotype and / or the benign / malignant indication with a percentage of greater incidence, based on the values of the said plurality of morphological and / or consistency parameters previously used in input to the second machine learning algorithm (300).

5. Method according to any of the previous claims, wherein the second algorithm (300) is trained to also receive as input a parameter representative of the ascites.

6. Method according to any of the preceding claims, wherein the first algorithm is trained with images with solid lesions already labeled as benign or malignant.

7. Method according to any of the preceding claims, wherein the plurality of morphological parameters comprises the level of vascularization.

8. Method according to any of the preceding claims, wherein the outputs of the first algorithm and the second algorithm respectively are presented together on a graphical user interface.

9. Method according to any of the preceding claims, wherein said second algorithm is random forest.

Images