Machine learning brain ventricle size quantification via transcranial ultrasound

Abstract

Machine learning techniques of quantifying a patient brain ventricle size are presented. The techniques include: acquiring a patient transcranial ultrasound image, where the transcranial ultrasound image depicts a patient brain ventricle; passing the patient transcranial ultrasound image to a trained machine learning system, where the trained machine learning system produces a patient transcranial ultrasound image segmentation, and where the trained machine learning system is trained with training data that includes a plurality of labeled individual transcranial ultrasound images; determining, from the patient transcranial ultrasound image segmentation, a patient brain ventricle size quantification; and providing the patient brain ventricle size quantification.

Description

MACHINE LEARNING BRAIN VENTRICLE SIZE QUANTIFICATION VIA TRANSCRANIAL ULTRASOUNDRelated Application

[0001] This application claims the benefit of U.S. Provisional Patent Application No. 63 / 650,989, entitled “Machine Learning Brain Ventricle Size Quantification Via Transcranial Ultrasound,” filed May 23, 2024, which is hereby incorporated by reference in its entirety.Field

[0002] This disclosure relates generally to head imaging, e.g., for cerebrospinal fluid disorders.Background

[0003] Hydrocephalus, a buildup of cerebrospinal fluid (CSF), is a disease of the brain and its CSF system, which may cause incontinence, gait unsteadiness, and dementia in the elderly. Hydrocephalus is characterized by the enlargement of the brain ventricles. Hydrocephalus can be treated with shunt valve placement, and patients usually remain shunt dependent for the rest of their life. A ventriculoperitoneal shunt connects the brain ventricle to the peritoneal cavity and leaves the patient with a hole in the skull where the catheter enters. Post shunt placement, patients require ongoing computed tomography (CT) scans to monitor their disease, sometimes ranging in the hundreds; with each acquisition exposing the patient to ionizing radiation. Cumulative exposures increase the risk of tumorigenesis, particularly with exposures in childhood. Alternatively, magnetic resonance imaging (MRI) spares theradiation but is more expensive, requires a longer acquisition, and is less available (e.g., impossible for patients with some implants).Summary

[0004] According to various embodiments, a machine learning method of quantifying a patient brain ventricle size is presented. The method includes: acquiring a patient transcranial ultrasound image, where the transcranial ultrasound image depicts a patient brain ventricle; passing the patient transcranial ultrasound image to a trained machine learning system, where the trained machine learning system produces a patient transcranial ultrasound image segmentation, and where the trained machine learning system is trained with training data including a plurality of labeled individual transcranial ultrasound images; determining, from the patient transcranial ultrasound image segmentation, a patient brain ventricle size quantification; and providing the patient brain ventricle size quantification.

[0005] Various optional features of the above method embodiments include the following. The method may include: repeating the acquiring, the passing, the determining, and the providing, whereby a change in size of the patient brain ventricle is detected; and providing an indication of the change in size of the patient brain ventricle. The method may include treating the patient based on the change in size of the patient brain ventricle, where the treating includes adjusting a patient ventricular shunt flow rate. The patient transcranial ultrasound image may be captured through a sonolucent burr hole cover at a cranium location of a burr hole in a cranium of the patient. The trained machine learning system may include at least one trained machine learning subsystem for each of a plurality of burr hole locations. The plurality of burr hole locations may include: a frontal location, and an occipito-parietal location.The method may include accepting a user input indicative of a selected burr hole location, where the passing includes passing to a trained machine learning subsystem corresponding to the selected burr hole location. The patient transcranial ultrasound image may include a still image from an ultrasound video sequence. The determining may be performed by the trained machine learning system. The patient brain ventricle size quantification may include at least one of: an area, a diameter, or a width.

[0006] According to various embodiments, a system for quantifying a patient brain ventricle size is presented. The system includes: a non-transitory computer readable medium including instructions; and at least one electronic processor that executes the instructions to perform operations including: acquiring a patient transcranial ultrasound image, where the transcranial ultrasound image depicts a patient brain ventricle; passing the patient transcranial ultrasound image to a trained machine learning system, where the trained machine learning system produces a patient transcranial ultrasound image segmentation, and where the trained machine learning system is trained with training data including a plurality of labeled individual transcranial ultrasound images; determining, from the patient transcranial ultrasound image segmentation, a patient brain ventricle size quantification; and providing the patient brain ventricle size quantification.

[0007] Various optional features of the above system embodiments include the following. The operations may include: repeating the acquiring, the passing, the determining, and the providing, whereby a change in size of the patient brain ventricle is detected; and providing an indication of the change in size of the patient brain ventricle. The patient may be treated based on the change in size of the patient brain ventricle, where the treating includes adjusting a patient ventricular shunt flow rate. The patient transcranial ultrasound image may be captured through a sonolucent burrhole cover at a cranium location of a burr hole in a cranium of the patient. The trained machine learning system may include at least one trained machine learning subsystem for each of a plurality of burr hole locations. The plurality of burr hole locations may include: a frontal location, and an occipito-parietal location. The operations may further include accepting a user input indicative of a selected burr hole location, where the passing includes passing to a trained machine learning subsystem corresponding to the selected burr hole location. The patient transcranial ultrasound image may include a still image from an ultrasound video sequence. The determining may be performed by the trained machine learning system. The patient brain ventricle size quantification may include at least one of: an area, a diameter, or a width.

[0008] Combinations, (including multiple dependent combinations) of the above-described elements and those within the specification have been contemplated by the inventors and may be made, except where otherwise indicated or where contradictory.Brief Description of the Drawings

[0009] Various features of the examples can be more fully appreciated, as the same become better understood with reference to the following detailed description of the examples when considered in connection with the accompanying figures, in which:

[0010] Fig. 1 illustrates transcranial ultrasound, according to various embodiments;

[0011] Fig. 2 shows a transcranial ultrasound image, according to various embodiments;

[0012] Fig. 3 shows an original transcranial ultrasound image, a corresponding cropped transcranial ultrasound image, and a corresponding labeled cropped transcranial ultrasound image, according to various embodiments;

[0013] Fig. 4 presents schematic diagrams of a machine learning neural network two-dimensional convolutional layer and a machine learning neural network two-dimensional coordinate convolution layer, as used in two reductions to practice;

[0014] Fig. 5 shows original transcranial hydrocephalus ultrasound images, their ground truth ventricle annotations, segmentations produced by the first reduction to practice, and segmentations produced by the second reduction to practice; and

[0015] Fig. 6 is a flow diagram illustrating a machine learning method of quantifying a patient brain ventricle size, according to various embodiments.Description of the Examples

[0016] Reference will now be made in detail to example implementations, illustrated in the accompanying drawings. Wherever convenient, the same reference numbers will be used throughout the drawings to refer to the same or like parts. In the following description, reference is made to the accompanying drawings that form a part thereof, and in which is shown by way of illustration specific exemplary examples in which the invention may be practiced. These examples are described in sufficient detail to enable those skilled in the art to practice the invention and it is to be understood that other examples may be utilized and that changes may be made without departing from the scope of the invention. The following description is, therefore, merely exemplary.

[0017] Some embodiments utilize transcranial ultrasound, which is a radiation- free, relatively inexpensive, and optionally a point-of-care alternative, to provide headimaging, e.g., for patients with hydrocephalus and other cerebrospinal fluid (CSF) disorders. Some embodiments utilize machine learning (e.g., deep learning) to automate the process of brain ventricle segmentation from transcranial ultrasound images. Some embodiments automatically provide a brain ventricle size determination based on the brain ventricle segmentation.

[0018] A reduction to practice achieved an average Dice Similarity Coefficient (DSC) of 0.690 ± 0.247 using a ll-Net model. Another reduction to practice augmented a ll-Net model with coordinate convolution (CoordConv). In comparison to the reduction to practice without the CoordConv, the augmented reduction to practice achieved a statistically significant improvement in the mean DSC to 0.805 ± 0.051. Thus, some embodiments utilize deep learning in automating hydrocephalus assessment from ultrasound imaging.

[0019] These and other features and advantages are shown and described herein in reference to the figures.

[0020] Fig. 1 illustrates transcranial ultrasound 100, according to various embodiments. As shown in Fig. 1 , a transcranial right frontal burr hole 104 is introduced into a patient’s cranium 102. The burr hole allows for placement of a ventriculoperitoneal shunt 110, which passes CSF from the patient’s brain ventricles to the patient’s peritoneum. The ventriculoperitoneal shunt 110 may be equipped with a valve according to some embodiments. A burr hole cover 106 is implanted in the patient’s cranium 102 to overlay the burr hole 104. The burr hole cover 106 as shown in Fig. 1 , by way of non-limiting example, defines a catheter tunnel 108, which accommodates the ventriculoperitoneal shunt 110. Moreover, the burr hole cover 106 is sonolucent, which preserves a natural ultrasound window. For example, the sonolucent burr hole cover 106 allows for a visualization of the lateral ventricles 112and their septum 114, as well as the third ventricle 116, with transcutaneous ultrasound obtained using an ultrasound transducer 118. The ultrasound transducer 118 is communicatively coupled, e.g., using a wired or wireless connection, to an ultrasound computer, which may be implemented as a dedicated ultrasound computer, a laptop computer, a desktop computer, or a smart phone, according to various embodiments.

[0021] Fig. 2 shows a transcranial ultrasound image 200, according to various embodiments. The ultrasound image 200 depicted in Fig. 2 is of a coronal plane orientation. Visible in the ultrasound image 200 are the right lateral ventricle 202, the left lateral ventricle 204, and the septum 206. The ultrasound image 200 may be captured using an ultrasound transducer and ultrasound computer through a burr hole, as shown and described herein in reference to Fig. 1. According to various embodiments, the ultrasound image 200 may be a still image taken from an ultrasound video.

[0022] Fig. 3 shows an original transcranial ultrasound image 310, a corresponding cropped transcranial ultrasound image 320, and a corresponding labeled cropped transcranial ultrasound image 330, according to various embodiments. The original transcranial ultrasound image 310 depicts an ultrasound image acquired using an ultrasound system that includes an ultrasound transducer and an ultrasound computer, e.g., as shown and described herein in reference to Fig. 1 . The cropped transcranial ultrasound image 320 represents a cropping of the original transcranial ultrasound image 310, e.g., 512 x 512 pixels. As described herein, some embodiments utilize cropped transcranial ultrasound images, such as the cropped transcranial ultrasound image 320. The labeled cropped transcranial ultrasound image 330 represents the cropped transcranial ultrasound image 320 withthe pixels of the right lateral ventricle labeled 332 as such. Note that the cropped labeled transcranial ultrasound image 330 shown in Fig. 3 is representative. For example, embodiments may provide labels of brain ventricle pixels in any of a variety of ways, e.g., as binary encodings (such as binary masks), ventricle pixel coordinates, etc. According to some embodiments, a machine learning system is trained using cropped labeled ultrasound images, such as the labeled cropped transcranial ultrasound image 330.

[0023] The ventricle detection process is inherently difficult due to imagining issues such as, among other things, the poor resolution of the ultrasound images and the various artifacts inherent to the modality (e.g., speckle, refraction, beam width, etc.), as illustrated in Fig. 3. Further, the ventricle detection process is inherently difficult due to anatomical issues, which arise from the wide variations in ventricle shape and size due to the hydrocephalus-induced enlargement. Consequently, automating the segmentation and measurement of brain ventricles from hydrocephalus ultrasound images poses significant challenges, which are overcome by various embodiments as disclosed herein.

[0024] A description of two example non-limiting embodiments that were reduced to practice follows in reference to Figs. 4 and 5. The embodiments used trained machine learning systems to automatically segment a brain ventricle in transcranial ultrasound images acquired through burr holes. A first such embodiment used a trained machine learning system that included a ll-Net deep learning neural network. A second such embodiment used a trained machine learning system that included a ll-Net deep learning neural network modified to include a coordinate convolution layer, CoordConv. The segmentations of the reductions to practice may be used to determine a brain ventricle size quantification as described herein. Further,some embodiments may include an additional machine learning subsystem, e.g., one or more neural network layers, which may be additional layers to those shown and described in reference to the reductions to practice, to directly determine the a brain ventricle size quantification.

[0025] A description of the datasets and pre-processing utilized by the reductions to practice follows. Hydrocephalus patients underwent a clinical assessment and subsequent surgery involving shunt placement and burr hole cover implantation. Transcranial ultrasound images were captured using a Philips system at Johns Hopkins Hospital, an example image of which is shown in Fig. 3 (reference 310). Twenty three ultrasound video sequences from fifteen hydrocephalus patients were collected, with each video sequence being comprised of between 68 and 379 (viable) frames. These ultrasound video sequences were obtained following a standardized imaging protocol that involved capturing the cross-sectional area of the ipsilateral lateral ventricle within the standard frame encompassing the ipsilateral foramen of Monroe.

[0026] Prior to utilization in model training, evaluation, or manual delineation, each image was cropped to dimensions of 512 x 512 pixels, as shown in Fig. 3 (reference 320). Every image within the dataset, 4,093 2D ultrasound images, underwent manual annotation by a proficient observer, followed by review by a neurosurgeon. The manual delineations created masks of the right lateral ventricle, an example of which is depicted in Fig. 3 (reference 330). These manual delineation served as the ground truth labels for both training and evaluating the reductions to practice. Within the dataset, the images within the ultrasound video sequences from eleven patients were used for training the deep network models (3,391 images). The reductions to practice used the images in the ultrasound video sequences from onepatient for validation during training (134 images), and the images in the ultrasound video sequences from the remaining three patients were used in testing (568 images).

[0027] The machine learning system of the first reduction to practice included a two-dimensional ll-Net architecture. The training process involved 100 epochs, employing a batch size of two, and utilizing the Adam optimizer with momentum ([3) set to 0.999 and a learning rate of 1 x 10-5, with a binary cross-entropy loss, and no data augmentation.

[0028] The machine learning system of the second reduction to practice integrated a coordinate convolution layer with the ll-Net model of the first reduction to practice. In comparison to the convolution layer (Fig. 4, reference 402), the coordinate convolution layer (Fig. 4, reference 404) functions by augmenting the input layer with additional channels containing pixel coordinates. In the specific two-dimensional context of the second reduction to practice, this entailed appending two supplementary channels — one containing x-coordinates and the other containing / -coordinates. In particular, the first convolutional layer of each convolutional block within the ll-Net encoding path was replaced with the convolution coordinate layer for the second reduction to practice.

[0029] Fig. 4 presents schematic diagrams of a machine learning neural network two-dimensional convolutional layer 402 and a machine learning neural network two-dimensional coordinate convolution layer 404, as used in the reductions to practice described herein. Note that an embodiment may include multiple such layers or combinations of such layers. The convolutional layer 402 maps from a two- dimensional image (or a representation block from a previous layer in the neural network) to a new representation block. The coordinate convolution layer 404 has the same output signature as the convolutional layer 402. However, in the coordinateconvolution layer, two extra channels are introduced respectively to represent the x and y coordinates of the input two-dimensional images (or representation block from the previous layer in the neural network). The values of the coordinate channels are normalized to the range from [-1 , 1 ],

[0030] A description of the results produced by, and the evaluations of, the reductions to practice, follows.

[0031] Fig. 5 shows original transcranial hydrocephalus ultrasound images 502, their ground truth ventricle annotations 504, segmentations 506 produced by the first reduction to practice, and segmentations 508 produced by the second reduction to practice. Thus, Fig. 5 illustrates qualitative segmentation results produced by the reductions to practice. As is clearly shown in Fig. 5, both reductions to practice produced segmentations that were very close to the ground truths, with the second reduction to practice producing segmentations that were nearly indistinguishable from the ground truths.

[0032] The reductions to practice were quantitatively evaluated using the Dice Similarity Coefficient (DSC) (Equation (1 )) and the symmetric Hausdorff distance (HD) (Equation (2)).

[0033] In Equations (1 ) and (2), S represents the regions of the brain ventricle as segmented by the respective reduction to practice, and T represents the ground truth region of the brain ventricle as manually annotated. The function d represents Euclidean distance. The DSC values are in the range [0, 1 ], where 1 indicates perfect overlap between S and T, and 0 indicates no overlap. HD measures the proximitybetween the segmentation and ground truth, with smaller numbers indicating better agreement. As is customary, the 95% HD is reported herein. The quantitative evaluation of the reductions to practice is summarized in the Table below.Table

[0034] In the Table, all results are presented as the mean with standard deviation across the whole testing dataset. It is evident that the second reduction to practice obtained superior performance, as evidenced by a higher mean DSC score and a lower mean 95% HD score. Notably, the differences between the DSC and 95% HD scores of the first and second reductions to practice were verified to be statistically significant with a Wilcoxon signed rank test (at a 0.05 significance level), with p-values of 0.0213 and 0.0114, respectively.

[0035] Fig. 6 is a flow diagram illustrating a machine learning method 600 of quantifying a patient brain ventricle size, according to various embodiments. The method 600 may be performed using an ultrasound system, e.g., as shown and described herein in reference to Fig. 1 . By way of non-limiting example, the actions of 602 may be performed by an ultrasound transducer, and the actions of 604, 606, 608, and 610 may be performed by an ultrasound system computer or other computer. The actions of 612 may be performed by a clinician, e.g., a neurosurgeon.

[0036] At 602, the method 600 includes acquiring a patient transcranial ultrasound image, where the transcranial ultrasound image depicts a patient brain ventricle. The patient transcranial ultrasound image may be as described herein inreference to Figs. 1 , 2, 3 (e.g., reference 310), and / or 5 (e.g., reference 502). The patient transcranial ultrasound image may be acquired through a sonolucent burr hole cover at a cranium location of a burr hole in a cranium of the patient. The location may be a frontal location or an occipito-parietal location, by way of non-limiting examples. The patient transcranial ultrasound image may be acquired directly from an ultrasound system, or may be acquired from electronic storage of a previously-obtained ultrasound image, according to various non-limiting embodiments.

[0037] At 604, the method 600 includes passing the patient transcranial ultrasound image to a trained machine learning system. By way of non-limiting embodiments, the trained machine learning system may include a deep learning neural network architecture, e.g., as shown and described herein, e.g., in reference to Figs. 3, 4, and / or 5. Alternately, the trained machine learning system may include a transformer architecture. The trained machine learning system produces a patient transcranial ultrasound image segmentation, e.g., as shown and described herein in reference to Fig. 5. The trained machine learning system is trained with training data that includes a plurality of labeled individual transcranial ultrasound images, e.g., as shown and described herein in reference to Figs. 3, 4, and / or 5. The labeled individual transcranial ultrasound images may be from a plurality of different individuals. Each such ultrasound image may be captured through a sonolucent burr hole cover at a cranium location of a burr hole in a respective cranium of an individual. The location may be a frontal location or an occipito-parietal location, by way of non-limiting examples. The labeled individual transcranial ultrasound images may be labeled to indicate one or more ventricles, e.g., at least a ventricle that is closest to the location of a burr hole through which the ultrasound image is obtained. The segmentation produced by the actions of 604 may include a segmentation (e.g., a mask) thatindicates the location of (e.g., pixels) one or more ventricles in the patient transcranial ultrasound image.

[0038] At 606, the method 600 includes determining, from the patient transcranial ultrasound image segmentation, a patient brain ventricle size quantification. By way of non-limiting examples, the brain ventricle size quantification may be a measurement that is indicative of an area of a segmented ventricle from 604, a diameter of a segmented ventricle from 604 (e.g., the longest span from one side of the segmented ventricle from 604 to the other), or a width of a segmented ventricle from 604 (e.g., a measurement from one side of the segmented ventricle from 604 to the other at a predetermined location, such as parallel or perpendicular to the septum). According to some embodiments, the actions of 606 may be performed by the trained machine learning system, e.g., by one or more layers of a trained deep learning system. According to some embodiments, the actions of 606 may be performed by a system other than the trained machine learning system. By way of non-limiting examples, for the latter embodiments, an area of the segmented brain ventricle may be determined by a process that counts pixels in the segmentation, or by a process that determines a length of a span of the segmented brain ventricle, e.g., by counting pixels along the span or otherwise.

[0039] At 608, the method 600 includes providing the patient brain ventricle size quantification. According to some embodiments, the actions of 608 may be performed by displaying the patient brain ventricle size to a user of an ultrasound system, e.g., on a computer monitor of such a system. According to some embodiments, the actions of 608 may be performed by providing the patient brain ventricle size quantification to persistent electronic memory, e.g., of a medical information system.

[0040] At 610, the method 600 includes repeating the acquiring of 602, the passing of 604, the determining of 606, and the providing of 608, such that a change in size value of the patient brain ventricle is detected, e.g., by determining a mathematical difference between the determined patient brain ventricle size quantifications. The repetitions may occur once or more than once. The repetitions may be temporally spaced, e.g., such that they form part of a longitudinal study of the patient. By way of non-limiting example, the repetitions may be separated by a week, a month, two months, three months, four months, six months, nine months, a year, etc. The method 600 may further include providing an indication of the change in size value of the patient brain ventricle at 610. The indication of the change in size value of the patient brain ventricle may be provided, e.g., by displaying the patient brain ventricle size to a user of an ultrasound system or by storage in persistent electronic memory, according to various non-limiting embodiments.

[0041] At 612, the method 600 includes treating the patient based on the change in size value of the patient brain ventricle. Any of a variety of interventions can be performed as a result of the change in ventricle size value. The treating can include adjusting a patient ventricular shunt flow rate. For example, the patient ventricular shunt flow rate may be adjusted invasively. Alternately, or in addition, a surgical intervention may be performed. Note that, according to various embodiments, if the change in size value of the patient brain ventricle detected at 610 is indicative of a lack of change in size, then the patient may be spared imaging, e.g., by a CT or MRI scan.

[0042] Many variations of the disclosed embodiments are possible. For example, an embodiment may include multiple trained machine learning subsystems, each trained using training data that includes a plurality of labeled individualtranscranial ultrasound images captured through a sonolucent burr hole cover at a different burr hole cranium location. Each subsystem may include a system as shown and described herein in reference to Figs. 3, 4, and / or 5. Thus, a trained machine learning system may include at least one trained machine learning subsystem for each of a plurality of burr hole locations. Such locations include, by way of non-limiting example, a frontal location or an occipito-parietal location. For example, a trained machine learning system may include a trained machine learning subsystem trained using ultrasound images captured through a frontal burr hole location, as well as a trained machine learning subsystem trained using ultrasound images captured through an occipito-parietal burr hole location. In use, a user may provide a user input, e.g., through a user interface, indicative of a selected burr hole location that corresponds to the location at which the patient transcranial ultrasound image was captured. The trained machine learning system may pass the patient transcranial image to a trained machine learning subsystem corresponding to the selected burr hole location.

[0043] Thus, embodiments that leverage deep learning for the automation of brain ventricle segmentation from hydrocephalus ultrasound images are disclosed. The reductions to practice demonstrated successful accomplishment of this challenging task.

[0044] Certain examples can be performed using a computer program or set of programs. The computer programs can exist in a variety of forms both active and inactive. For example, the computer programs can exist as software program (s) comprised of program instructions in source code, object code, executable code or other formats; firmware program(s), or hardware description language (HDL) files. Any of the above can be embodied on a transitory or non-transitory computer readablemedium, which include storage devices and signals, in compressed or uncompressed form. Exemplary computer readable storage devices include conventional computer system RAM (random access memory), ROM (read-only memory), EPROM (erasable, programmable ROM), EEPROM (electrically erasable, programmable ROM), flash memory, and magnetic or optical disks or tapes.

[0045] Aspects of the present disclosure are described herein with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the disclosure. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented using computer readable program instructions that are executed by an electronic processor.

[0046] These computer readable program instructions may be provided to a processor of a general-purpose computer, special purpose computer, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the electronic processor of the computer or other programmable data processing apparatus, create means for implementing the functions / acts specified in the flowchart and / or block diagram block or blocks. These computer readable program instructions may also be stored in a computer readable storage medium that can direct a computer, a programmable data processing apparatus, and / or other devices to function in a particular manner, such that the computer readable storage medium having instructions stored therein comprises an article of manufacture including instructions which implement aspects of the function / act specified in the flowchart and / or block diagram block or blocks.

[0047] In embodiments, the computer readable program instructions may be assembler instructions, instruction-set-architecture (ISA) instructions, machine instructions, machine dependent instructions, microcode, firmware instructions, statesetting data, configuration data for integrated circuitry, or either source code or object code written in any combination of one or more programming languages, including an object oriented programming language such as Smalltalk, C++, or the like, and procedural programming languages, such as the C programming language or similar programming languages. The computer readable program instructions may execute entirely on a user's computer, partly on the user's computer, as a stand-alone software package, partly on the user's computer and partly on a remote computer or entirely on the remote computer or server.

[0048] As used herein, the terms “A or B” and “A and / or B” are intended to encompass A, B, or {A and B}. Further, the terms “A, B, or C” and “A, B, and / or C” are intended to encompass single items, pairs of items, or all items, that is, all of: A, B, C, {A and B}, {A and C}, {B and C}, and {A and B and C}. The term “or” as used herein means “and / or.”

[0049] As used herein, language such as “at least one of X, Y, and Z,” “at least one of X, Y, or Z,” “at least one or more of X, Y, and Z,” “at least one or more of X, Y, or Z,” “at least one or more of X, Y, and / or Z,” or “at least one of X, Y, and / or Z,” is intended to be inclusive of both a single item (e.g., just X, or just Y, or just Z) and multiple items (e.g., {X and Y}, {X and Z}, {Y and Z}, or {X, Y, and Z}). The phrase “at least one of” and similar phrases are not intended to convey a requirement that each possible item must be present, although each possible item may be present.

[0050] The techniques presented and claimed herein are referenced and applied to material objects and concrete examples of a practical nature thatdemonstrably improve the present technical field and, as such, are not abstract, intangible or purely theoretical. Further, if any claims appended to the end of this specification contain one or more elements designated as “means for [performing [a function]...” or “step for [performing [a function]...”, it is intended that such elements are to be interpreted under 35 U.S.C. § 112(f). However, for any claims containing elements designated in any other manner, it is intended that such elements are not to be interpreted under 35 U.S.C. § 112(f).

[0051] While the invention has been described with reference to the exemplary examples thereof, those skilled in the art will be able to make various modifications to the described examples without departing from the true spirit and scope. The terms and descriptions used herein are set forth by way of illustration only and are not meant as limitations. In particular, although the method has been described by examples, the steps of the method can be performed in a different order than illustrated or simultaneously. Those skilled in the art will recognize that these and other variations are possible within the spirit and scope as defined in the following claims and their equivalents.

Claims

What is claimed is:1 . A machine learning method of quantifying a patient brain ventricle size, the method comprising: acquiring a patient transcranial ultrasound image, wherein the transcranial ultrasound image depicts a patient brain ventricle; passing the patient transcranial ultrasound image to a trained machine learning system, wherein the trained machine learning system produces a patient transcranial ultrasound image segmentation, and wherein the trained machine learning system is trained with training data comprising a plurality of labeled individual transcranial ultrasound images; determining, from the patient transcranial ultrasound image segmentation, a patient brain ventricle size quantification; and providing the patient brain ventricle size quantification.

2. The method of claim 1 , further comprising: repeating the acquiring, the passing, the determining, and the providing, whereby a change in size of the patient brain ventricle is detected; and providing an indication of the change in size of the patient brain ventricle.

3. The method of claim 2, further comprising treating the patient based on the change in size of the patient brain ventricle, wherein the treating comprises adjusting a patient ventricular shunt flow rate.

4. The method of claim 1 , wherein the patient transcranial ultrasound image is captured through a sonolucent burr hole cover at a cranium location of a burr hole in a cranium of the patient.

5. The method of claim 1 , wherein the trained machine learning system comprises at least one trained machine learning subsystem for each of a plurality of burr hole locations.

6. The method of claim 5, wherein the plurality of burr hole locations comprises: a frontal location, and an occipito-parietal location.

7. The method of claim 5, further comprising accepting a user input indicative of a selected burr hole location, wherein the passing comprises passing to a trained machine learning subsystem corresponding to the selected burr hole location.

8. The method of claim 1 , wherein the patient transcranial ultrasound image comprises a still image from an ultrasound video sequence.

9. The method of claim 1 , wherein the determining is performed by the trained machine learning system.

10. The method of claim 1 , wherein the patient brain ventricle size quantification comprises at least one of: an area, a diameter, or a width.11 . A system for quantifying a patient brain ventricle size, the system comprising: a non-transitory computer readable medium comprising instructions; and at least one electronic processor that executes the instructions to perform operations comprising: acquiring a patient transcranial ultrasound image, wherein the transcranial ultrasound image depicts a patient brain ventricle; passing the patient transcranial ultrasound image to a trained machine learning system, wherein the trained machine learning system produces a patient transcranial ultrasound image segmentation, and wherein the trained machine learning system is trained with training data comprising a plurality of labeled individual transcranial ultrasound images; determining, from the patient transcranial ultrasound image segmentation, a patient brain ventricle size quantification; and providing the patient brain ventricle size quantification.

12. The system of claim 11 , wherein the operations further comprise: repeating the acquiring, the passing, the determining, and the providing, whereby a change in size of the patient brain ventricle is detected; and providing an indication of the change in size of the patient brain ventricle.

13. The system of claim 12, wherein the patient is treated based on the change in size of the patient brain ventricle, wherein the treating comprises adjusting a patient ventricular shunt flow rate.

14. The system of claim 11 , wherein the patient transcranial ultrasound image is captured through a sonolucent burr hole cover at a cranium location of a burr hole in a cranium of the patient.

15. The system of claim 11 , wherein the trained machine learning system comprises at least one trained machine learning subsystem for each of a plurality of burr hole locations.

16. The system of claim 15, wherein the plurality of burr hole locations comprises: a frontal location, and an occipito-parietal location.

17. The system of claim 5, wherein the operations further comprise accepting a user input indicative of a selected burr hole location, wherein the passing comprises passing to a trained machine learning subsystem corresponding to the selected burr hole location.

18. The system of claim 11 , wherein the patient transcranial ultrasound image comprises a still image from an ultrasound video sequence.

19. The system of claim 11 , wherein the determining is performed by the trained machine learning system.

20. The system of claim 11 , wherein the patient brain ventricle size quantification comprises at least one of: an area, a diameter, or a width.

Images