Method and system for determining fetal ventricle volume from diffusion MRI and nmr evaluation method

Abstract

The invention relates to a computer implemented method for determining fetal ventricle volume from diffusion weighted magnetic resonance imaging, referred to as DWI, comprising the steps of: A. acquiring DWI images (112) of the fetal ventricle with a preset voxel height, wherein the acquisition is performed with a single b-value (111) selected in the range between 200 and 1000 s / mm 2 B. selecting ROIs (122, 123, 124, 125) on each DWI image around the ventricle; C. automatically clustering (127) the pixels in the ROIs, thereby obtaining clustered DWI images; and D. calculating the ventricle volume (130) based on the number of pixels in the ROIs of each DWI image and the preset voxel height. The invention further relates to an NMR evaluation system and an NMR evaluation method using the method of the invention.

Description

[0001] Applicants: National Research Council; University of Rome

[0002] Inventors: Silvia Capuani; Mattia Borelli; Giacomo Platese; Maria Giovanna di Trani; Lucia Manganaro

[0003] This invention relates to a computer-implemented method and system for determining fetal ventricular volume based on diffusion-weighted magnetic resonance imaging, and related NMR ventricular volume assessment methods. Background Technology

[0004] In recent years, the advantages and benefits of three-dimensional investigation of tissues and objects of interest using techniques such as magnetic resonance imaging have become a consensus among most experts in the field. General MRI methods in the field of fetuses and newborns can be found in

[13] and

[14] .

[0005] Especially when dealing with the topic of ventricular enlargement (VM, which occurs in about 1 or 2 out of every 1000 cases ([2])), and in terms of technological improvements, the study of the volume of the lateral ventricles must be considered a necessity and an important step towards a more accurate understanding of the pathology. Since VM is associated with abnormal enlargement of the lateral ventricles (see Figure 1 Therefore, its diagnosis is only related to the size measurement of these brain regions. However, to date, the diagnosis is based on the linear size of the three-dimensional object (ventricle) (specifically the diameter of the atrioventricular portion of the ventricle), that is, the diameter of the atrioventricular portion of the fetal lateral ventricle measured using ultrasound or MRI spin-echo T2-weighted images

[11] . Guidelines for the diagnosis of VM recommend a threshold diameter of 10 mm, so fetuses larger than 10 mm are considered to be affected by VM. Ventricular enlargement can have a variety of causes, but in most cases there is no single identifiable cause, and to date there is no model that can explain the evolution of the disease.

[0006] Therefore, because the ventricles are not spatially homogeneous and isotropic, images obtained through two-dimensional projection of the ventricles cannot satisfactorily quantify abnormal growth of the ventricular cavity. This means that classification cannot adequately distinguish the level of postnatal pathology. This is why diameter-based VM diagnosis often fails to reflect expected outcomes and further obscures treatment options for postnatal children. Furthermore, early diagnosis or effective postnatal treatment becomes impossible. Because ventricular diameter is independent of gestational age, current methods cannot distinguish intermediate levels or capture signals of the initial pathological state: thus, many "marginal" cases are difficult to classify.

[0007] It is believed that it is necessary to develop a method for measuring fetal ventricular volume that can overcome current diagnostic limitations.

[0008] As a separate, additional problem, the volume is calculated by manually drawing the outline of the ventricles on radiographs. This outline is usually (slightly) larger than the actual volume of the ventricles and is inaccurate because there is no time for physicians to draw a precise outline. A separate, additional requirement exists to allow for accurate volume calculations, even with a rough outline of the ventricles. This would allow for faster volume calculations and therefore faster diagnosis of VM. Summary of the Invention

[0009] The object of this invention is to provide a computer-implemented method for determining fetal ventricular volume in the diagnosis of ventricular enlargement. Another object of this invention is to provide an NMR system capable of implementing this method, as well as an NMR assessment method for ventricular volume.

[0010] The subject matter of this invention is a computer-implemented method and NMR system for determining fetal ventricular volume according to the appended claims. Another subject matter of this invention is an NMR assessment method for ventricular volume according to the corresponding appended claims. Attached Figure Description

[0011] The present invention is named a VM detector and will now be described for illustrative but not limiting purposes, with particular reference to the illustrations in the accompanying drawings, in which:

[0012] - Figure 1 An anatomical diagram of the fetal ventricles involved in this invention is shown;

[0013] - Figure 2 A flowchart of one embodiment of the method according to the present invention is shown;

[0014] - Figure 3 A graph is shown highlighting the behavior of DWI signals as a function of the b-value of water in the ventricles and brain material;

[0015] - Figure 4 The method according to the invention is shown at b values ​​of 50, 200, and 700 s / mm. 2 The DWI images obtained below (using the k-means tool with k=4);

[0016] - Figure 5 This demonstrates how to manually implement ROI using roiply;

[0017] - Figure 6 Two different ROI selections are shown: (a) the ROI selection does not include other brain regions around the ventricles, and (b) the ROI selection includes these regions. The segmentation results according to the invention are shown on the right.

[0018] - Figure 7A comparison is shown between the real DWI image magnified in the ventricle (right side of the image) in (a) and the segmentation of the present invention performed by k-means (k=4) in (a);

[0019] - Figure 8 The comparison between the real DWI image magnified in the ventricle (right side of the image) in (b) and the segmentation performed by k-means (k=3) in (a) is shown;

[0020] - Figure 9 3D diagrams of the two fetal lateral ventricles realized by the method of the present invention are shown in (a) and (b), respectively;

[0021] - Figure 10 The selected ROI is shown through highly accurate contour mapping of the ventricles;

[0022] - Figure 11 The diagram shows a region of interest (ROI) selected by drawing an imprecise outline of the ventricles, where the outlined region completely encompasses the ventricles, but is larger than them.

[0023] - Figure 12 It shows in Figure 11 The result of segmenting the contour at k=3 using k-means; and

[0024] -exist Figure 13 In (a), (b), and (c), k = 4, 5, and 6 respectively show the values ​​with the same characteristics as... Figure 10 Results for the same ROI.

[0025] It will be clearly pointed out here with reference to the accompanying drawings that elements of different embodiments can be combined together to provide an unlimited number of further embodiments while respecting the technical concept of the invention, as will be readily understood by those skilled in the art from the description.

[0026] Regarding detailed features not described, such as less important elements that are typically used in similar solutions in known technologies, this specification also relates to implementations of known technologies.

[0027] When an element is introduced, it always means that it can be "at least one" or "one or more".

[0028] When a list of elements or features is provided in this specification, it means that the invention is based on an invention that "comprises" or alternatively "is composed of" these elements. Detailed Implementation

[0029] General Methods Introduction

[0030] The computer-implemented VM detector according to the present invention has been developed as part of nuclear magnetic resonance (NMR) neuroimaging studies that are based on the study of molecular diffusion of biological water in tissues and aim to optimize prenatal diagnosis.

[0031] The diagnostic problem addressed by the inventors involves optimizing the objectivity and sensitivity of measurements and automation in prenatal diagnosis of lateral ventricles (VMs). The main objective of this project is to validate a pathological assessment method that demonstrates superiority of lateral ventricle volume analysis over traditional ventricular diameter measurements.

[0032] The VM detector according to the invention proposes a novel measurement that is more correlated with the biological properties of the effective ventricular structure, intended to be a new discriminative factor in diagnosing fetal VM. In fact, the computer-implemented method of the invention begins with an image sequence obtained by diffusion-weighted magnetic resonance imaging (DWI), reconstructing the ventricular volume using a standard MRI protocol routinely performed in suspected pathological cases but not used to determine the aforementioned volume.

[0033] Thus, in the method of the present invention, volume analysis is based on automatic classification by a ventricle reconstruction algorithm starting from the behavior of water diffusion within the brain

[11] , which is generally an excellent classifier for the biological properties of brain tissue. The possibility of including measurements of diameter to perform cross-analysis of comparative data (diameter and volume) is also considered advantageous.

[0034] The prototype was developed in the MATLAB@ environment, and the machine learning algorithms used came from packages downloaded from MATLAB@ itself.

[0035] Examples of general methods

[0036] According to a specific embodiment, the method of the present invention is as follows: Figure 2 It was depicted.

[0037] Although this scheme is described in relation to specific tools such as MatLab@, it should be understood that once the technical concept of the present invention is known, any tool can be used for this purpose.

[0038] Process 100 begins in 110 with image acquisition and preprocessing, wherein a specific NMR acquisition scheme is used in 111 to obtain DWI images 112. Such images are optionally preprocessed in 113 to reduce noise and affect (as is the case with) rearrangement of acquired NMR slices in the prior art.

[0039] After image acquisition and preprocessing, the core method of the present invention is executed in 120, which includes blocks 121-128.

[0040] In step 121, the DWI image is loaded and appropriately converted into a 4D matrix (optional) so that it can be used with MATLAB@ functions, and in particular, the data in MATLAB@ for processing DICOM images (load_untouch_nii).

[0041] In step 122, a coarse selection of the region of interest (ROI) around the ventricles is performed for each slice, as detailed below. The ROI selection result is determined in step 123 to be smaller than the ventricular area, or in step 124 to be larger than the ventricular area, including other brain regions (relative to the same reference scale) with voxel intensities comparable to those of the ventricles. Both results 123 and 124 are considered unfavorable or unsatisfactory. In this case, step 122 (125) is repeated until the ROI selection is considered good.

[0042] In box 126, a ventricular mask is created by multiplying the good ROI by the acquired matrix (corresponding to a 2D matrix for each slice acquired each time). In box 127, the pixels of the ventricular mask are automatically clustered, thus obtaining the correct choice of the number of clusters in box 128. The MATLAB functions kmenes, roiply, and getpts are used in one example.

[0043] Finally, the ventricular volume is calculated in various ways in 130.

[0044] The procedures and experimental details for each of the above boxes are given below.

[0045] collection

[0046] The VM detection program of the present invention provides a specific acquisition scheme 111, 112, including using only a single b value (e.g., b = 700 s / mm). 2 Acquire DWI images.

[0047] Preprocessing

[0048] More specifically, regarding the preprocessing of images 113, according to one aspect of the invention, preprocessing has been performed on an MRI scanner installed at the "Umberto I" hospital in Rome at a specific b-value (e.g., equal to 700 s / mm). 2 The acquired DWI images underwent rearrangement and optional denoising procedures.

[0049] This preprocessing rearrangement and / or denoising is known in itself, but is used for entirely different purposes, such as quantifying ADCs in different brain regions of healthy and VM fetal brains (Front.Phys.7:160.doi:10.3389 / fphy.2019.00160). Specifically, different b values ​​(at least three values, e.g., 50, 200, 700 s / mm) are used in the following manner. 2 The obtained DWI image: For each acquired pixel, the corresponding signal intensity of the DWI image under different b values ​​is fitted to an exponential function to obtain the ADC value of each pixel.

[0050] Post-processing of DWI images

[0051] Following the optional preprocessing described above, the VM detection procedure of the present invention provides, in 120, images that better separate the fluid within the ventricles from the gray / white matter in the fetal brain than conventional MRI. This allows for a better definition of the ventricles' contours and a more accurate estimation of ventricle volume through clustering.

[0052] Typically, each tissue is characterized by different diffusion behaviors of biological water, which depend on the specific tissue microstructure.

[0053] Therefore, the optimal clustering is not known a priori.

[0054] In the case of the fetal brain, using DWI images, the inventors found approximately b = 700 s / mm. 2 The diffusion weight is the optimal value for enhancing the outline of the ventricles relative to the white and gray matter of the fetal brain. It was also found that by using other b-values ​​or different images other than DWI, such as conventional T2-weighted images, it was not possible to distinguish the ventricles from the rest of the brain tissue well.

[0055] In the tests, DWI was performed at 1.5T (Siemens Avanto, Erlangen, Germany). The MRI protocol included DW spin-echo EPI, with TR / TE = 4000 / 79; bandwidth = 1628 Hz / px; matrix = 192x192; FOV = 379x379 mm². In-plane resolution = 2x2 mm², slice thickness = 4 mm, NSA = 2, where the b-value along the three (x, y, z) orthogonal axes was 700 s / mm. 2 .

[0056] The inventors' research shows that clustering remains feasible even when deviating from the aforementioned b-value. More specifically, the first broad applicable range for the b-value is between 200 and 1000 s / mm. 2 Between 600 and 800 s / mm, with the optimal range being between 600 and 800 s / mm. 2Between. Figure 3 In the chart, the exemplary vertical line highlights the signal difference proportional to the image contrast between water in the ventricles and water in the brain tissue.

[0057] When the value is higher than b = 800s / mm 2 At this point, problems begin to emerge due to insufficient SNR. Typically, a sufficiently high-quality DWI for diagnostic purposes is characterized by an SNR greater than 5. At lower b-values, problems begin to arise due to perfusion components (a confounding factor in diffusion contrast).

[0058] exist Figure 4 The image shows the k-means tool with values ​​of 50, 200, and 700 s / mm. 2 The obtained DWI images and their corresponding segmentation at k=4. From the images, we can observe the segmentation differences produced by acquisitions of the same slice from the same acquisition point with different b values, using the same number of clusters in k-means (k=4, which is the optimal k value we found). The b values ​​are 700, 200, and 50 s / mm, respectively. 2 As can be seen from the figure, at b = 50 s / mm 2 At this point, the intensity contrast between pixels is insufficient to distinguish between ventricles and brain tissue. (At b = 200 s / mm) 2 At that time, the situation was slightly better, but for b = 700 s / mm 2 This allows for correct segmentation.

[0059] The aforementioned range discovered by the inventors is specific to distinguishing between ventricles and brain tissue in the fetal brain. Indeed, diffusion-weighted images obtained at different b-values ​​are insufficient to guarantee excellent segmentation in other investigations, especially in the case of heterogeneous tumors (such as glioblastoma). There is a large body of literature on this problem, and several segmentation algorithms have been developed, but they all have significant limitations. The issue is that in the case of heterogeneous tumors (80% of tumors), there are many dynamics associated with different degrees of tumor, and therefore it is difficult to separate all these components simply by using diffusion-weighted images. Instead, in the case of the VM detector, we essentially deal with only two distinct dynamics: the dynamics of free water in the ventricles and the dynamics of impeded and restricted water in the fetal brain material (it is worth noting that the dynamics of water in the adult brain differ from those in the fetus). These two dynamics are surprisingly resolved by the method of the present invention at the aforementioned b-values, a property not shown in the literature.

[0060] The acquisitions described above are in the traditional DICOM world format and can be opened and managed using a specific program called @FSL: the inventors utilize this software to control and verify the results and acquired data in parallel. Due to the load_untouch_nii function above, these images are converted into a 4D matrix of size 192×192×30×1, where 1 is the number of b-values; 30 is the number of axial sections of the fetal head, and the number ranges from 1 to 30 along the fetal head (e.g., along the z-axis); and 192×192 is the size of each 2D section (e.g., in the xy-plane).

[0061] The physical size of each pixel in the slice is 1.97 × 1.97 mm. 2 Each voxel is 4 mm high, and each slice is taken every 4 mm to cover the entire fetus during acquisition (we know these values ​​from the parameters set in the acquisition scheme, where the resolution on the plane defines the cardinality of the voxels (given by dividing the field of view by the acquisition matrix), and the height of the voxels is given by the thickness of each acquisition slice).

[0062] After carefully segmenting the fetal brain within the image itself, the estimation of ventricular volume and diameter is valid. To this end, in order to eliminate potential motion artifacts, the inventors used FSL flirt to rearrange the DWI images; this is a feature of the @FSL software specifically designed to remove noise and motion artifacts from images.

[0063] ROI definition

[0064] Regarding the ROI definition in boxes 112-126, according to the present invention, initially and optionally, the dataset for each slice is reduced to polygons around the ventricles, excluding cerebrospinal fluid (CSF) and the third ventricle: the resulting polygonal regions are the segmented regions of interest (ROIs). This is accomplished using roipoly, which allows users to manually draw and select polygons in the image. Figure 5 The regions selected using roiply are best approximated around the ventricles, so that the pixels transferred to k-means also include other brain regions, maximizing segmentation accuracy. See also: Figure 6 The differences shown are illustrated.

[0065] Even when using algorithms for automatic contour drawing, the present invention will achieve its purpose.

[0066] The pixels of the polygon are then passed to k-means for grouping and segmentation based on the intensity of each pixel, resulting in... Figure 5 The typical results shown are explained below.

[0067] Volume division

[0068] Regarding the segmentation of box 127, the ventricle segmentation was performed automatically using an existing algorithm, which is essentially a machine learning-based algorithm called k-means. This algorithm consists of a clustering operation on a set of elements, which are divided into k distinct clusters. Elements in these clusters are similar to other elements in the same cluster in some attributes, and are different from other elements in different clusters.

[0069] Any other suitable algorithm can be understood to be used in this invention, such as specific methods typically included in general grouping of connection-based clustering (hierarchical clustering), distribution-based clustering, density-based clustering, and grid-based clustering. Other future clustering techniques will be applicable to this invention, as this invention is not itself about such clustering techniques.

[0070] In the case of the DWI matrix, the element intensity of each pixel is different. Therefore, when the algorithm is implemented with the aforementioned b value, it has been found to distinguish them in a way consistent with the biological characteristics of the fetal brain region to which they belong. As a result, ventricle pixels are placed in the same cluster, unlike the clustering of surrounding regions. The resulting segmentation is surprisingly accurate, thus minimizing the error of isolating ventricle pixels from other pixels, such as... Figure 7 As shown, an extreme case (a very small ventricle only a few pixels thick) is illustrated to highlight the precision of the segmentation.

[0071] Different numbers of clusters have been tried to ensure that k-means works as well as possible. The best candidate is k=4, which is also able to... Figure 8 As seen in the image, it shows the relationship with... Figure 5 Segmentation of the same ventricle in the brain with k=3 (initial candidate).

[0072] This result is specific to a particular clustering algorithm known as k-means. Other clustering algorithms will have functions that adjust for the selection of ROIs by taking one or more other parameters as parameters. In fact, ROIs can be selected using very precise ventricular contour mapping, such as... Figure 10 As shown, and in this case, experiments using the specific algorithm k-means demonstrate that the present invention works well starting from k=1, and for k=2 and larger values, the results of the present invention are significant.

[0073] However, regardless of the specific segmentation algorithm, as mentioned above, one of the practical problems in radiography is the availability of time for accurate contour mapping. Most of the time, physicians do not have the time to perform such precise contour mapping. Therefore, an independent but additional synergistic problem addressed by this invention is to allow for rapid segmentation and thus rapid volumetric computation. This is achieved by allowing ROI selection to be performed by cropping DWI images in a manner that includes brain regions other than the ventricles and adjusting the parameters of the segmentation algorithm to achieve sufficiently good segmentation.

[0074] In the case of k-means, such as Figure 6 Top left image or Figure 11 The outline shown has been proven to be acceptablely valid for k values ​​of k=3, such as... Figure 12 As shown (obtained with k=3): the outline of the ventricles is well highlighted from the segmentation and is part of the same ROI from the surrounding pixels.

[0075] exist Figure 13 In (a), (b), and (c), k = 4, 5, and 6 respectively show the values ​​with the same characteristics as... Figure 10 Results for the same ROI. For k=4, we can see that the ventricles are well distinguished from the rest of the pixels, and even from a series of pixels whose intensities appear similar to those of the ventricles at first glance, but are not actually so. This effect could be due to motion artifacts, or because the intensity of the ventricles pixels also contaminates the intensity of neighboring pixels. For k=5, the clustering highlights the outline of the ventricles very well and separates them from the rest of the pixels. For k=6, the k-means segmentation is somewhat confusing (though still acceptable), as there are always more ventricles pixels missing and separated from the core cluster to be assigned to other secondary clusters.

[0076] In summary, based on experiments using k-means and the rough outline above, the acceptable range for k is between 3 and 5, with the optimal value being 4.

[0077] Similar experiments have been conducted using other segmentation algorithms with adjusted clustering parameters.

[0078] Volume calculation

[0079] Regarding the volume calculation in box 127, after segmenting the ventricles, the pixels of interest can be separated from other pixels. The sum of the number of ventricle pixels in each slice gives the total number of pixels in the examined ventricle. Starting from this number, it is possible to calculate the volume by multiplying the total number of pixels by the volume of each voxel (1.97 × 1.97 × 4 mm). 3This is used to extract the volume. In this way, the space between one slice and another is perfectly covered. After ventricle segmentation, any other method is suitable.

[0080] Due to resolution differences, partial volume effects can occur in images, risking overestimation of volume measurements. Partial volume effects typically occur when different types of tissue appear within the same voxel, a common occurrence in the first and last slices. To overcome this problem, it was decided to halve the voxel volume in the first and last slices. The error associated with the measurement is precisely the amount of volume removed. In this way, it is possible to ensure that the true size of the ventricles is within the optimal estimate and its margin of error. Clearly, acquiring DWI with smaller slice thicknesses (e.g., 2 mm or less) and larger matrices (e.g., 256x256 or greater) yields smaller voxels, which helps make the method more sensitive and accurate. In particular, the error in estimating ventricular volume will be significantly reduced. However, increasing the resolution of DWI images (thinner slices and better in-plane resolution) leads to a decrease in the signal-to-noise ratio (SNR). We have developed a VM detector using a 1.5T scanner, and 1.5T is currently the maximum magnetic field strength used for fetal MRI surveys. However, 3T scanners are already in use in Europe and worldwide, allowing for better SNR, as SNR is proportional to magnetic field strength. Therefore, in the long run, VM detectors used with DWI acquisition achieve higher magnetic fields, making them more sensitive and specific in quantifying differences between ventricular volumes.

[0081] exist Figure 9 In the diagram, two examples of three-dimensional ventricular reconstruction performed using the procedure of this invention are shown in (a) and (b), respectively. It is readily apparent that the resulting ventricular shape closely matches the anatomical shape of the ventricles, confirming that this procedure reconstructs the true volume of the ventricles.

[0082] Calculation of the diameter of the ventricles

[0083] Regarding the optional calculation of the diameter, Figure 1Not shown as a process block in section A, a MATLAB function called `getpts` was used to extract the diameter measurement. This function allows the user to select specific points within the image and returns their corresponding coordinates. Using this function on an image previously segmented by k-means, the extreme points of the lateral ventricle diameter can be selected on the largest slice of the highly accurate image segmentation. After selecting the relevant points, the diameter can be simply calculated as the Cartesian distance between them (measured in pixels) multiplied by the linear size of each pixel (1.97 mm). The error associated with the measurement is exactly the linear size of the pixels (1.97 mm). Again, in this case, due to the extreme precision of the segmentation provided by the present invention, the present invention provides better results than diameter calculations performed in the same manner but on the original DWI.

[0084] experiment

[0085] The inventors tested the VM detection method of the present invention in a cohort of healthy subjects and subjects with VM by comparing the results with the volume and diameter of the ventricles. The results showed no statistically significant correlation with diameter (p = 0.05), while the results showed a highly statistically significant correlation with volume (p = 0.0099), as reported in the table below:

[0086] Table 1 – Correlation Coefficient Table

[0087] Calculated volume 0.6842 0.5469 medical diameter 0.5540 0.3567

[0088] Table 2 – P-value Table

[0089] Calculated volume 0.0099 0.0154 medical diameter 0.0495 0.1339

[0090] Furthermore, unlike diameter, the volume of the lateral ventricle is correlated with gestational age. This is an incredible advantage because it allows for early diagnosis with a wealth of available statistical data, enabling specialists to conduct regular and targeted monitoring of fetal brain development in specific prenatal diagnostic contexts. Additionally, the prototype developed allows users to compare volume measurements with diameter measurements (measurements are performed in the same manner, i.e., manually selecting the limiting dimensions of the segmentation the user wants to know), in order to maintain information on current guidelines for VMs while simultaneously utilizing new, more reliable volume-based methods.

[0091] Therefore, according to the present invention, an NMR diagnostic method is provided, comprising the following steps:

[0092] - Provide an NMR system according to the above description;

[0093] - Obtain the gestational age of the fetus;

[0094] - The steps AD of performing the above methods on the fetus using an NMR system; and

[0095] - Selectively (automatically by computer) diagnose conditions of ventricular enlargement based on ventricular volume and gestational age.

[0096] The diagnostic steps of this method can be based on the difference between the volume calculated by the volume calculation step and the ventricular volume of a healthy fetus as a reference curve for gestational age.

[0097] Regarding gestational age, this invention has been tested on a total statistical dataset of 32 patients: at prenatal check-up, 19 patients were declared clinically healthy (HEALTH), and 13 patients were declared clinically ill (ILL). According to the currently provided lateral... Department of Housing A diagnostic protocol using reference values ​​for ventricular diameter was developed for medical diagnosis of specific GA (gestational age) or gestational gestational age. The hypothesis of a correlation between gestational age and ventricular volume was evaluated in both healthy and diseased cases.

[0098] Use Spearman's method to perform relevant tests:

[0099]

[0100] Where D is the distance between the ranks of the two columns and N is the number of elements in the dataset. This method calculates the correlation between two values ​​without making any assumptions about the linearity of the relationship, measuring the correlation of any monotonic relationship between the two quantities.

[0101] For each patient, report the value of the larger of the two ventricles. In healthy patients, volume is positively correlated with GA, as shown below:

[0102] Table 3

[0103] GA: Calculate volume 0.5469 0.0154 GA: medical diameter 0.3567 0.1339

[0104] Table 4

[0105]

[0106]

[0107] According to current literature, the patient diameter measured using the method according to the invention is not correlated with GA (<0.4). The volume of diseased patients is also not correlated with GA (<0.4): this result is because the disease varies in severity depending on the individual patient, rendering time-related hypotheses useless. This disease alters the natural ventricular growth behavior, which under normal conditions is positively correlated with GA.

[0108] The patient dataset is based on the volume calculated using the method according to the invention, the diameter value calculated by physicians using currently effective methods, and the actual pathological development in the first few years of a child's life, referred to as the outcomes in this analysis. The outcomes are divided into five different severity categories: the severity of cerebral edema and the actual symptoms reported after birth.

[0109] 1. VM cured;

[0110] 2. Confirmed as VM, but the child's psychomotor development is normal;

[0111] 3. Confirmed as VM, the child exhibits mild psychomotor developmental delay;

[0112] 4. Confirmed as VM, a severe psychomotor developmental disorder in children, potentially leading to epileptic seizures; and

[0113] 5. If confirmed as VM, the child's psychomotor development is severely impaired, and death may occur (due to various causes).

[0114] Table 5

[0115] Result category: Calculate volume 0.6842 0.0099 Result Category: Medical Diameter 0.5540 0.0495

[0116] Table 6

[0117]

[0118]

[0119] NMR evaluation methods

[0120] According to one aspect of the present invention, an NMR assessment and diagnostic method includes the following steps:

[0121] - Provides an NMR system including an NMR image acquisition device and a computer connected thereto, wherein the computer includes a coding device configured to execute method steps according to the above-described method when the computer is run;

[0122] - Obtain the gestational age of the fetus;

[0123] - The steps of the method of the present invention are performed by a computer, wherein the step of initially selecting the ROI on each DWI image around the ventricles is performed on DWI images acquired from computer memory; and

[0124] - Evaluate the difference between the volume calculated in the previous step and a reference curve of the ventricular volume of a healthy fetus as a function of gestational age.

[0125] Advantages of the present invention

[0126] This invention offers numerous advantages, as mentioned in the preceding sections, which can be summarized as improvements in diagnostic efficiency, accuracy, speed, and enhancement, as well as ease of use of the procedure. In particular, volumetric analysis shows better agreement with results compared to ventricular diameter analysis, making the VM detector a very powerful tool for prenatal diagnosis and postnatal treatment.

[0127] This should never be seen as an attempt to replace experts in diagnosis, but rather as a tool to minimize human assessment errors related to measurement, thereby enabling the use of VM detectors and thus making the diagnosis of VMs as unbiased as possible.

[0128] Since a database of ventricular volumes of healthy fetuses based on gestational age has already been established using the method of this invention, the data evaluated using the above procedure is immediately compared with the set of volume values. Therefore, patients and doctors can output the following information:

[0129] - How much the estimated individual volume differs from the average volume of a healthy person at that gestational age, denoted by σ (therefore σ, 2σ, 3σ, or more); and

[0130] - The value of the ventricular diameter is used to compare the new diagnosis (based on volume calculations) with the conventionally used diagnosis.

[0131] References

[0132] [1] Cardoza JD, Goldstein RB, et al., Exclusion of fetal ventriculomegaly with a single measurement: the width of the lateral ventricular atrium, Radiology. 1988; 169: 711-714.

[0133] [2] Achiron R., Schimmel M. et al., Fetal mild idiopathic lateral ventriculomegaly: is there a correlation with fetal trisomy? Ultrasound in Obstetrics & Gynecology. 1993; 3: 89-92.

[0134] [3]Studholme C. Mapping Fetal Brain Development in utero Using MRI: The Big Bang of Brain Mapping. Annual Review of Biomedical Engineering. 2011; 13: 345-368.

[0135] [4]Orit A. Gleen, MR imaging in fetal brain, Pediater radiol, January 2010; 40(1): 68–81.

[0136] [5]Pediaditis M., Leitgeb N, et al., RF-EMF exposure of fetus and mother during magnetic resonance imaging, Physics in Medicine & Biology. 2008; 53: 7187-7195.

[0137] [6]Budinger T.F., Nuclear magnetic resonance (NMR) in vivo studies: known thresholds for health effects. Journal of Computed Assist Tomography. 1981; 5: 800-811.

[0138] [7]G. S. Chilla et al., Diffusion weighted magnetic resonance imaging and its recent trend—a survey, Quant Imaging Medic Surg. June 2015; 5(3): 407–422.

[0139] [8]Atam P. Dhawan, H. K. Huang, Dae Shik Kim, Principles and Advanced Methods in Medical Imaging and Image Analysis, 2008

[0140] [9] Congenital Hydrocephalus, Adré J. du Plessis et al., Volpe's Neurology of the Newborn (6th ed.), 2018

[0141]

[10] First-Trimester Detection of Fetal Anomalies, Raffaele Napolitano, Aris T Papageorghiou, Twining's Textbook of Fetal Abnormalities (Third Edition), 2015

[0142]

[11] Mascalchi M, Filippi M, Floris R, Fonda C, Gasparotti R, Villari R. Diffusion-weighted MR of the brain: methods and clinical application. Radiol Med 2005 Mar;109(3):155-97.

[0143]

[12] Malcolm I. Levene, Frank A. Chervenak, Fetal and Neonatal Neurology and Neurosurgery (Fourth Edition). 2009

[0144]

[13] JESSICA DUBOIS: "MRI of the Neonatal Brain: A Review ofMethodological Challenges and Neuroscientific Advances", February 18, 2020 (2020-02-18), XP009528962

[0145]

[14] MAKROPOULOS ANTONIOS et al.: “A review on automatic fetal and neonatal brain MRI segmentation”, NEUROIMAGE, ELSEVIER, AMSTERDAM, NL, Vol. 170, 28 June 2017 (2017-06-28), pp. 231-248, XP085367734, ISSN:1053-8119, DOI:10.1016 / J.NEUROIMAGE.

[0146] Preferred embodiments have been described above, and variations of the invention have been proposed. However, it should be understood that those skilled in the art will be able to make modifications and alterations without departing from the relative scope of protection defined by the appended claims.

Claims

1. A computer-implemented method for determining fetal ventricular volume based on diffusion-weighted magnetic resonance imaging (DWI), comprising performing the following steps: A. Acquire DWI images of the fetal ventricles with a preset voxel height (112), wherein, The data acquisition was performed at 200 s / mm. 2 and 1000 s / mm 2 (111) is performed by selecting a single b value within the range between them. B. Select ROIs (122, 123, 124, 125) on each DWI image around the ventricle. C. Perform automatic clustering (127) on the pixels in the ROI to obtain a clustered DWI image; as well as D. The ventricular volume (130) is calculated based on the number of pixels in the ROI of each DWI image and the preset voxel height.

2. The method according to claim 1, wherein, The single b value is at 600 s / mm 2 and 800 s / mm 2 between.

3. The method according to claim 1 or 2, wherein, Step C is performed by a machine learning-based clustering algorithm.

4. The method according to claim 1 or 2, wherein, Step B is performed by cropping the DWI image so that the cropped DWI image includes portions of the brain region that are different from the ventricles.

5. The method according to claim 1 or 2, wherein, Step D is performed by summing the number of ventricular pixels in each DWI image and multiplying the sum by the voxel volume.

6. The method according to claim 1 or 2, wherein, In step A, the DWI image is denoised and rearranged to remove motion artifacts.

7. The method according to claim 1 or 2, wherein, Perform a further step E, in which the diameter of the ventricles is measured on the DWI images of the cluster.

8. The method according to claim 1 or 2, wherein, The clustering in step C is performed using k-means, where the pixels in the ROI are divided into k distinct clusters.

9. The method according to claim 8, wherein, The number of clusters, k, is chosen in the range of 3 to 5.

10. The method according to claim 9, wherein, k＝4。 11. An NMR system comprising an NMR image acquisition device and a computer connected thereto, wherein the computer's apparatus is configured to perform the steps of the method according to any one of claims 1 to 10.

12. An NMR evaluation method, comprising the following steps: - Obtain the gestational age of the fetus; - Perform step BD of the method according to one or more claims 1 to 10 on the fetus by a computer, wherein in step B, a DWI image including a representation of the fetus is initially acquired from a computer memory; as well as - Evaluate the difference between the volume calculated in the previous step and a reference curve of the ventricular volume of a healthy fetus as a function of gestational age.

Images