Systems and methods for treating vitreous abnormalities with ultrasound

Abstract

Systems, methods and devices are provided for treating vitreous disorders using focused ultrasound. An ultrasound device, configured for both ultrasound imaging of the eye and the delivery of focused ultrasound energy into the eye, is employed to obtain ultrasound imaging data characterizing at least the vitreous. This image data is autonomously processed, via a deep learning algorithm, to perform object detection and / or segmentation, for the identification of an abnormal region associated with opacity. The therapeutic ultrasound transducer is controlled to deliver therapeutic ultrasound energy to the abnormal vitreous regions to generate cavitation sufficient for disruption of the opacities. The imaging ultrasound transducer and the therapeutic ultrasound transducer can be moved in unison to facilitate coordinated imaging, deep-learning-based vitreous abnormality detection, and targeted ultrasound therapy.

Description

SYSTEMS AND METHODS FOR TREATING VITREOUS ABNORMALITIES WITH ULTRASOUNDCROSS-REFERENCE TO RELATED APPLICATION

[0001] This application claims priority to U.S. Provisional Patent Application No. 63 / 571 ,269 titled “SYSTEMS AND METHODS FOR TREATING VITREOUS ABNORMALITIES WITH” and filed on March 28, 2024, the entire contents of which is incorporated herein by reference.BACKGROUND

[0002] The present disclosure relates to the use of therapeutic ultrasound for ophthalmological interventions. More particularly, the present disclosure relates to the use of therapeutic ultrasound for treatment of abnormalities of the vitreous.

[0003] The vitreous body is a transparent, gel-like substance that fills the space between the lens and the retina in the eye. It plays a crucial role in maintaining the shape of the eye and supporting its structures. However, various pathologies can affect the vitreous, leading to vision problems and potential complications.

[0004] One common vitreous pathology is vitreous hemorrhage, which occurs when blood vessels within the vitreous leak blood, often due to trauma, diabetes, or age-related changes. This can result in blurred vision, floaters, and, in severe cases, vision loss. Patients suffering from vitreous hemorrhages that do not naturally resolve are treated surgically, most often with vitrectomy.

[0005] Floaters are another common issue associated with the vitreous. Floaters are tiny clumps of cells or debris that float in the vitreous fluid, casting shadows on the retina and causing visual disturbances. While floaters are usually harmless, they can be bothersome and may indicate underlying eye conditions.

[0006] Floaters are typically managed conservatively, with most cases resolving on their own over time. However, in cases where floaters significantly impair vision or are associated with other eye conditions, surgical interventions such as vitrectomy may be considered.

[0007] Surgical interventions, such as vitrectomy, for treating vitreous abnormalities, are uncomfortable for the patient and may result in infection. Such surgical methods are also invasive and have inherent risks, including retinal detachment and infection, both of which if they were to occur, are severe complications that can lead to permanent blindness. Furthermore, currently allvitreous surgery is performed in operating rooms, which increases the cost of the treatment.SUMMARY

[0008] Systems, methods and devices are provided for treating vitreous disorders using focused ultrasound. An ultrasound device, configured for both ultrasound imaging of the eye and the delivery of focused ultrasound energy into the eye, is employed to obtain ultrasound imaging data characterizing at least the vitreous. This image data is autonomously processed, via a deep learning algorithm, to perform object detection and / or segmentation, for the identification of an abnormal region associated with opacity. The therapeutic ultrasound transducer is controlled to deliver therapeutic ultrasound energy to the abnormal vitreous regions to generate cavitation sufficient for disruption of the opacities. The imaging ultrasound transducer and the therapeutic ultrasound transducer can be moved in unison to facilitate coordinated imaging, deep-learning-based vitreous abnormality detection, and targeted ultrasound therapy.

[0009] Accordingly, in a first aspect, there is provided a system for treating a vitreous disorder via focused ultrasound therapy, the system comprising: a coupling device capable of contacting an eye of a subject such that a chamber suitable for receiving an acoustic coupling medium is formed above the eye; an ultrasound device capable of docking with the coupling device to acoustically couple a distal ultrasound energy emitting surface of the ultrasound device with the eye, the ultrasound device comprising: an imaging ultrasound transducer array, the imaging ultrasound transducer array being controllable, when the ultrasound device is docked with the coupling device and acoustically coupled to the eye, to image at least a portion of the eye; and a therapeutic ultrasound transducer, the therapeutic ultrasound transducer being controllable, when the ultrasound device is docked with the coupling device and acoustically coupled to the eye, to direct focused ultrasound energy within the eye; andcontrol circuitry operatively coupled to the ultrasound device, the control circuitry comprising a processor and memory, the memory comprising instructions executable by the processor for performing operations comprising: a) controlling the imaging ultrasound transducer array to obtain ultrasound image data characterizing at least a portion of a vitreous of the eye; b) employing a deep learning algorithm to process the ultrasound image data and autonomously identify an abnormal region within the vitreous; and c) controlling the ultrasound device to deliver the focused ultrasound energy to the abnormal region with conditions suitable for generating cavitation.

[0010] In one example implementation of the system, the therapeutic ultrasound transducer is a therapeutic ultrasound transducer array, and wherein the control circuitry is configured to control the therapeutic ultrasound transducer array to focus the focused ultrasound energy at the abnormal region.

[0011] In one example implementation of the system, the control circuitry is further configured such that the deep learning algorithm is a vitreous abnormality deep learning algorithm trained to autonomously identify the abnormal region within segmented vitreous image data associated with the vitreous, and wherein employing the deep learning algorithm to process the ultrasound image data and autonomously identify the abnormal region within the vitreous comprises operations including: processing the ultrasound image data to segment a vitreous region corresponding to the vitreous and employing the vitreous region to obtain the segmented vitreous image data from the ultrasound image data; and employing the vitreous abnormality deep learning algorithm to process the segmented vitreous image data and autonomously identify the abnormal region within the vitreous.

[0012] In one example implementation of the system, the control circuitry is further configured to perform operations comprising: determining that the abnormal region resides within the vitreous region and beyond a pre-determined distance from a boundary of the vitreous region prior to controlling the ultrasound device to direct the focused ultrasound energy to the abnormal region with conditions suitable for generating cavitation.

[0013] In one example implementation of the system, the control circuitry is further configured such that processing the ultrasound image data to segment the vitreous region corresponding to the vitreous comprises employing a vitreous segmentation deep learning algorithm to process the ultrasound image data toautonomously segment the vitreous region. The vitreous segmentation deep learning algorithm may be capable of intraoperative execution in real-time or near- real-time, with a latency of less than 1 second. The vitreous abnormality deep learning algorithm may be capable of intraoperative execution in real-time or near- real-time, with a latency of less than 1 second.

[0014] In one example implementation of the system, the control circuitry is further configured such that processing the ultrasound image data to segment the vitreous region and obtain the segmented vitreous image data from the ultrasound image data is performed in the absence of deep learning.

[0015] In one example implementation of the system, the abnormal region is a first abnormal region, and wherein the control circuitry is further configured to perform the following operations after controlling the ultrasound device to direct the focused ultrasound energy to the first abnormal region with conditions suitable for generating cavitation: controlling the imaging ultrasound transducer array to obtain additional ultrasound image data; employing the deep learning algorithm to process the additional ultrasound image data to identify a second abnormal region requiring further insonification; and controlling the ultrasound device to direct the focused ultrasound energy to the second abnormal region with conditions suitable for generating cavitation. The control circuitry may be further configured to constrain the second abnormal region to reside within the first abnormal region.

[0016] In one example implementation of the system, the control circuitry is further configured such that the deep learning algorithm is further configured to: classify a vitreous abnormality present within the abnormal region; and determine, according to a classification of the vitreous abnormality, ultrasound parameters suitable for achieving disruption, via cavitation, of the vitreous abnormality present within the abnormal region; and wherein the ultrasound device is controlled to deliver the focused ultrasound energy to the abnormal region according to the ultrasound parameters.

[0017] In one example implementation of the system, the control circuitry is further configured to perform operations comprising: after identifying the abnormal region and prior to controlling the ultrasound device to direct the focused ultrasound energy to the abnormal region: displaying, on a user interface, an image identifying the abnormal region within the vitreous; and receiving input from an operator authorizing focused ultrasound treatment of the abnormal region.

[0018] In one example implementation of the system, the imaging ultrasound transducer array and the therapeutic ultrasound transducer are spatially aligned to insonify a common planar region during ultrasound imaging and ultrasound therapy, and wherein the imaging ultrasound transducer array is controllable to generate a two-dimensional image slice characterizing the common planar region; wherein the imaging ultrasound transducer array and the therapeutic ultrasound transducer are supported by a support structure, and wherein the imaging ultrasound transducer array, the therapeutic ultrasound transducer array, and the support structure together form an ultrasound transducer assembly, the ultrasound transducer assembly being movable to vary a location of the common planar region within the eye; wherein the control circuitry is configured to: after the ultrasound transducer assembly is moved, such that the common planar region is moved to a different location within the eye, repeat steps a) to c), thereby facilitating treatment of vitreous abnormalities associated with the different location.

[0019] The ultrasound device may include an ultrasound transducer assembly positioning mechanism for moving the ultrasound transducer assembly, and wherein the control circuitry is configured to control the ultrasound transducer assembly positioning mechanism to move the ultrasound transducer assembly such that the common planar region is moved to the different location. The control circuitry may be configured to control the ultrasound transducer assembly positioning mechanism to move the ultrasound transducer assembly such that the common planar region is moved to the different location after receiving input selecting the different location. The control circuitry may be configured to control the ultrasound transducer assembly positioning mechanism to autonomously move the ultrasound transducer assembly such that the different location of the common planar region is adjacent to a current location of the common planar region. The control circuitry may be configured such to provide an alert when, as a consequence of motion of the ultrasound transducer assembly, the common planar region is moved to a previous location that has been previously treated with focused ultrasound energy.

[0020] In another aspect, there is provided a system for treating a vitreous disorder via focused ultrasound therapy, the system comprising: a coupling device capable of contacting an eye of a subject such that a chamber suitable for receiving an acoustic coupling medium is formed above the eye;an ultrasound device capable of docking with the coupling device to acoustically couple a distal ultrasound energy emitting surface of the ultrasound device with the eye, the ultrasound device comprising: an imaging ultrasound transducer array, the imaging ultrasound transducer array being controllable, when the ultrasound device is docked with the coupling device and acoustically coupled to the eye, to image at least a portion of the eye; a therapeutic ultrasound transducer array, the therapeutic ultrasound transducer array being controllable, when the ultrasound device is docked with the coupling device and acoustically coupled to the eye, to direct focused ultrasound energy within the eye; the imaging ultrasound transducer array and the therapeutic ultrasound transducer array being spatially aligned to insonify a common planar region during ultrasound imaging and ultrasound therapy, and wherein the imaging ultrasound transducer array is controllable to generate a two-dimensional image slice characterizing the common planar region; wherein the imaging ultrasound transducer array and the therapeutic ultrasound transducer array are supported by a support structure, and wherein the imaging ultrasound transducer array, the therapeutic ultrasound transducer array, and the support structure together form an ultrasound transducer assembly, the ultrasound transducer assembly being movable to vary a location of the common planar region within the eye; and an ultrasound transducer assembly positioning mechanism for moving the ultrasound transducer assembly; and control circuitry operatively coupled to the ultrasound device, the control circuitry comprising a processor and memory, the memory comprising instructions executable by the processor for performing operations comprising: controlling the ultrasound transducer assembly positioning mechanism to move the ultrasound transducer assembly such that the common planar region is moved among of a plurality of locations within the eye, and such that the following operations are performed at each location: controlling the imaging ultrasound transducer array to obtain ultrasound image data characterizing a portion of the vitreous;employing a deep learning algorithm to process the ultrasound image data and autonomously identify an abnormal region within the portion of the vitreous; and providing a user interface enabling an operator to view the abnormal regions identified among the plurality of locations within the eye; receiving input from the operator identifying selected abnormal regions for treatment with the focused ultrasound energy; controlling the ultrasound transducer assembly positioning mechanism to move the ultrasound transducer assembly such that the common planar region is moved to each location having a selected abnormal region; and at each location having a selected abnormal region, controlling the ultrasound device to deliver the focused ultrasound energy to the abnormal region with conditions suitable for generating cavitation.

[0021] In another aspect, there is provided a method of treating a vitreous disorder via focused ultrasound therapy, the system comprising: providing a coupling device capable of contacting an eye of a subject such that a chamber suitable for receiving an acoustic coupling medium is formed above the eye; docking an ultrasound device with the coupling device to acoustically couple a distal ultrasound energy emitting surface of the ultrasound device with the eye, the ultrasound device comprising: an imaging ultrasound transducer array, the imaging ultrasound transducer array being controllable, when the ultrasound device is docked with the coupling device and acoustically coupled to the eye, to image at least a portion of the eye; and a therapeutic ultrasound transducer array, the therapeutic ultrasound transducer array being controllable, when the ultrasound device is docked with the coupling device and acoustically coupled to the eye, to direct focused ultrasound energy within the eye; and the method further comprising: a) controlling the imaging ultrasound transducer array to obtain ultrasound image data characterizing at least a portion of a vitreous of the eye; b) employing a deep learning algorithm to process the ultrasound image data and autonomously identify an abnormal region within the vitreous; andc) controlling the ultrasound device to deliver the focused ultrasound energy to the abnormal region with conditions suitable for generating cavitation.

[0022] A further understanding of the functional and advantageous aspects of the disclosure can be realized by reference to the following detailed description and drawings.BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Embodiments are described with reference to the accompanying drawings. In the drawings, like reference numbers can indicate identical or functionally similar elements.

[0024] FIG. 1 is a schematic of an example system for treating vitreous abnormalities using focused ultrasound.

[0025] FIGS. 2A, 2B and 2C show example configurations of a combined imaging and therapeutic ultrasound device for treating vitreous abnormalities.

[0026] FIGS. 3A, 3B and 3C show example variations of the ultrasound devices of FIGS. 2A, 2B and 2C, respectively, in which an ophthalmoscope and associated light source are included.

[0027] FIG. 4 schematically illustrates an example control and processing circuitry for controlling an ultrasound system for treating vitreous abnormalities using focused ultrasound.

[0028] FIG. 5 shows the overall architecture of the VGG-UNet deep learning algorithm.DETAILED DESCRIPTION

[0029] Various embodiments and aspects of the disclosure will be described with reference to details discussed below. The following description and drawings are illustrative of the disclosure and are not to be construed as limiting the disclosure. Numerous specific details are described to provide a thorough understanding of various embodiments of the present disclosure. However, in certain instances, well- known or conventional details are not described in order to provide a concise discussion of embodiments of the present disclosure.

[0030] As used herein, the terms “comprises” and “comprising” are to be construed as being inclusive and open ended, and not exclusive. Specifically, when used in the specification and claims, the terms “comprises” and “comprising” andvariations thereof mean the specified features, steps or components are included. These terms are not to be interpreted to exclude the presence of other features, steps or components.

[0031] As used herein, the term “exemplary” means “serving as an example, instance, or illustration,” and should not be construed as preferred or advantageous over other configurations disclosed herein.

[0032] As used herein, the terms “about” and “approximately” are meant to cover variations that may exist in the upper and lower limits of the ranges of values, such as variations in properties, parameters, and dimensions. Unless otherwise specified, the terms “about” and “approximately” mean plus or minus 25 percent or less.

[0033] It is to be understood that unless otherwise specified, any specified range or group is as a shorthand way of referring to each and every member of a range or group individually, as well as each and every possible sub-range or sub-group encompassed therein and similarly with respect to any sub-ranges or sub-groups therein. Unless otherwise specified, the present disclosure relates to and explicitly incorporates each and every specific member and combination of sub-ranges or subgroups.

[0034] As used herein, the term "on the order of", when used in conjunction with a quantity or parameter, refers to a range spanning approximately one tenth to ten times the stated quantity or parameter.

[0035] Referring now to FIG. 1 , an example system is disclosed for the treatment of vitreous disorders via focused ultrasound. The example system includes an ultrasound device 100 that includes an imaging ultrasound transducer 110 and a therapeutic ultrasound transducer 120. In some example implementations the imaging ultrasound transducer 110 and therapeutic ultrasound transducer 120 are mechanically supported by a common mechanical support structure. Together, they define what is referred to herein as an ultrasound transducer assembly. Although not shown in the figure, the ultrasound device 100 may also include a housing, within which the ultrasound transducer assembly is supported.

[0036] The figure also shows a coupling device 130 capable of contacting an eye of a subject and forming a chamber above the eye that is suitable for receiving an acoustic coupling medium. The ultrasound device 100 is capable of docking with the coupling device 130 to acoustically couple a distal ultrasound energy emitting surface (or surfaces) of the imaging ultrasound transducer 110 and therapeuticultrasound transducer 120 with the eye through the acoustic coupling medium. The acoustic coupling medium may be, for example, a gel or liquid. The coupling device 130 may be shaped to secure the ultrasound device 100 in place during a procedure, such that the distal energy emitting surface (or surfaces) of the imaging ultrasound transducer 110 and therapeutic ultrasound transducer 120 are fully submerged.

[0037] A lower portion of the coupling device 130 may include a thin layer, such as a thin layer of plastic or a silicone membrane, to facilitate the coupling of acoustic energy from the acoustic medium into the eye. As also shown in the figure, a layer of coupling medium, such as an ultrasound gel, may be applied between a membrane of the coupling device 130 and the eye of the patient to ensure proper acoustic coupling. Accordingly, the ultrasound device 100 and its docking with the coupling device 130 ensures proper contact, alignment and acoustic coupling between the transducers and the eye of the patient, thus facilitating the delivery of focused ultrasound to a target area while minimizing energy loss.

[0038] A degasser 140 may be provided in fluid communication with the coupling medium residing within the coupling device 130 to remove air bubbles or dissolved gases from the coupling medium, thereby ensuring the efficiency and safety of the ultrasound transmission. By eliminating bubbles or gas pockets, acoustic coupling between the transducer and the eye of the patient is improved, allowing for more effective transmission of focused ultrasound energy.

[0039] The imaging ultrasound transducer and the therapeutic ultrasound transducer may be supported (e.g. rigidly supported by a support / frame) such that an imaging axis of the imaging ultrasound transducer 110 is colinear with a therapeutic beam delivery axis of the therapeutic ultrasound transducer 120 for treatment control and monitoring. It will be understood that an imaging axis may be, for example, a beam propagation axis associated with a single-element imaging transducer that is mechanically scanned to collect image data, or may be, for example, an axis associated with an array of imaging ultrasound transducers, such as a central axis perpendicular to a 1 D or 2D imaging ultrasound transducer array. Likewise, a therapeutic beam delivery axis may be, for example, a beam propagation axis associated with a single-element therapeutic transducer that is mechanically scanned to delivery ultrasound therapy to different regions of the eye, or may be, for example, an axis associated with an array of therapeutic ultrasound transducers,such as a central axis perpendicular to a 1 D or 2D therapeutic ultrasound transducer array.

[0040] FIGS. 2A, 2B and 2C illustrate various example configurations of the ultrasound transducer device. FIG. 2A illustrates the case in which the therapeutic ultrasound transducer is a single-element ultrasound transducer 120A, which is movable, e.g. via movement of the ultrasound transducer assembly, to vary a position of a focused ultrasound beam within the eye. FIG. 2B illustrates an example case in which the therapeutic ultrasound transducer is provided in the form of an ultrasound transducer array capable of operation as a phased array. Although not shown, such an ultrasound transducer array may be configured as a 1 D or 2D transducer array. As shown in FIG. 2C, the therapeutic ultrasound transducer 120C may alternatively be provided as an array of single-element ultrasound transducers controlled with relative time delays for achieving suitable focusing.

[0041] In FIGS. 2A, 2B and 2C, the central imaging ultrasound transducer 110, shown in the present non-limiting example as being co-linearly aligned with the therapeutic ultrasound transducer 120A-120C, may be a linear imaging ultrasound transducer array configured for B-mode imaging (i.e. configured to generate a 2D image, e.g. an image slice), or may be a 2D imaging array capable of generating a volumetric image. As shown in the FIGS. 3A-3C, the ultrasound transducer assembly may also include an integrated optical ophthalmoscope 190 and associated light source 195.

[0042] In some example embodiments, the imaging ultrasound transducer 110 is fixed relative to the therapeutic ultrasound transducer 120. In other example embodiments, imaging ultrasound transducer 110 and the therapeutic ultrasound transducer 120 can rotate and / or translate relative to one another (e.g. with 6 degrees of freedom).

[0043] In some example embodiments, the ultrasound transducer assembly is manually movable, for example, with up to 6 degrees of freedom (3 translations and 3 rotations). In other example embodiments, the ultrasound transducer assembly is autonomously movable, for example, via a motorized-arm system, e.g. with up to 6 degrees of freedom (3 translations and 3 rotations). Such an embodiment is illustrated in FIG. 1 , which shows a motorized arm 150 (e.g. a set of mechanically coupled arms that can be actuated to pivot via a set of motors), and an associated motor controller 160.

[0044] The example three-axial positioning system allows precise control of the position and orientation of the imaging ultrasound transducer and the therapy ultrasound transducer. The autonomous control of the position and / or orientation of the ultrasound transducer assembly enables specific areas within the eye to be targeted for treatment, ensuring that the focused ultrasound is directed to the intended tissue volume, also ensuring spatial overlap between the imaging field of view and the transducer focal region as the ultrasound transducer assembly is moved relative to the eye.

[0045] Referring again to FIG. 1 , various example components of an electrical drive subsystem of the therapeutic ultrasound transducer are shown. The example system includes therapeutic ultrasound transducer drive electronics that include, but are not limited to, a function generator 170, a power amplifier 172, and matching circuitry 174 (shown as a matching circuit box). The function generator 170, or more generally, waveform generating circuitry, produces electrical signals that define the ultrasound waveform, and example inputs of the function generator include the type of wave form (sinusoidal), frequency, voltage amplitude, pulse period (or interval), and number of cycles. A function generator can serve as trigger for another function generator, in order to achieve precise control of the treatment duration.

[0046] The power amplifier 172 receives, as input, the electrical signal from the function generator and amplifies it to a level suitable for driving the ultrasound transducer. It ensures that the ultrasound transducer receives sufficient power to create the desired focused ultrasound field for treatment. In some non-limiting example implementations, the output power of the power amplifier may be approximately 800-1500 W, 800-2000 W, or, for example, 800-3000 W for the treatment of vitreous hemorrhage or floaters in vitreous humor.

[0047] The matching circuitry 174 (matching box) is provided to match the impedance of the power amplifier to the impedance of the ultrasound transducer. It is connected in between the power amplifier and the therapeutic transducer. By matching impedances, the matching box helps to efficiently transfer the electrical energy from the power amplifier to the transducer. The matching circuitry 174 is beneficial for generating the desired focused ultrasound field with sufficient power and accuracy. It also provides a level of protection for both the power amplifier and the ultrasound transducer by minimizing reflections that could otherwise damage thecomponents. In some example implementations, the impedance of the matching box may be optimized for 50 ohms, and its phase is optimized for 0°.

[0048] The parameters of the focused ultrasound treatment, such as energy level (pressure, intensity, acoustic power), duration (pulse duration, total treatment duration), and target coordinates (point clouds, area, volume), may be controlled by control and processing circuitry 500 (for example, a PC). The control and processing circuitry 500 may be operably connected to the therapeutic ultrasound transducer driving circuitry.

[0049] Although not shown in the figure, the control and processing circuitry 500 is also connected to the imaging ultrasound transducer 110, optionally through one or more addition transducer circuitry components, such as, but not limited to, beamforming circuitry and a transmit / receive switch.

[0050] The control and processing circuitry 500 may also be connected to the motor controller 160 for autonomous (or operator directed) control of the position and / or orientation of the ultrasound transducer assembly, for example, to synchronize ultrasound imaging and ultrasound therapy with changes in position and / or orientation of the ultrasound transducer assembly. The control and processing circuitry may render, on a display device, a user interface that facilitates the planning, monitoring, and / or adjustment of the ultrasound treatment, for example, intraoperatively, in real-time. The user interface may provide feedback on treatment progress, ensures safety limits are not exceeded, and allows for customization of the therapy based on individual patient needs.

[0051] The therapeutic ultrasound transducer 120 may be operated, for example, according to control signals provided by the control and processing circuitry 500 to the therapeutic ultrasound transducer drive circuitry, to deliver a burst of ultrasound pulses for generating of cavitation (histotripsy) within the vitreous for treating an abnormality within the vitreous, as detected based on processing of image data obtained by from the imaging ultrasound transducer.

[0052] The pulse (or burst) period refers to the time interval between the start of one pulse and the start of the next pulse in a train of pulses (burst). In some example implementations, this value may be 0-1 ms, 0-5 ms, 0-10 ms, or 0-100 ms. Decreasing the pulse (or burst) period leads to more precise control over the cavitation area (treatment zone). The pulse repetition frequency (PRF) is the inverse of the pulse (or burst) period, which may be, for example, 1 ,000-10,000 Hz, 200-10,000 Hz, or 1-10,000 Hz. In some example implementations, the duty cycle may be set to 0-1 % to avoid producing significant thermal effects. The pulse (or burst) duration is the pulse (or burst) period multiplied by the duty cycle. This parameter may be, for example, 0.5-5 ps, 0.5-20 ps, or 0.5-100 ps. The parameters listed above have been found by the present inventors to be suitable for the disruption, via the generation of cavitation, of vitreous opacities including opacities due to vitreous hemorrhage and / or floaters. In the case of floaters, the present inventors the minimum amplitude of negative pressure may exceed that necessary to achieve disruption of a vitreous hemorrhage.

[0053] The therapeutic transducer can be a concavely shaped single-element transducer, a phased-array system incorporating multiple single-element transducers, or a fully populated phased-array system with electronic steering capabilities. Its operating frequency may be, for example, 2-3 MHz, 1-3 MHz, 0.5-3 MHz, or 0.25-3 MHz

[0054] The imaging ultrasound transducer can be a single-element transducer configured for imaging via mechanical sweeping / scanning, or it can be a provided as a phased array. Example operating frequencies for the imaging ultrasound transducer include 10-15 MHz, 10-20 MHz, or 10-25 MHz.

[0055] In order to generate cavitation within the vitreous sufficient for achieving disruption of vitreous opacities such as vitreous hemorrhage and / or floaters, the amplitude of negative pressure may be, for example, within the range of 10-30 MPa, 10-40 MPa, 10-50 MPa, or 10-60 MPa.

[0056] In one example implementation, a suitable driving parameter (e.g. a driving voltage) that is sufficient for treatment can be determined as follows. Prior to commencing treatment, a plurality of test pulses (e.g. a number between 2 and 10 test pulses) are delivered at a selected location of vitreous opacity (e.g. a hemorrhage location corresponding to a blood clot or a floater location) within the vitreous humor. The test pulses are provided with an increasingly negative pressure within the lowest range (10 MPa), which can be adjusted by the input voltage of the ultrasound wave, since the pressure amplitude is positively correlated with the input voltage. The input voltage is (e.g. gradually) increased as the test pulses are delivered, so as to increase the negative pressure achieved at the focus of the therapeutic wave. The imaging ultrasound transducer is employed to perform imaging during the delivery of test pulses in order to monitor for the presence ofcavitation, and to determine the lowest input voltage (or pressure amplitude) that can achieve detectable cavitation (as verified by hyper-echoic clouds on ultrasonic B- scan image), optionally using a deep learning model as described further below. This minimum voltage can then be employed to determine a suitable voltage for use during the delivery of ultrasound treatment. For example, this minimum voltage, or a higher voltage determined according to a prescribed mathematical dependence on the minimum voltage, can be used for the subsequent ultrasound treatment process.

[0057] According to various example embodiments of the present disclosure, one or more deep learning algorithms (models) are employed to process the ultrasound image data obtained by the imaging ultrasound transducer to perform object detection and / or segmentation of vitreous opacities to facilitate targeted and image- guided ultrasound treatment. The therapeutic imaging transducer, when employed as a phased array, can be controlled to position the focus of the therapeutic ultrasound beam within a detected region of vitreous abnormality.

[0058] The choice of deep learning model often involves trade-offs between function (object detection vs. instance segmentation vs. semantic segmentation), speed, performance, and required computational resources. Non-limiting examples of models of different size and capability include the following. U-Net is a convolutional neural network architecture widely used for medical image segmentation tasks, including ophthalmic ultrasound segmentation. Its symmetric encoder-decoder architecture is effective for capturing detailed structures in ultrasound images, facilitating accurate segmentation of eye tissues and abnormalities. DeepLab, with its Atrous convolution mechanism, is well-suited for semantic segmentation tasks in medical imaging. It can efficiently process ophthalmic ultrasound images to identify and segment different eye structures and abnormalities with high accuracy. RetinaNet is a single-stage object detection model designed for detecting objects in images with high precision and speed. It can be adapted for eye abnormality detection in ophthalmic ultrasound images, providing rapid processing and accurate localization of abnormalities.

[0059] YOLOv4 is a state-of-the-art object detection model known for its speed and accuracy. It can be utilized for eye abnormality detection in ophthalmic ultrasound images, offering real-time processing capabilities while accurately identifying and localizing abnormalities within the eye. Faster R-CNN is a popular object detection framework that can be applied to detect abnormalities in ophthalmicultrasound images. By utilizing a Region Proposal Network (RPN) and subsequent bounding box regression, Faster R-CNN can efficiently process ultrasound images and identify regions of interest associated with abnormalities. EfficientNet is a family of convolutional neural network architectures optimized for resource-efficient image processing. It can be utilized for various tasks in ophthalmic ultrasound imaging, including segmentation and abnormality detection, providing a balance between speed and accuracy. Mask R-CNN extends the Faster R-CNN framework to perform instance segmentation in addition to object detection. It can accurately segment different eye structures and abnormalities in ophthalmic ultrasound images, facilitating precise localization and characterization of abnormalities.

[0060] These deep learning algorithms are well suited for rapid image processing tasks in ophthalmology, including ultrasound segmentation and eye abnormality detection. U-Net and DeepLab are primarily used for semantic segmentation, while RetinaNet, YOLOv4, Faster R-CNN, and Mask R-CNN are utilized for object detection and instance segmentation. EfficientNet is a versatile architecture that can be adapted for various tasks, including both semantic and instance segmentation. The choice of algorithm depends on factors such as the specific requirements of the task, available computational resources, and desired trade-offs between speed and accuracy.

[0061] In some example embodiments, the deep learning algorithm, or combination of deep learning algorithms, is selected to facilitate real-time feedback (defined here as either providing inference within one second) to an operator. The ability to perform real-time inference is particularly useful for the intraoperative implementations considered in the present disclosure, and in implementations in which multiple regions of the vitreous are scanned and processed to determine which areas of the vitreous should be treated, as such implementations require rapid feedback from a deep learning algorithm to enable a useful workflow.

[0062] Comparing deep learning models in terms of the speed of their inference mode involves measuring the time it takes for each model to process input data and generate predictions. Small sized (shallow) models which typically have fewer than 1 million parameters (fewer than 10 layers) generally offer faster inference times and real-time prediction response. Medium sized models generally range from 1 million to 100 million parameters (between 10 to 100 layers) strike a balance between model complexity and inference speed.

[0063] Real-time prediction response with medium-sized models can be acceptable for many applications, providing a good compromise between speed and accuracy. Large and deep models have more than 100 million parameters (more than 100 layers) can have slower inference times but can achieve higher predictive performance. Real-time prediction response with large and deep models might not be feasible for applications where low-latency and real-time feedback is needed. Accordingly, in some example implementations, a customized, small deep learning model can be employed when performing object detection and / or segmentation of abnormalities within the vitreous.

[0064] While some models can offer sufficiently accurate performance for detecting vitreous hemorrhages or floaters, others can be better suited for real-time intraoperative use. The choice of model depends on the specific requirements of the application, including the desired level of accuracy, speed, and computational resources available. Additionally, fine-tuning and optimization may be beneficial to tailor the models to the specific use case and ensure optimal performance. Real-time performance is crucial for intraoperative use cases where quick decision-making is essential. Deep learning models that perform object detection, as opposed to semantic segmentation, may be employed to provide fast, real-time inference, while also providing sufficient localization of vitreous abnormalities to facilitate focal ultrasound therapy. These models offer fast inference times, enabling rapid processing of images during surgery.

[0065] In some example embodiments, the deep learning model employed to perform object detection and / or segmentation can be specific to the type of vitreous pathology that is present. Indeed, a dual-model approach that first classifies the type of abnormality and then applies a pathology-specific deep-learning model architecture (e.g. for vitreous hemorrhage vs. vitreous floaters) could potentially enhance accuracy by tailoring the model architecture to each type of abnormality, optimizing performance for detection and segmentation tasks.

[0066] For example, an initial deep learning algorithm may be employed to first classify the type of vitreous abnormality that is present, without localizing the abnormality. A second deep learning model, capable of performing object detection and / or segmentation, may then be employed to determine a vitreous region associated with the abnormality, where the second deep learning model is selected among a set of pathology-specific deep learning models capable of object detectionand / or segmentation. For example, the second deep learning model may be selected from set of deep learning models including a first deep learning model trained to perform object detection and / or segmentation of vitreous hemorrhage regions, and a second deep learning model trained to perform object detection and / or segmentation of floaters in the vitreous.

[0067] The type of deep learning model employed for pathology-specific object detection and / or segmentation may be dependent on the type of pathology (e.g. vitreous hemorrhage vs. floaters). For accurate detection of vitreous hemorrhages, models like RetinaNet, YOLOv4, and Faster R-CNN are suitable due to their ability to detect small and overlapping abnormalities with high precision. These models provide precise bounding box predictions, which can aid in accurately identifying vitreous hemorrhages in images. However, the accuracy of detection may still vary depending on factors such as image quality and the complexity of the scene.

[0068] Detecting floaters in ophthalmic images can be challenging due to their subtle appearance and variability. Models like U-Net and DeepLab, which are specialized for segmentation tasks, may be more appropriate for accurately identifying and segmenting floaters from background tissue. These models excel in capturing detailed structures and can provide pixel-level segmentation, which may be beneficial for accurately delineating floaters.

[0069] Also, in the case of floaters, where multiple instances may exist within an image, it may be beneficial to employ a deep learning model that is capable of performing instance segmentation. Instance segmentation enables the identification and segmentation of individual instances within an image, providing precise delineation of each floater. This level of granularity may be beneficial for accurate detection and characterization of floaters.

[0070] In some example embodiments, it may be beneficial, from a speed and / or performance basis, to employ multiple deep learning algorithms, such that a first deep learning algorithm is employed to perform segmentation of the vitreous to obtain segmented image data associated with the vitreous region, and a second deep learning model is employed to process the segmented vitreous image data to perform object detection and / or segmentation of abnormal regions with the vitreous. Optionally, an intermediate deep learning algorithm can be employed to perform pathology classification on the segmented vitreous image data, prior to performingobject detection and / or segmentation with a pathology-specific second deep learning algorithm.

[0071] Utilizing a two-step approach involving a first deep learning model for vitreous segmentation followed by a second deep learning model for abnormality detection or segmentation could enhance inference speed and accuracy. Segmenting the vitreous image data beforehand allows the subsequent model to focus specifically on abnormality detection within the segmented region, potentially improving efficiency and reducing computational load. Furthermore, segmentation of the vitreous border enables the autonomous determination of spatial safety margins during treatment. For example, the system can determine that an abnormal region identified and localized by the deep learning model resides within the vitreous region and beyond a pre-determined distance from a boundary of the vitreous region prior to controlling said ultrasound device to direct the focused ultrasound energy to the abnormal region with conditions suitable for generating cavitation.

[0072] In an alternative implementation of such a two-step approach that employs initial segmentation of the vitreous followed by the use of a deep learning algorithm to autonomously determine regions having vitreous pathology, the initial vitreous segmentation can be performed using a method that does not employ deep learning. For example, the vitreous region can be segmented using a model that first involves receiving, from a user, a selection of points on the vitreous, and subsequently fitting a circular or elliptical shape the vitreous using border detection methods. The model may be improved by receiving, from the user, points defining the lens, fitting a circular or elliptical shape to the lens, and subtracting the lens region from the initially segmented circular or elliptical vitreous region. The segmented vitreous region may be further improved, for example, by refining the vitreous boundary based changes in intensity along a plurality of lines extending from the center of the initially segmented circular or elliptical vitreous region. An alternative example method of vitreous segmentation from ultrasound images, without using a deep learning algorithm, is described in Cirute et al., An efficient segmentation method for ultrasound images based on a semi-supervised approach and patch-based features, 2011 IEEE International Symposium on Biomedical Imaging: From Nano to Macro, 2011 , DOI: 10.1109 / ISBI.2011.5872564, which employs an algorithm based on a semi-supervised approach (user labels) and the use of image patches as data features.

[0073] In some example embodiments, the deep learning model employed for object detection and / or segmentation may be dependent on the type of therapeutic ultrasound transducer employed for therapy. For example, the choice between a fixed-focus transducer and a transducer array may influence the selection of a suitable deep learning model. A transducer array might provide higher-resolution focusing suitable for more complex deep learning models, while a fixed-focus transducer might be sufficient for simpler deep learning model. The selection of the suitable deep learning model may also be dependent on the properties of the imaging ultrasound transducer. For example, the properties of the imaging ultrasound transducer may influence the selection deep learning models based on factors such as image resolution, depth penetration, and the level of detail required for analysis.

[0074] As described above, a deep learning algorithm may be employed to process the ultrasound image data obtained by the ultrasound imaging transducer and to identify (i.e. locate, via object detection, such as bounding box determination, and / or semantic and / or instance segmentation) one or more regions that exhibit an abnormality, and the regions may be subsequently disrupted via cavitation generated by focused ultrasound energy generated by the therapeutic ultrasound transducer, with the therapeutic ultrasound beam being focused and / or steered to insonify the abnormal regions.

[0075] In some example implementations, after having first identified and delivered therapeutic ultrasound energy to a first abnormal region, the imaging ultrasound transducer array is employed to obtain additional ultrasound image data, and the deep learning model is employed to process the additional ultrasound image data to identify a second abnormal region requiring further insonification. The ultrasound device is then controlled to direct focused ultrasound energy to the second abnormal region with conditions suitable for generating cavitation. The second abnormal region can be constrained to reside within the first abnormal region.

[0076] In some example implementations, the therapeutic ultrasound transducer is autonomously controlled to deliver the therapeutic ultrasound energy to a region of vitreous opacity after having identified the region using a deep learning algorithm. In other example implementations, after identifying the abnormal region, but prior to controlling the ultrasound device to direct the focused ultrasound energy to theabnormal region, an image is displayed on a user interface, the image identifying the abnormal region within the vitreous, and delivery of the focused ultrasound to the abnormal region is only initiated after receiving input from an operator authorizing treatment.

[0077] The imaging ultrasound transducer array and the therapeutic ultrasound transducer can be spatially aligned to insonify a common planar region during ultrasound imaging and ultrasound therapy, with the imaging ultrasound transducer array configured to generate a two-dimensional image slice the common planar region. In some example implementations, the ultrasound transducer assembly (formed by the imaging ultrasound transducer array, the therapeutic ultrasound transducer and a support structure) is movable relative (e.g. relative to a housing) to vary a location of the common planar region within the eye. The ultrasound transducer assembly may be moved one or more times, such that the common planar region is moved to a different location within the eye, thereby facilitating treatment of vitreous abnormalities at different locations within the eye. For example, in some example implementations, the common planar region can be moved among a set of mutually adjacent locations to scan different regions of the eye. The movement of the ultrasound transducer assembly may be manual, semi-automated (e.g. motorized and controlled by an operator), or fully automated. In some example implementations, an alert is generated when, as a consequence of motion of the ultrasound transducer assembly, the common planar region is moved to a previous location that has been previously treated with focused ultrasound energy.

[0078] While the previous example embodiment pertains to an example serial workflow involving performing the steps of imaging, deep-learning-based vitreous abnormality treatment, and delivery of therapeutic ultrasound at each of several locations, it will be understand that other variations of this workflow are contemplated by the present disclosure. For example, in an alternative workflow, regions having vitreous abnormalities may initially be determined by performing ultrasound imaging at the plurality of locations of the common planar region, and subsequently moving the ultrasound transducer assembly to selected locations having identified vitreous abnormalities to deliver the focused ultrasound. In one example implementation, after having first identified vitreous abnormalities by scanning the common planar region among a plurality of locations within the eye, an operator can select, on a user interface, a set of locations for subsequent therapeutic ultrasound treatment. Theultrasound transducer assembly and the generation of therapeutic ultrasound can then be autonomously controlled to deliver focused ultrasound therapy to the selected regions of vitreous abnormality.

[0079] A non-limiting example workflow for employing the system according to FIG. 1 , or variations thereof, for the treatment of vitreous disorders via ultrasound therapy causing cavitation, is described below. An operator (e.g. surgeon) controls the position motorized arm via a user interface to move the ultrasound transducer assembly until a desired imaging slice is obtained. For example, the operator may prefer to initiate treatment when the image slice is aligned with the optic nerve. The user interface identifies this slice as the central slice and enables the user to lock onto this position. The eye (already immobilized) and the ultrasound transducer are docked and held constant.

[0080] On the screen of the user interface, once the eyeball is in view, vitreous opacities will be directly visualized as hyperreflective areas within vitreous. The deep learning algorithm is employed to perform object location and / or segmentation of these vitreous opacities. The vitreous body will also be segmented. This will provide a safety zone, as all treatment will fall well within this region. If a treatment is planned that is outside of this safe zone, the treatment will not commence.

[0081] According to the present non-limiting example workflow, the operator confirms which of the autonomously identified regions within the vitreous are to be treated. Once a single region is confirmed (e.g. via pressing a button or providing another form of input), the delivery of focused ultrasound (sonication) commences.

[0082] The presence of sonication may be indicated or communicated according to one or more example modalities in order to provide one or more levels of feedback to the operator. For example, feedback indicative of sonication could include auditory feedback (e.g. a “buzzing” sound). In other examples, visual feedback can be provided, for example via display of ultrasound image data acquired during therapeutic sonication.

[0083] In some example implementations, an ultrasound image may be segmented to indicate the ultrasound cavitation signal, for example, using a deep learning segmentation algorithm. The segmented cavitation image (e.g. bubble) can be overlayed on the vitreous opacities that were initially imaged and identified, for example, to provide a clear indication of the area that is receiving active treatment relative to the abnormal regions that were initially identified.

[0084] Upon completion of the sonication, the operator, viewing the ultrasound image on the user interface, can confirm that the opacity has been treated based on a change in the ultrasound signature. For example, in dense clotted blood, it is hyperechoic, while if the clot is disrupted and hemolyzed, it becomes hypoechoic. The deep learning algorithm can be employed to confirm a sufficiency of treatment, and / or to identify one or more regions that require further treatment. If any further areas that require treatment are identified by either the operator or autonomously, the desired region can again be highlighted, and treatment can be repeated.

[0085] Once treatment in this specific plane (image slide) is deemed complete (and confirmed by the deep learning algorithm), the operator can then reposition the ultrasound transducer assembly to an adjacent slice (manually or semi- autonomously), or the system can autonomously reposition the ultrasound transducer assembly to the adjacent slice. The operator can decide how many slices to assess and potentially treat. The axial slices may be stored registered so the system can identify slices that have been previously treated.

[0086] In some cases, the system can autonomously scan the ultrasound transducer assembly through the treated slices, and identify if there are any further vitreous opacities that can be treated via focused ultrasound, thereby providing decision support to the operator. Finally, after the procedure is completed, the ultrasound device is removed from the coupling device, and the coupling device removed from the patient.

[0087] FIG. 4 illustrates an example implementation of a control and processing circuitry / hardware 500 for controlling the ultrasound device for the treatment of vitreous disorders via focused ultrasound. The example control and processing hardware 500 may include a processor 510, a memory 515, a system bus 505, one or more input / output devices 520, and a plurality of optional additional devices such as communications interface 525, external storage 530, and a data acquisition interface 535. In one example implementation, a display (not shown) may be employed to provide a user interface for facilitating input to control the operation of the system 500. The display may be directly integrated into a control and processing device (for example, as an embedded display), or may be provided as an external device (for example, an external monitor).

[0088] The control and processing system 500 may include or be connectable to a console that provides an interface for facilitating an operator to control theultrasound device. The console may include, for example, one or more input devices, such, but not limited to, a keypad, mouse, joystick, touchscreen, and may optionally include a display device.

[0089] The methods described herein, such as methods for controlling the position and / or orientation of the ultrasound transducer assembly, and the controlled acquisition of ultrasound image data and the controlled delivery of focused ultrasound, and other example methods described herein, can be implemented via processor 510 and / or memory 515. As shown in FIG. 4, control of the motor controller, deep-learning-based inference for classification, object localization, and / or segmentation, transmit and receive beamforming of imaging ultrasound signals, and control of ultrasound therapy parameters and beam focusing parameters (in the case of a phased array for ultrasound therapy delivery) may be implemented by control and processing circuitry 500, via executable instructions represented as motor control module 550, deep learning algorithm module 560, imaging beamforming module 570, and ultrasound therapy control module 580, respectively. Such executable instructions may be stored, for example, in the memory 515 and / or other internal storage.

[0090] The methods described herein can be partially implemented via hardware logic in processor 510 and partially using the instructions stored in memory 515. Some embodiments may be implemented using processor 510 without additional instructions stored in memory 515. Some embodiments are implemented using the instructions stored in memory 515 for execution by one or more microprocessors. Thus, the disclosure is not limited to a specific configuration of hardware and / or software.

[0091] It is to be understood that the example system shown in the figure is not intended to be limited to the components that may be employed in a given implementation. For example, the system may include one or more additional processors. Furthermore, one or more components of control and processing hardware 500 may be provided as an external component that is interfaced to a processing device. Furthermore, although the bus 505 is depicted as a single connection between all of the components, it will be appreciated that the bus 505 may represent one or more circuits, devices or communication channels which link two or more of the components. For example, the bus 505 may include amotherboard. The control and processing hardware 500 may include many more or less components than those shown.

[0092] Some aspects of the present disclosure can be embodied, at least in part, in software, which, when executed on a computing system, transforms an otherwise generic computing system into a specialty-purpose computing system that is capable of performing the methods disclosed herein, or variations thereof. That is, the techniques can be carried out in a computer system or other data processing system in response to its processor, such as a microprocessor, executing sequences of instructions contained in a memory, such as ROM, volatile RAM, non-volatile memory, cache, magnetic and optical disks, or a remote storage device. Further, the instructions can be downloaded into a computing device over a data network in a form of compiled and linked version. Alternatively, the logic to perform the processes as discussed above could be implemented in additional computer and / or machine- readable media, such as discrete hardware components as large-scale integrated circuits (LSI's), application-specific integrated circuits (ASIC's), or firmware such as electrically erasable programmable read-only memory (EEPROM's) and field- programmable gate arrays (FPGAs).

[0093] A computer readable storage medium can be used to store software and data which when executed by a data processing system causes the system to perform various methods. The executable software and data may be stored in various places including for example ROM, volatile RAM, nonvolatile memory and / or cache. Portions of this software and / or data may be stored in any one of these storage devices. As used herein, the phrases “computer readable material” and “computer readable storage medium” refers to all computer-readable media, except for a transitory propagating signal perse.EXAMPLES

[0094] The following examples are presented to enable those skilled in the art to understand and to practice embodiments of the present disclosure. They should not be considered as a limitation on the scope of the disclosure, but merely as being illustrative and representative thereof.

[0095] In non-limiting example, the deep learning algorithm VGG16 was integrated into the U-Net architecture as the encoder component. The pre-trainedVGG16 layers were used to extract high-level features from the input ultrasound slices, providing a strong foundation for subsequent segmentation by the U-Net decoder. By combining the benefits of the U-Net's skip connections and the featurerich encoder from VGG16, the model aims to achieve accurate and detailed semantic segmentation of various structures of the eye.

[0096] The U-Net architecture is a widely employed neural network architecture for semantic segmentation tasks, particularly in medical image analysis. Its unique design, featuring a contracting path (encoder) and an expansive path (decoder), makes it well-suited for capturing fine-grained details while maintaining contextual information. The U-Net architecture is characterized by the following key components:Contracting Path (Encoder)

[0097] Convolutional Blocks: The contracting path consists of multiple convolutional layers, each followed by a rectified linear unit (ReLU) activation function. These layers are responsible for capturing and encoding features from the input data, gradually reducing the spatial dimensions.

[0098] Pooling Layers: Periodically placed pooling layers (typically max pooling) further downsample the feature maps, helping the network learn hierarchical representations.

[0099] Skip Connections: A critical innovation in U-Net is the introduction of skip connections that connect layers in the contracting path to corresponding layers in the expansive path. These skip connections enable the network to retain high-resolution feature information.Expansive Path (Decoder)

[0100] Transposed Convolutional Layers: In the expansive path, transposed convolutional layers (also known as upsampling or deconvolutional layers) are used to upsample the feature maps, gradually increasing the spatial dimensions.

[0101] Concatenation: The upsampled feature maps are concatenated with the corresponding feature maps from the contracting path. This merging of high-level and low-level features allows the network to refine segmentation boundaries.

[0102] Convolutional Blocks: Similar to the contracting path, the decoder includes convolutional layers with ReLU activations for feature refinement.

[0103] Output Layer: The final layer of the decoder typically consists of a convolutional layer with an appropriate activation function (e.g., softmax for multiclass segmentation) to produce the segmentation mask.

[0104] Backbone Encoder (VGG16)

[0105] The utilization of a pretrained backbone encoder like VGG16 in the U-Net architecture (shown in FIG. 5) offers several advantages for semantic segmentation tasks, including the segmentation of various structures of the eye:

[0106] Transfer Learning: VGG16 has been pre-trained on large-scale image classification tasks, such as ImageNet. As a result, it has learned rich and generalizable features that can be beneficial for a variety of image analysis tasks, including semantic segmentation.

[0107] Feature Extraction: The deeper layers of VGG16 capture hierarchical and abstract features, which can be crucial for segmenting complex structures like optic nerve and opacities. These features can significantly improve the model's ability to recognize and delineate regions of interest.

[0108] Robustness: Pre-trained models like VGG16 have demonstrated robustness to variations in input data, which can be particularly advantageous in medical imaging where data quality can vary.

[0109] Reduction in Training Time: Using a pre-trained backbone allows for faster convergence during training, as the model starts with meaningful feature representations.

[0110] Fine-Tuning: If necessary, you have the option to fine-tune the pre-trained VGG16 layers to adapt them specifically to the task of eye’s structures segmentation, further enhancing performance.

[0111] Compatibility: VGG16's architecture, particularly its convolutional layers, aligns well with the U-Net design, making it straightforward to integrate into the U- Net as an encoder.Model Training

[0112] In this section, we'll discuss the key parameters and configurations used for training your U-Net model for semantic segmentation of ultrasound images, focusing on vitreous, optic nerve, and opacities segmentation.

[0113] Activation Function: The activation function used in the final layer of the model is Softmax. Softmax is often employed for multi-class segmentation tasks as itassigns probability scores to each class, making it suitable for cases where multiple categories, such as vitreous body, optic nerve, opacities inside vitreous, need to be segmented.

[0114] Learning Rate: The learning rate determines the step size taken during the optimization process. A learning rate of 0.0001 was selected, which is a common starting point for many segmentation tasks. It allows for gradual convergence while ensuring stability during training.

[0115] Optimizer: The chosen optimizer is Adam. Adam is an adaptive optimization algorithm that is well-suited for deep learning tasks. It adjusts the learning rate during training, helping the model converge efficiently.

[0116] Loss Function: The loss function is a combination of Dice loss and Focal loss. This combination of loss functions is designed to address the challenges associated with class imbalance in medical image segmentation. The Dice loss measures the overlap between predicted and ground truth masks, while the Focal loss focuses on difficult-to-segment pixels, such as vitreous body boundaries.

[0117] Metrics: During training and evaluation, two primary metrics are used: Intersection over Union (loU) and Dice (F1 -score). loU measures the overlap between predicted and ground truth regions, while the Dice coefficient quantifies the similarity between the two. These metrics provide insight into the model's segmentation accuracy and are particularly valuable in medical image analysis.

[0118] Backbone: The U-Net architecture is augmented with the VGG16 backbone encoder. VGG16 is a pre-trained deep convolutional neural network that has been pre-trained on a vast dataset (Imagenet) for image classification tasks. Its feature extraction capabilities enhance the model's ability to capture complex patterns in ultrasound images.

[0119] Encoder Weights: The VGG16 backbone is initialized with weights pretrained on the ImageNet dataset. These pre-trained weights serve as an excellent starting point for feature extraction, allowing the model to leverage knowledge learned from a wide range of natural images.

[0120] Batch Size: The batch size is set to 8. Batch size determines the number of data samples used in each forward and backward pass during training. A batch size of 8 balances computational efficiency and model stability. Smaller batch sizes can lead to noisy updates, while larger batch sizes may require more memory.

[0121] The specific embodiments described above have been shown by way of example, and it should be understood that these embodiments may be susceptible to various modifications and alternative forms. It should be further understood that the claims are not intended to be limited to the particular forms disclosed, but rather to cover all modifications, equivalents, and alternatives falling within the spirit and scope of this disclosure.

Claims

CLAIMS1 . A system for treating a vitreous disorder via focused ultrasound therapy, the system comprising: a coupling device capable of contacting an eye of a subject such that a chamber suitable for receiving an acoustic coupling medium is formed above the eye; an ultrasound device capable of docking with said coupling device to acoustically couple a distal ultrasound energy emitting surface of said ultrasound device with the eye, said ultrasound device comprising: an imaging ultrasound transducer array, said imaging ultrasound transducer array being controllable, when said ultrasound device is docked with said coupling device and acoustically coupled to the eye, to image at least a portion of the eye; and a therapeutic ultrasound transducer, said therapeutic ultrasound transducer being controllable, when said ultrasound device is docked with said coupling device and acoustically coupled to the eye, to direct focused ultrasound energy within the eye; and control circuitry operatively coupled to said ultrasound device, said control circuitry comprising a processor and memory, said memory comprising instructions executable by said processor for performing operations comprising: a) controlling said imaging ultrasound transducer array to obtain ultrasound image data characterizing at least a portion of a vitreous of the eye; b) employing a deep learning algorithm to process the ultrasound image data and autonomously identify an abnormal region within the vitreous; and c) controlling said ultrasound device to deliver the focused ultrasound energy to the abnormal region with conditions suitable for generating cavitation.

2. The system according to claim 1 wherein said therapeutic ultrasound transducer is a therapeutic ultrasound transducer array, and wherein said control circuitry is configured to control said therapeutic ultrasound transducer array to focus the focused ultrasound energy at the abnormal region.

3. The system according to claim 1 or 2 wherein said control circuitry is further configured such that the deep learning algorithm is a vitreous abnormality deep learning algorithm trained to autonomously identify the abnormal region within segmented vitreous image data associated with the vitreous, and wherein employing the deep learning algorithm to process the ultrasound image data and autonomously identify the abnormal region within the vitreous comprises operations including: processing the ultrasound image data to segment a vitreous region corresponding to the vitreous and employing the vitreous region to obtain the segmented vitreous image data from the ultrasound image data; and employing the vitreous abnormality deep learning algorithm to process the segmented vitreous image data and autonomously identify the abnormal region within the vitreous.

4. The system according to claim 3 wherein said control circuitry is further configured to perform operations comprising: determining that the abnormal region resides within the vitreous region and beyond a pre-determined distance from a boundary of the vitreous region prior to controlling said ultrasound device to direct the focused ultrasound energy to the abnormal region with conditions suitable for generating cavitation.

5. The system according to claim 3 wherein said control circuitry is further configured such that processing the ultrasound image data to segment the vitreous region corresponding to the vitreous comprises employing a vitreous segmentation deep learning algorithm to process the ultrasound image data to autonomously segment the vitreous region.

6. The system according to claim 5 wherein the vitreous segmentation deep learning algorithm is capable of intraoperative execution in real-time or near-real-time, with a latency of less than 1 second.

7. The system according to claim 5 wherein the vitreous abnormality deep learning algorithm is capable of intraoperative execution in real-time or near-real-time, with a latency of less than 1 second.

8. The system according to claim 3 wherein said control circuitry is further configured such that processing the ultrasound image data to segment the vitreous region and obtain the segmented vitreous image data from the ultrasound image data is performed in the absence of deep learning.

9. The system according to claim 2 wherein the abnormal region is a first abnormal region, and wherein said control circuitry is further configured to perform the following operations after controlling said ultrasound device to direct the focused ultrasound energy to the first abnormal region with conditions suitable for generating cavitation: controlling said imaging ultrasound transducer array to obtain additional ultrasound image data; employing the deep learning algorithm to process the additional ultrasound image data to identify a second abnormal region requiring further insonification; and controlling said ultrasound device to direct the focused ultrasound energy to the second abnormal region with conditions suitable for generating cavitation.

10. The system according to claim 9 wherein said control circuitry is further configured to constrain the second abnormal region to reside within the first abnormal region.11 . The system according to any one of claims 1 to 10 wherein said control circuitry is further configured such that: the deep learning algorithm is further configured to: classify a vitreous abnormality present within the abnormal region; and determine, according to a classification of the vitreous abnormality, ultrasound parameters suitable for achieving disruption, via cavitation, of the vitreous abnormality present within the abnormal region; and wherein said ultrasound device is controlled to deliver the focused ultrasound energy to the abnormal region according to the ultrasound parameters.

12. The system according to any one of claims 1 to 11 wherein said control circuitry is further configured to perform operations comprising:after identifying the abnormal region and prior to controlling said ultrasound device to direct the focused ultrasound energy to the abnormal region: displaying, on a user interface, an image identifying the abnormal region within the vitreous; and receiving input from an operator authorizing focused ultrasound treatment of the abnormal region.

13. The system according to any one of claims 1 to 12 wherein said imaging ultrasound transducer array and said therapeutic ultrasound transducer are spatially aligned to insonify a common planar region during ultrasound imaging and ultrasound therapy, and wherein said imaging ultrasound transducer array is controllable to generate a two-dimensional image slice characterizing the common planar region; wherein said imaging ultrasound transducer array and said therapeutic ultrasound transducer are supported by a support structure, and wherein said imaging ultrasound transducer array, said therapeutic ultrasound transducer, and said support structure together form an ultrasound transducer assembly, said ultrasound transducer assembly being movable to vary a location of the common planar region within the eye; wherein said control circuitry is configured to: after said ultrasound transducer assembly is moved, such that the common planar region is moved to a different location within the eye, repeat steps a) to c), thereby facilitating treatment of vitreous abnormalities associated with the different location.

14. The system according to claim 13 wherein said ultrasound device comprises an ultrasound transducer assembly positioning mechanism for moving said ultrasound transducer assembly, and wherein said control circuitry is configured to control said ultrasound transducer assembly positioning mechanism to move said ultrasound transducer assembly such that the common planar region is moved to the different location.

15. The system according to claim 14 wherein said control circuitry is configured to control said ultrasound transducer assembly positioning mechanism to move saidultrasound transducer assembly such that the common planar region is moved to the different location after receiving input selecting the different location.

16. The system according to claim 14 wherein said control circuitry is configured to control said ultrasound transducer assembly positioning mechanism to autonomously move said ultrasound transducer assembly such that the different location of the common planar region is adjacent to a current location of the common planar region.

17. The system according to any one of claims 14 to 16 wherein said control circuitry is configured such to provide an alert when, as a consequence of motion of said ultrasound transducer assembly, the common planar region is moved to a previous location that has been previously treated with focused ultrasound energy.

18. A system for treating a vitreous disorder via focused ultrasound therapy, the system comprising: a coupling device capable of contacting an eye of a subject such that a chamber suitable for receiving an acoustic coupling medium is formed above the eye; an ultrasound device capable of docking with said coupling device to acoustically couple a distal ultrasound energy emitting surface of said ultrasound device with the eye, said ultrasound device comprising: an imaging ultrasound transducer array, said imaging ultrasound transducer array being controllable, when said ultrasound device is docked with said coupling device and acoustically coupled to the eye, to image at least a portion of the eye; a therapeutic ultrasound transducer array, said therapeutic ultrasound transducer array being controllable, when said ultrasound device is docked with said coupling device and acoustically coupled to the eye, to direct focused ultrasound energy within the eye; said imaging ultrasound transducer array and said therapeutic ultrasound transducer array being spatially aligned to insonify a common planar region during ultrasound imaging and ultrasound therapy, and wherein said imagingultrasound transducer array is controllable to generate a two-dimensional image slice characterizing the common planar region; wherein said imaging ultrasound transducer array and said therapeutic ultrasound transducer array are supported by a support structure, and wherein said imaging ultrasound transducer array, said therapeutic ultrasound transducer array, and said support structure together form an ultrasound transducer assembly that is movable to vary a location of the common planar region within the eye; and an ultrasound transducer assembly positioning mechanism for moving said ultrasound transducer assembly; and control circuitry operatively coupled to said ultrasound device, said control circuitry comprising a processor and memory, said memory comprising instructions executable by said processor for performing operations comprising: controlling said ultrasound transducer assembly positioning mechanism to move said ultrasound transducer assembly such that the common planar region is moved among of a plurality of locations within the eye, and such that the following operations are performed at each location: controlling said imaging ultrasound transducer array to obtain ultrasound image data characterizing a portion of the vitreous; employing a deep learning algorithm to process the ultrasound image data and autonomously identify an abnormal region within the portion of the vitreous; and providing a user interface enabling an operator to view the abnormal regions identified among the plurality of locations within the eye; receiving input from the operator identifying selected abnormal regions for treatment with the focused ultrasound energy; controlling said ultrasound transducer assembly positioning mechanism to move said ultrasound transducer assembly such that the common planar region is moved to each location having a selected abnormal region; and at each location having a selected abnormal region, controlling said ultrasound device to deliver the focused ultrasound energy to the abnormal region with conditions suitable for generating cavitation.

19. A method of treating a vitreous disorder via focused ultrasound therapy, the system comprising: providing a coupling device capable of contacting an eye of a subject such that a chamber suitable for receiving an acoustic coupling medium is formed above the eye; docking an ultrasound device with said coupling device to acoustically couple a distal ultrasound energy emitting surface of said ultrasound device with the eye, said ultrasound device comprising: an imaging ultrasound transducer array, said imaging ultrasound transducer array being controllable, when said ultrasound device is docked with said coupling device and acoustically coupled to the eye, to image at least a portion of the eye; and a therapeutic ultrasound transducer array, said therapeutic ultrasound transducer array being controllable, when said ultrasound device is docked with said coupling device and acoustically coupled to the eye, to direct focused ultrasound energy within the eye; and the method further comprising: a) controlling said imaging ultrasound transducer array to obtain ultrasound image data characterizing at least a portion of a vitreous of the eye; b) employing a deep learning algorithm to process the ultrasound image data and autonomously identify an abnormal region within the vitreous; and c) controlling said ultrasound device to deliver the focused ultrasound energy to the abnormal region with conditions suitable for generating cavitation.

20. A system for treating a vitreous disorder via focused ultrasound therapy, the system comprising: an ultrasound device comprising: an imaging ultrasound transducer array, said imaging ultrasound transducer array being controllable, when said ultrasound device is acoustically coupled to the eye, to image at least a portion of the eye; and a therapeutic ultrasound transducer, said therapeutic ultrasound transducer being controllable, when said ultrasound device is acoustically coupled to the eye, to direct focused ultrasound energy within the eye; andcontrol circuitry operatively coupled to said ultrasound device, said control circuitry comprising a processor and memory, said memory comprising instructions executable by said processor for performing operations comprising: a) controlling said imaging ultrasound transducer array to obtain ultrasound image data characterizing at least a portion of a vitreous of the eye; b) employing a deep learning algorithm to process the ultrasound image data and autonomously identify an abnormal region within the vitreous; and c) controlling said ultrasound device to deliver the focused ultrasound energy to the abnormal region with conditions suitable for generating cavitation.21 . A system for treating a vitreous disorder via focused ultrasound therapy, the system comprising: an ultrasound device comprising: an imaging ultrasound transducer array, said imaging ultrasound transducer array being controllable, when said ultrasound device acoustically coupled to the eye, to image at least a portion of the eye; a therapeutic ultrasound transducer array, said therapeutic ultrasound transducer array being controllable, when said ultrasound device acoustically coupled to the eye, to direct focused ultrasound energy within the eye; said imaging ultrasound transducer array and said therapeutic ultrasound transducer array being spatially aligned to insonify a common planar region during ultrasound imaging and ultrasound therapy, and wherein said imaging ultrasound transducer array is controllable to generate a two-dimensional image slice characterizing the common planar region; wherein said imaging ultrasound transducer array and said therapeutic ultrasound transducer array are supported by a support structure, and wherein said imaging ultrasound transducer array, said therapeutic ultrasound transducer array, and said support structure together form an ultrasound transducer assembly that is movable to vary a location of the common planar region within the eye; and an ultrasound transducer assembly positioning mechanism for moving said ultrasound transducer assembly; andcontrol circuitry operatively coupled to said ultrasound device, said control circuitry comprising a processor and memory, said memory comprising instructions executable by said processor for performing operations comprising: controlling said ultrasound transducer assembly positioning mechanism to move said ultrasound transducer assembly such that the common planar region is moved among of a plurality of locations within the eye, and such that the following operations are performed at each location: controlling said imaging ultrasound transducer array to obtain ultrasound image data characterizing a portion of the vitreous; employing a deep learning algorithm to process the ultrasound image data and autonomously identify an abnormal region within the portion of the vitreous; and providing a user interface enabling an operator to view the abnormal regions identified among the plurality of locations within the eye; receiving input from the operator identifying selected abnormal regions for treatment with the focused ultrasound energy; controlling said ultrasound transducer assembly positioning mechanism to move said ultrasound transducer assembly such that the common planar region is moved to each location having a selected abnormal region; and at each location having a selected abnormal region, controlling said ultrasound device to deliver the focused ultrasound energy to the abnormal region with conditions suitable for generating cavitation.

Images