Tracheal tube guidance with ultrasound input

Abstract

The disclosure relates to systems and methods for tracheal tube guidance with ultrasound input. A method for tracheal tube guidance includes receiving an ultrasound image of a subject's trachea, providing the image as input into a trained model, and receiving as output a segmentation of tissue structure in the ultrasound image. The segmentation is displayed on the ultrasound image, prompting a user confirmation. Upon receiving the user confirmation, the system identifies potentially compatible tracheal tubes based on tracheal measurements from the segmentation.

Description

Attorney Docket No. A0013140W001TRACHEAL TUBE GUIDANCE WITH ULTRASOUND INPUTCROSS-REFERENCE TO RELATED APPLICATION

[0001] This application claims the benefit of U.S. Provisional Patent ApplicationSerial No. 63 / 715,584, filed November 3, 2024, the entire contents of which are incorporated herein by reference.FIELD

[0002] The disclosure is generally related to medical devices and methods including tracheal tubes and ultrasound images.BACKGROUND

[0003] Complications from placement of airway tubes into a patient’s trachea, including tracheostomy tubes and endotracheal tubes, can lead to airway injury, delayed recovery, and increased cost. Selection of a proper size and type of airway tube is one factor that can help prevent these complications. However anatomical variations between patients can limit the ability of a medical caregiver to select the appropriate size or type of tube and achieve a good placement on the first attempt. When a less optimal tube size is initially selected, reintubation may be required, which increases the risk of complications.

[0004] Ultrasound imaging is a noninvasive, rapid, and inexpensive imaging modality that may be able to improve outcomes in these procedures.SUMMARY

[0005] The techniques of this disclosure generally relate to systems and methods for tracheal tube guidance with ultrasound input.

[0006] According to an embodiment, a method is provided for tracheal tube guidance with ultrasound input. The method includes receiving, from a probe, an ultrasound image of a subject’s trachea, and displaying the ultrasound image on a display screen. The method also includes providing the ultrasound image as input into a trained model, and receiving, as output from the trained model, a segmentation of tissue structure in the ultrasound image. The segmentation is displayed on the display screen by overlaying one or more boundaries onto the displayed ultrasound image. The method also includesAttorney Docket No. A0013140W001 prompting a user confirmation of the displayed segmentation, and upon receiving the user confirmation, identifying first and second candidate tracheal tubes. The first and second tracheal tubes are displayed on the display screen.

[0007] According to an embodiment, a method is provided for tracheal tube guidance with ultrasound input. The method includes receiving, from a probe, an ultrasound image of a subject’s trachea and displaying the ultrasound image on a display screen. The method also includes providing the ultrasound image as input into a trained model and receiving, as outputs from the trained model, a segmentation of tissue structure in the ultrasound image and an identification of a tracheal tube in the ultrasound image. The method includes displaying an annotated ultrasound image comprising an indicator of the identified tracheal tube, retrieving a stored image of the subject’s trachea, and displaying the stored image and the annotated ultrasound image side by side on the display screen.

[0008] According to an embodiment, an ultrasound guidance system includes a probe generating an ultrasound signal, a display screen coupled to the probe and displaying an ultrasound image from the ultrasound signal, and a processor communicating with the display screen. The processor is programmed to perform a set of instructions including providing the ultrasound image as an input to a trained machine learning model, receiving, as an output from the trained machine learning model, a segmentation of tissue structure in the ultrasound image, and displaying the segmentation on the display screen by overlaying one or more boundaries onto the displayed ultrasound image. The instructions also include prompting a user confirmation of the displayed segmentation, and upon receiving the user confirmation, identifying first and second candidate tracheal tubes and displaying the identified first and second candidate tracheal tubes on the display screen.

[0009] Examples are set forth in the accompanying drawings and the description below. Other features, objects, and advantages of the techniques described in this disclosure will be apparent from the description and drawings, and from the claims.BRIEF DESCRIPTION OF THE DRAWINGS

[0010] FIG. 1 A illustrates a tracheal tube guidance system with ultrasound input according to an embodiment of the present disclosure.

[0011] FIG. IB illustrates a block diagram of a tracheal tube guidance system according to an embodiment of the present disclosure.Attorney Docket No. A0013140W001

[0012] FIG. 2 A illustrates a displayed ultrasound image of airway anatomy according to an embodiment of the present disclosure.

[0013] FIG. 2B illustrates an ultrasound image with displayed tissue segmentation according to an embodiment of the present disclosure.

[0014] FIG. 3 A illustrates a segmented ultrasound image with a user prompt according to an embodiment of the present disclosure.

[0015] FIG. 3B illustrates a segmented ultrasound image with tracheal measurements according to an embodiment of the present disclosure.

[0016] FIG. 4 illustrates a side-by-side view of an ultrasound image and a depiction of an identified tracheal tube according to an embodiment of the present disclosure.

[0017] FIG. 5A illustrates long axis and short axis ultrasound views according to an embodiment of the present disclosure.

[0018] FIG. 5B illustrates depictions of identified tracheal tubes according to an embodiment of the present disclosure.

[0019] FIG. 6A and 6B illustrate an animation of insertion of a tracheostomy tube, according to an embodiment of the present disclosure.

[0020] FIG. 7 illustrates a method of training a machine learning model that outputs tissue segmentation as a function of ultrasound images, according to an embodiment of the present disclosure.

[0021] FIG. 8 illustrates a method of providing ultrasound guidance according to an embodiment of the present disclosure.

[0022] FIG. 9 illustrates a mapping of tracheal measurements to tracheal tube sizes, according to an embodiment of the present disclosure.DETAILED DESCRIPTION

[0023] The techniques of this disclosure generally relate to systems and methods for guiding selection and placement of tracheal tubes, with input from ultrasound images. In an embodiment, an ultrasound probe is used at the bedside with the patient to provide live, real-time ultrasound images of the patient’s airway anatomy. The system receives live ultrasound images and generates a segmented view identifying the patient’s unique airway anatomy and tracheal measurements. These measurements are used to identify potentially compatible tracheal tubes, and / or to identify a target location for percutaneous entry intoAttorney Docket No. A0013140W001 the trachea. In an embodiment, ultrasound images and segmentation are also generated after the tube is placed, to confirm proper functioning, sizing, and positioning.

[0024] A system 100 for tracheal tube guidance according to an embodiment is illustrated in FIG. 1A. The system 100 includes an ultrasound wand or probe 110 communicating (through cable 108 or wirelessly) with a monitor 112. The monitor 112 includes a portable tablet computer 114 with a display screen 116 and user input 118, such as a button or keypad. The display screen 116 may be a touch screen to enable user inputs by touching the display.

[0025] The ultrasound probe 110 emits ultrasound waves 120 and generates an ultrasound image 122, which is displayed on the display screen 116. When the probe is placed against a patient’s neck, the ultrasound image can depict the patient’s airway anatomy. Two example patients Pl and P2 are shown in FIG. 1, to illustrate variations in individual airway anatomy, including a trachea 124, thyroid cartilage 126, and tracheal cartilage rings 128. A few tracheal measurements will be described in reference to patients Pl, P2 in FIG. 1. The trachea 124 includes a lumen along a long axis A, and the lumen has an inner diameter (DI, D2 respectively for patients Pl, P2) that may be referred to as the tracheal diameter. The trachea is also positioned at a depth below the patient’s skin, which can be measured from the skin to the center of the trachea, shown as line L. This depth is identified as LI, L2 for patients Pl and P2 respectively. Another tracheal measurement is the angle 9 between the lumen axis A and the line L. This angle is 91 for patient Pl and 92 for patient P2. Note that in FIG. 1 the diameter DI, D2 is shown perpendicular to the axis A, but the ultrasound probe may measure perpendicularly to the user’s skin instead. Such a measurement can still be used as an indicator of the patient’s tracheal diameter (the diameter perpendicular to axis A), and / or can be mapped to the tracheal diameter DI, D2 based on the angle 9.

[0026] These measurements can vary significantly between patients, as illustrated by patients Pl and P2. Patient P2 has a larger tracheal diameter D2 (larger than DI), a longer distance L2 (longer than LI), and a larger angle 92 (larger than 91). Typical ranges for these measurements include: 1cm to 3cm for tracheal diameter D; 2cm to 7cm for depth distance L; and 110 to 150 degrees for angle 9. Demographic information such as patient gender, age, or height may not fully predict these anatomical differences.Attorney Docket No. A0013140W001

[0027] System 100 also includes a tracheal tube, such as a tracheostomy tube 130 or an endotracheal tube 132. The tracheostomy tube 130 is intended to be inserted into the patient’s trachea percutaneously or surgically, passing through an opening or incision in the patient’s skin and through the front of the tracheal wall. The tracheostomy tube 130 includes a curved cannula 134 and a tapered balloon cuff 136. While a tapered cuff is shown, tracheostomy tubes may have cuffs with barrel, spherical, curved, or other shapes. The cannula 134 curves approximately 90 degrees from a front (proximal) end, where the cannula is oriented approximately horizontally (to receive an airway tube, filter, or other device), to a distal end (below the cuff) where the cannula is oriented approximately vertically into the trachea. Different tracheostomy tubes have cannulas with different sizes (such as inner and outer diameters), curvatures, lengths (such as from the proximal end to the curve), features (such as fenestrations), and cuff geometries and sizes. The endotracheal tube 132 is intended to be inserted into the patient’s trachea through the mouth or nose and includes an elongated cannula 138 and barrel-shaped balloon cuff 140. While a barrel cuff is shown, endotracheal tubes may have cuffs with tapered, spherical, curved, or other shapes. The cuffs 136, 140 are inflated inside the trachea to form a seal against the inner wall of the trachea, to facilitate the transmission of pressurized air through the tube and into or out of the patient’s lungs. Various endotracheal tubes have cannulas with different sizes (such as inner and outer diameters), curvatures, lengths, features, and cuff geometries and sizes. The tubes 130, 132 in FIG. 1 are provided as examples only, and many other types of cuffed and cuffless airway tubes are available with different geometries, shapes, sizes, and features.

[0028] A block diagram view of system 100 is illustrated in FIG. IB. The system 100 includes the ultrasound wand 110 communicating with the monitor 112. The monitor 112 is a computing device that includes one or more processors 142, the display screen 116, a hardware memory 144, one or more user inputs or interfaces 118, and a wireless transceiver 146. The monitor 112 may be a desktop monitor, a portable tablet, a smartphone, a laptop, or other similar computing device. The monitor 112 communicates with a network 148 which is also in communication with one or more databases 152. The databases 152 may include an EMR (electronic medical record) system, a server with storage, or a cloud storage system, and may store training data, clinical data, validation data, patient data, or other data. The system 100 also includes a trained model 150. TheAttorney Docket No. A0013140W001 dashed lines in FIG. IB indicate that the ML model 150 can be housed in various places - it may be stored in the database 152 and retrieved via the network 148, or it may be stored in its own cloud storage location accessible via the network 148, or it may be stored in the system memory 144 on board the monitor 112. The trained model 150 will be described in further detail below.

[0029] The network 148 may comprise a local area network (LAN), a wide area network (WAN) (e.g., the Internet), a wireless network, a cellular network (e.g., radio access network (RAN)), or any combination thereof, among other examples. Various computing devices, medical devices, databases, and storage devices communicate with each other over the network.

[0030] Automatic segmentation of a live ultrasound image will be described next in connection with FIGs. 2A and 2B. According to an embodiment, a monitor 212 displays an ultrasound image 222A on a display screen 216. The image 222A is received from an ultrasound wand or probe positioned at a patient’s neck, to obtain a view of the patient’s airway anatomy. In FIG. 2 A, the ultrasound image 222 A is a live image from the ultrasound wand, providing a live, real-time ultrasound image of the patient. An indicator 219 such as a red dot, the word “LIVE” or another graphical element is displayed at the bottom corner of the display screen to indicate that the image is live.

[0031] In FIG. 2B, an automatic segmentation has been applied to the ultrasound image to identify and distinguish different anatomical structures present in the image. The monitor 212 in FIG. 2B displays the segmentation by overlaying boundaries onto the ultrasound image, producing an annotated ultrasound image 222B. The annotated ultrasound image 222B may be a snapshot taken from an earlier live ultrasound image, or may be a live image in which the segmentation is continuously applied and updated. The segmentation identifies anatomical structures and landmarks in the image, such as the trachea, thyroid, vessels, and other structures.

[0032] The annotated image 222B includes one or more boundaries that indicate various tissue structures. For example in FIG. 2B, the boundaries include a dermal boundary 260 identifying a dermal tissue layer 261, a thyroid boundary 262 identifying a thyroid tissue 263, a tracheal boundary 264 identifying a trachea 265, and vessel boundaries 266 identifying blood vessels 267. The display screen also includes anAttorney Docket No. A0013140W001 indicator 256 such as “AUTO-SEGMENTATION” to indicate that the segmentation model is applied in the image.

[0033] The segmentation shown in FIG. 2B includes boundary lines drawn around the identified tissue structures and landmarks. These boundary lines can be displayed in color, dashed or dotted lines, or solid lines. These boundary lines are just one example of a graphical indicators of the segmentation on the display screen, and other types of graphical indicators can be used, such as cross-hatched or shaded areas in the ultrasound image, arrows pointing to certain areas, highlighting, flashing or pulsing areas, relative brightening or darkening of areas, text labels, and other indicators.

[0034] Additionally, a snapshot of the annotated image with the graphical indicators of the segmentation can be saved and stored in the patient’s medical record. For example, the image 222B in FIG. 2B with the boundaries 260, 262, 264, 266 can be saved as an image file and transferred into a local patient record or sent to an EMR system over a network (see FIG. IB). This information may be useful for later analysis of the patient’s intubation or tracheostomy procedure, their recovery process, or future procedures. The image can also be de-identified and shared with other systems or users for training.

[0035] In an embodiment, the segmentation is generated by a machine learning (ML) model. ML models may include decision trees, neural networks (NN), convolutional neural networks (CNN), transformers, support vector machines (SVM), and the like, that can be trained to recognize patterns or make predictions based on input data. Training a ML model involves utilizing a dataset (e.g., training data) to enable the ML model to learn relationships between input variables (e.g., input features) and output variables (e.g., labels, if available). For example, training data used to prepare a ML model can include labeled data (e.g., supervised learning), unlabeled data (e.g., unsupervised learning), or a combination of labeled and unlabeled data (e.g., semi-supervised learning). The performance of a trained ML model can be evaluated using a validation dataset and, in some cases, further training can be conducted. Once trained and validated, a ML model is able to make predictions and / or identify patterns in new data (e.g., data not included in the training dataset and / or the validation dataset).

[0036] According to an embodiment, a tissue segmentation ML model is a supervised model in which features are learned using training data which is labeled input data, in this case labeled ultrasound images. These ultrasound images are labeled with airway anatomyAttorney Docket No. A0013140W001 that is present in the ultrasound image, such as dermal tissue, thyroid tissue, trachea lumen, tracheal cartilage, blood vessels, and other areas of tissue (muscle, dermal layers, cartilage, vocal cords, glands, others). The output from the ML model is a segmented ultrasound image, with the distinct anatomical structures identified and measured. The model thus outputs anatomical structures and measurements (size, shape, relative locations, boundaries) as a function of the input ultrasound image. Example tracheal measurements that can be included in the ML output include tracheal diameter, anterior- posterior distance, lateral distance, estimation of the trachea centerline (location of axis A), tracheal depth (from the skin), and tracheal angle. In an embodiment, the tracheal diameter is measured about 2 cm below the second tracheal ring, or between the second and third rings.

[0037] One example ML model is a deep neural network (such as a convolutional NN). In this case, the ultrasound images are provided to the input layer of the NN. The NN propagates the input through a number of layers to the output layer. The output layer produces the segmentation.

[0038] In an embodiment, the ML model operates with one or more ultrasound images as the only input or type of input into the model. The ML model does not require other patient-specific information such as height, weight, age, or gender. This enables the system to be used more quickly in real-time at the bedside, without being delayed by manual entry or lookup of a patient’s demographic information. In addition to being streamlined, the system is also more robust, as these pieces of demographic information do not necessarily predict unique tracheal anatomy for individual patients. The input can be a single ultrasound image of the patient taken by the ultrasound probe, or several ultrasound images taken at a few different angles or positions (such as long and short axis ultrasound images), such as 2, 3, 4, or more images taken and accepted by the user.

[0039] In an embodiment, the output of the ML model is two-dimensional. The segmentation of structures and the measurements are all in a two-dimensional plane within the ultrasound image. This two-dimensional output requires less computational resources as compared to generating a 3 -dimensional model.

[0040] Additionally, in an embodiment, the output of the ML model stops short of actually recommending a tracheal tube that would be potentially a good fit for the patient. Rather than move to that endpoint, the output of the ML model is the tissue segmentationAttorney Docket No. A0013140W001(structures and measurements). This output is able to be labeled on the training data set with high confidence, and it results in a ML model that operates with high accuracy. Thus, the ML model is strategically deployed where it is most useful, and it is not extended to other endpoints (such as selection of a tube) that are better served by a rules-based algorithm or with user input. An output that identifies a tube with a good or appropriate fit would require training data with labeling that is more subjective and debatable in nature (identifying a “good” fit) as compared to the labeling here (tissue structure).

[0041] Turning to FIG. 3 A, a segmented ultrasound image 322Ais displayed along with a legend 370 and a prompt 372. The legend 370 labels the specific tissue structures in the segmented image, such as dermis and muscle (area 1), thyroid (area 2), vessels (area 3), and trachea (area 4). These labels and the associated boundaries and measurements are the output of the ML model described above.

[0042] Prior to selecting a tracheal tube for the patient, the system provides the user with a chance to accept or reject the auto-generated segmentation, at the prompt 372. At this point, a user such as a medical professional can assess the auto-generated segmentation for any apparent errors or problems, such as tissue structures that the user thinks are mis-identified (the thyroid being included in the trachea, for example), structures that are not identified, signal quality issues, distortions, or other issues. If the user does not accept the proposed segmentation, the user can continue to obtain new ultrasound images and continue to obtain new auto-generated segmentation results until the user is satisfied with the results. For example, the user can move the ultrasound wand, adjust the patient’s position, change ultrasound settings, or make other adjustments to obtain a cleaner or more useful ultrasound image. The auto-segmentation runs continuously, providing a proposed output in the annotated image 322A. When the user is satisfied with the result, the user touches the prompt 372 to accept.

[0043] When the system receives the user input in response to the prompt 372, the system moves into a next operation to list proposed tracheal tubes that may provide a good fit for the patient. This is shown in FIG. 3B, where the monitor 312 displays an image 322B with tracheal measurement outputs 374 and identified tracheal tubes 376. In the example of FIG. 3B, the measurements include a first tracheal depth A (distance from the skin surface to the front wall of the trachea), a second tracheal depth B (distance from the skin surface to the center of the tracheal lumen), and a tracheal inner diameter C.Attorney Docket No. A0013140W001

[0044] The system uses the measurements 374 as input into a rules-based algorithm to map the measurements to candidate tracheal tubes that are potentially compatible with the patient’s anatomy. The identified tubes 376 (such as specific sizes or types) are then displayed on the display screen. This list of tubes is generated based on the patient’s specific unique anatomy from the live ultrasound image, and not based on averages or trends across populations, and not based on prior or historical information from the patient on a prior visit. The list of identified tubes is produced based on custom information about the individual patient at the time of the procedure. The rules-based algorithm can be a formula, a lookup table, a decision tree, or another similar set of steps that maps the tracheal measurements to tracheal tubes.

[0045] For endotracheal tubes, the primary tube measurements utilized in this rules- based algorithm are the cuff diameter (inflated) and the outer diameter (OD) of the cannula of the tube. For a tracheostomy tube, the primary tube measurements are cuff diameter (inflated), tube OD, curve angle of the cannula, and cannula length (overall, proximal and distal).

[0046] In one example, a lookup table is used to identify tubes that have an inflated cuff diameter that is 110-150% the size of the tracheal inner diameter output from the ML model. In another example, the tracheal inner diameter is input into a formula that calculates a range of appropriate inflated cuff diameters, and then a table is consulted to identify tubes with balloon cuffs within that range.

[0047] In another example, the tracheal inner diameter is input into a formula to calculate a range of outer diameters of a potentially compatible tracheal tube cannula. In an example, a target cuff diameter is within a range of 110-150% of the tracheal diameter, to provide a cuff large enough to expand to the inner surfaces of the trachea and create a seal. Then a table is consulted to identify tubes with outer diameters within that range. A similar operation can be performed based on an appropriate inner diameter of the tracheal tube.

[0048] In another example, the skin-to-trachea distance (either distance A or B from image 422A) is used to identify tracheostomy tubes that have an appropriate cannula length between the proximal (forward) end of the tube to the curve (see FIG. 1 A). This is important to select a tube that will fit within the patient’s anatomy and not abut the forward or rear tracheal walls.Attorney Docket No. A0013140W001

[0049] A chart showing tracheal diameter ranges and potentially compatible tracheal tubes is shown in FIG. 9. The y-axis has several tubes with different diameters (including inflated cuff diameter), and the x-axis has tracheal diameter. The solid bars indicate ranges where the associated tube has an inflated cuff diameter that is 110-150% of the tracheal diameter. Dashed bars indicate ranges where the associated tube has an inflated cuff diameter that is 100-110% of the tracheal diameter, indicating a cuff that might fit but is close to being too small and causing leaks. Charts like this can be gathered from empirical data, test data, literature reviews, and other sources to produce a mapping of tracheal measurements to tubes that are potentially a good fit. Various charts, tables, or formulas like this can be created for different categories of tubes as well, such as cuffless endotracheal tubes, barrel-cuff endotracheal tubes, taper-cuff endotracheal tubes, cuffless tracheostomy tubes, etc. The system consults the appropriate chart to identify tubes to display to the user.

[0050] FIG. 4 illustrates an embodiment in which a live ultrasound image and a depiction of a tracheal tube are shown side-by-side. In this example, the monitor 412 includes a live ultrasound image 422A on the left, annotated with measurements A, B, and C that are output from the ML model.

[0051] An annotated ultrasound image 422B is shown on the right. The two images 422A, 422B are displayed simultaneously. The annotated image 422B can be a snapshot taken when the user accepts the segmentation (such as a snapshot of image 322Aupon receipt of the prompt in FIG. 3 A), or it can be a live image updated in realtime with the incoming signal from the ultrasound probe.

[0052] The annotated image 422B include a depiction 480 of a selected tracheal tube, positioned in the subject’s trachea, to indicate how the selected tube might fit. The depicted tube is selected by the user from the list of identified tubes 476. The user can input a selection of one of the tubes, such as by touching on the touch screen on the desired tube (or other user input). In FIG. 4, Tube 2 has been selected, as indicated by the check mark. The system then uses the size and shape of the selected tube to display the depiction 480.

[0053] In an embodiment, the depiction 480 is two-dimensional, like the ultrasound image itself. For example, the depiction 480 in FIG. 4 includes an outer circle 481 (showing the outer boundary of an inflated balloon cuff), and two inner circles 482, 483Attorney Docket No. A0013140W001(showing the inner and outer diameters of the cannula of the tracheal tube). The size of the circles 481, 482, 483 are based on the selected tube. The position of the circles within the image 422B is determined by centering the circles about the tracheal lumen of the subject, which can be determined by the segmentation output from the ML model above. The circles 481, 482, 483 are concentric, as the elements they represent on the tracheal tube (the cuff and the cannula) are also concentric about the axis of the tracheal tube.

[0054] The side-by-side view of FIG. 4 enables the user to visualize the size of the selected tracheal tube in the patient’s anatomy and assess whether it provides a good fit, prior to inserting the tube. For example, the inflated outer diameter of the cuff should be larger than the patient’s tracheal diameter, so that the cuff, when inflated, provides a good seal around the circumference of the trachea. If the diameter of the balloon cuff is smaller than the trachea, then gaps may exist between the outer surface of the inflated cuff and the tracheal wall, which can lead to leaks in ventilation of the patient.

[0055] By viewing the circles 482, 483, the user can also look for a tube that is as large as possible to maximize air flow through the inner diameter of the tube but no so large as to create force on the patient’s tracheal wall. In another example, the user may look for a smaller outer diameter to allow airflow around the tube when the cuff is deflated, which is important for weaning off a tracheostomy tube. The views in FIG. 4 help the user make these assessments prior to selecting and inserting a tube.

[0056] Another embodiment with different views and user interactions is shown in FIGs. 5A and 5B. In these example, a monitor 512 displays two ultrasound images simultaneously. In FIG. 5A, the top image 584 is a long-axis ultrasound image, which is obtained by orienting the length of the ultrasound wand vertically along the patient’s neck, such that the plane of the ultrasound waves is parallel with the axis of the tracheal lumen. This long axis image provides a view of the tracheal wall and a length of the tracheal lumen, but does not indicate the tracheal diameter. In image 584, the ML model has produced a segmentation identifying distances A and B and angle 9, the angle between the distance measurements and the axis of the trachea. These measurements are also shown numerically on the right side of the screen. Once the user accepts the segmentation of the long-axis image 584, the user enters a confirmation (such as via the “Acceptance” button), and then the user rotates the ultrasound wand 90 degrees to obtain a short-axis ultrasound view of the patient’s trachea.Attorney Docket No. A0013140W001

[0057] A short-axis ultrasound image 586 is shown in the bottom of the display screen in FIG. 5A. This image is obtained by orienting the ultrasound wand across the patient’s neck horizontally, so that the plane of the ultrasound waves are perpendicular to the axis of the trachea. This view provides an image including the diameter of the trachea. In the image 586, the ML model is providing a segmentation including length measurements A and B, and tracheal diameter C. These measurements are also displayed numerically on the display screen, and a prompt 572 is presented for the user to accept the measurements when ready.

[0058] Once the user has accepted both the long and short axis measurements (or just one of these, in other embodiments), the system identifies potential tubes and displays them as shown in FIG. 5B. In this view, the user can select a tube and view a depiction of the tube placed in the trachea, based on a live or snapshot ultrasound image. In FIG. 5B, two images are displayed simultaneously so that the user can compare the size and fit of two tubes. In the top image, Tube 1 is depicted in the upper image 586A (including circles 581, 582, 583 indicating the cuff diameter, cannula outer diameter, and cannula inner diameter respectively of Tube 1), and Tube 2 is depicted in the lower image 586 B (also including respective circles indicating the dimensions of Tube 2). The user can compare these two images to determine whether Tube 1 or Tube 2 provides a better fit, and then select the appropriate tube for placement. Other views can show additional tubes or additional comparisons of depicted tube sizes and geometries.

[0059] FIGs. 6A and 6B show an animation of insertion of a tracheostomy tube into a patient. In this embodiment, the display screen (such as any of display screens 116, 216, and others shown herein) displays this animation after the user selects a candidate tracheostomy tube, such as selecting Tube 1, Tube 2, or Tube 3 in FIG. 5B. Upon this selection (or another user input), the monitor displays the animation in FIGs. 6A-6B demonstrating placement of the tracheostomy tube. The size, position, and angle of the trachea in displayed in FIGs. 6A-6B is based on earlier measurements of the patient. The animation can be a two- or three-dimensional rendering of the trachea and the tube, or it can be a rendering of the tube displayed over an ultrasound image of the patient. In either case, the movement of the tube (as shown by the arrows and the sequence of images in FIGs. 6A-6B) demonstrates how it might fit inside the patient’s unique tracheal anatomy, as previously measured by the ultrasound guided systems and methods described herein.Attorney Docket No. A0013140W001The user can view the animation prior to deciding which tube to use, and prior to performing the procedure.

[0060] A method for training a ML model to output a tissue segmentation as a function of an ultrasound image is illustrated in FIG. 7, according to an embodiment. The method 700 includes collecting training data at 701. This includes a database of ultrasound images of airway anatomy with tissue structures labeled and measured in the images. For example, ultrasound images can be stored and a medical professional can label these images to indicate the location, size, and shape of the tissue structures (skin, thyroid, trachea, vessels) in the images. The method also includes training an ML model at 702. Training includes supervised learning, in which the ML learns relationships between input variables (features in the ultrasound images) and output variables (tissue segmentation). The method also includes validating the ML model at 703. Validation can be done by inputting a separate validation data set (ultrasound images that are not labeled) into the ML model, obtaining outputs from the ML model, and comparing the outputs to expected outputs based on the validation data set. In an embodiment, the method also includes updating or refining the ML model at 704, but re-training with new or additional training data sets to improve or update the accuracy and performance of the ML model.

[0061] In an embodiment, segmented images that are produced by the system in operation and accepted by the user (such as image 322A) are de-identified (removing all patient identifying data) and passed back to a server to use as additional training data to update the ML model. The labeling in these images has been verified by a human user and can be considered new truth data, and thus they can be used as re-training data at 704 in FIG. 7.

[0062] A method 800 for providing tracheal tube guidance with ultrasound input is illustrated in FIG. 8, according to an embodiment. The method includes receiving ultrasound images of a patient’s airway at 801, such as images from an ultrasound probe, and displaying an ultrasound image on a display screen at 802. The method includes providing the ultrasound image as input into a trained model at 803, and receiving, as output from the trained model, a segmentation of tissue structure in the ultrasound image at 804. The method includes displaying the segmentation on the display screen at 805, such as by overlaying one or more boundaries onto the displayed ultrasound image, and receiving a user confirmation of the segmentation at 806. The method includes identifyingAttorney Docket No. A0013140W001 first and second tracheal tubes at 807 (such as via a lookup table or formula), and displaying the identified tubes at 808. Optionally the method of FIG. 8 also includes receiving a selection of an identified tube at 809 and displaying a depiction of the selected tube at 810, such as by overlaying a geometric outline of the tube onto the displayed ultrasound image at 810.

[0063] The automatic segmentation of tissue structure in the airway image is performed continuously in realtime at the bedside, enabling the user to continue to adjust the ultrasound probe until a good position and good image are obtained. This is indicated by the return arrow back to 801 in FIG. 8. The user can view the real-time outputs of the ML model and determine when the outputs are ready to accept, such as moving the probe to obtain a good image with clear segmentation, or to get a better view of specific tissue structure. The user can also accept several images at different angles or positions, to get a complete view of the anatomy. Additionally, the segmentation can be added to a patient’s medical record, such as by storing it in an EMR so that it can be referenced in the future.

[0064] In an embodiment, the ML model can also identify a suggested target location for a percutaneous entry point for inserting a tracheostomy tube. In this case, the percutaneous entry point is another output of the ML model, based on training data that includes ultrasound images with appropriate entry points labeled on the images. An appropriate entry point is one that avoids incision of blood vessels and the thyroid gland. The segmentation output by the ML model includes identifying a location of an appropriate entry point, as an output of the model. This entry point can be displayed on the display screen along with other elements of the segmentation.

[0065] In an embodiment, further ultrasound inputs can be obtained after the tube has been placed into the patient’s trachea, such as an intubation with an endotracheal tube or a tracheostomy procedure for a tracheostomy tube. After placement, the inserted tube can be re-examined with the ultrasound probe and the resulting image can be compared to the modeled image (such as image 422B, 586 A, or 586B) for any variations and to assess cuff sealing. Doppler ultrasound may facilitate measurement of blood flow at the site to identify any bleeding or injury. The ultrasound image can also be examined to identify fluid collection above the cuff for suctioning, or any occlusions in the tube. For a tracheostomy tube with fenestrations, the ultrasound image can also reveal if the fenestrations are contacting tracheal tissue, which can cause irritation. These outputs canAttorney Docket No. A0013140W001 be provided by a second ML model that is trained on annotated images of inserted tubes (identifying an inserted tube within an ultrasound image) and outcome data including the performance of these inserted tubes, the presence of gaps or leaks, and the size and fit of the cuff.

[0066] The following numbered examples demonstrate one or more aspects of the disclosure.

[0067] Example 1. A method for tracheal tube guidance with ultrasound input, comprising receiving, from a probe, an ultrasound image of a subject’s trachea; displaying the ultrasound image on a display screen; providing the ultrasound image as input into a trained model; receiving, as output from the trained model, a segmentation of tissue structure in the ultrasound image; displaying the segmentation on the display screen by overlaying one or more boundaries onto the displayed ultrasound image; prompting a user confirmation of the displayed segmentation; upon receiving the user confirmation, identifying first and second candidate tracheal tubes; and displaying, on the display screen, the first and second candidate tracheal tubes.

[0068] Example 2. The method of Example 1, wherein the segmentation comprises a tracheal measurement output from the trained model, and wherein identifying the first and second tracheal tubes comprises selecting the first and second tracheal tubes based on the tracheal measurement.

[0069] Example 3. The method of Example 2, wherein the tracheal measurement comprises a tracheal diameter, a dermal to trachea depth, a tracheal angle, or a combination of these.

[0070] Example 4. The method of Example 2, wherein selecting the first and second tracheal tubes based on the tracheal measurement comprises utilizing a formula or a lookup table mapping the tracheal measurement to a dimension of the tracheal tubes.

[0071] Example 5. The method of Example 3, further comprising displaying the tracheal measurement on the display screen.

[0072] Example 6. The method of any preceding Example, further comprising receiving a user selection of one of the first or second tracheal tubes, and displaying a depiction of the selected tracheal tube in the segmented tissue structure.Attorney Docket No. A0013140W001

[0073] Example 7. The method of Example 6, wherein the depiction comprises an annotated ultrasound image comprising an outline of the selected tracheal tube centered on the subject’s trachea.

[0074] Example 8. The method of Example 7, wherein the outline comprises an inflated cuff diameter of the tracheal tube.

[0075] Example 9. The method of Example 6, further comprising receiving a subsequent user selection of the other of the first or second tracheal tubes, and displaying a second depiction of such other selected tracheal tube.

[0076] Example 10. The method of Example 1-9, wherein the first tracheal tube comprises a first endotracheal tube having a first size or type, and the second tracheal tube comprises a second endotracheal tube having a second size or type different from the first.

[0077] Example 11. The method of Example 1-9, wherein the first tracheal tube comprises a first tracheostomy tube having a first size or type, and the second tracheal tube comprises a second tracheostomy tube having a second size or type different from the first.

[0078] Example 12. The method of Example 11, wherein the segmentation comprises a blood vessel, and wherein the method further comprises identifying a percutaneous entry point outside of the blood vessel and labeling the percutaneous entry point on the display screen.

[0079] Example 13. The method of any preceding Example, wherein the trained model comprises a supervised machine learning model trained on labeled ultrasound images.

[0080] Example 14. The method of any preceding Example, wherein the one or more boundaries include one or more of: a first boundary indicating dermal tissue, a second boundary indicating thyroid tissue, a third boundary indicating a trachea, and a fourth boundary indicating a blood vessel.

[0081] Example 15. A method for tracheal tube guidance with ultrasound input, comprising: receiving, from a probe, an ultrasound image of a subject’s trachea; displaying the ultrasound image on a display screen; providing the ultrasound image as input into a trained model; receiving, as outputs from the trained model, a segmentation of tissue structure in the ultrasound image and an identification of a tracheal tube in the ultrasound image; displaying an annotated ultrasound image comprising an indicator of the identifiedAttorney Docket No. A0013140W001 tracheal tube; retrieving a stored image of the subject’s trachea; and displaying the stored image and the annotated ultrasound image side by side on the display screen.

[0082] Example 16. The method of Example 15, wherein the stored image comprises a prior ultrasound image of the subject’s trachea taken at an earlier time.

[0083] Example 17. The method of Example 15 or 16, wherein the annotated ultrasound image further comprises one or more of: a first boundary indicating dermal tissue, a second boundary indicating thyroid tissue, a third boundary indicating a trachea, and a fourth boundary indicating a blood vessel.

[0084] Example 18. An ultrasound guidance system comprising: a probe generating an ultrasound signal; a display screen coupled to the probe and displaying an ultrasound image from the ultrasound signal; a processor communicating with the display screen and programmed to perform a set of instructions including: providing the ultrasound image as an input to a trained machine learning model; receiving, as an output from the trained machine learning model, a segmentation of tissue structure in the ultrasound image; displaying the segmentation on the display screen by overlaying one or more boundaries onto the displayed ultrasound image; prompting a user confirmation of the displayed segmentation; and upon receiving the user confirmation, identifying first and second candidate tracheal tubes and displaying the identified first and second candidate tracheal tubes on the display screen.

[0085] Example 19. The system of Example 18, wherein the display screen and the processor are housed in a portable tablet computer.

[0086] Example 20. The system of Example 18, wherein the display screen and the processor are housed in a smartphone.

[0087] In one or more examples, the described techniques may be implemented in hardware, software, firmware, or any combination thereof. If implemented in software, the functions may be stored as one or more instructions or code on a computer-readable medium and executed by a hardware-based processing unit. Computer-readable media may include non-transitory computer-readable media, which corresponds to a tangible medium such as data storage media (e.g., RAM, ROM, EEPROM, flash memory, or any other medium that can be used to store desired program code in the form of instructions or data structures and that can be accessed by a computer).Attorney Docket No. A0013140W001

[0088] Instructions may be executed by one or more processors, such as one or more digital signal processors (DSPs), general purpose microprocessors, application specific integrated circuits (ASICs), field programmable logic arrays (FPGAs), or other equivalent integrated or discrete logic circuitry. Accordingly, the term “processor” as used herein may refer to any of the foregoing structure or any other physical structure suitable for implementation of the described techniques. Also, the techniques could be fully implemented in one or more circuits or logic elements.

[0089] It should be understood that various aspects disclosed herein may be combined in different combinations than the combinations specifically presented in the description and accompanying drawings. It should also be understood that, depending on the example, certain acts or events of any of the processes or methods described herein may be performed in a different sequence, may be added, merged, or left out altogether (e.g., all described acts or events may not be necessary to carry out the techniques). In addition, while certain aspects of this disclosure are described as being performed by a single module or unit for purposes of clarity, it should be understood that the techniques of this disclosure may be performed by a combination of units or modules associated with, for example, a medical device.

Claims

Attorney Docket No. A0013140W001WHAT IS CLAIMED IS:

1. A method for tracheal tube guidance with ultrasound input, comprising: receiving, from a probe, an ultrasound image of a subject’s trachea; displaying the ultrasound image on a display screen; providing the ultrasound image as input into a trained model; receiving, as output from the trained model, a segmentation of tissue structure in the ultrasound image; displaying the segmentation on the display screen by overlaying one or more boundaries onto the displayed ultrasound image; prompting a user confirmation of the displayed segmentation; upon receiving the user confirmation, identifying first and second candidate tracheal tubes; and displaying, on the display screen, the first and second candidate tracheal tubes.

2. The method of claim 1, wherein the segmentation comprises a tracheal measurement output from the trained model, and wherein identifying the first and second tracheal tubes comprises selecting the first and second tracheal tubes based on the tracheal measurement, and wherein the method further comprises displaying the tracheal measurement on the display screen.

3. The method of claim 2, wherein the tracheal measurement comprises a tracheal diameter, a dermal to trachea depth, a tracheal angle, or a combination of these.

4. The method of claim 2, wherein selecting the first and second tracheal tubes based on the tracheal measurement comprises utilizing a formula or a lookup table mapping the tracheal measurement to a dimension of the tracheal tubes.Attorney Docket No. A0013140W0015. The method of any preceding claim, further comprising receiving a user selection of one of the first or second tracheal tubes, and displaying a depiction of the selected tracheal tube in the segmented tissue structure, wherein the depiction comprises an annotated ultrasound image comprising an outline of the selected tracheal tube centered on the subject’s trachea.

6. The method of Claim 5, further comprising receiving a subsequent user selection of the other of the first or second tracheal tubes, and displaying a second depiction of such other selected tracheal tube.

7. The method of claim 1-6, wherein the first tracheal tube comprises a first endotracheal tube having a first size or type, and the second tracheal tube comprises a second endotracheal tube having a second size or type different from the first.

8. The method of claim 1-6, wherein the first tracheal tube comprises a first tracheostomy tube having a first size or type, and the second tracheal tube comprises a second tracheostomy tube having a second size or type different from the first.

9. The method of claim 8, wherein the segmentation comprises a blood vessel, and wherein the method further comprises identifying a percutaneous entry point outside of the blood vessel and labeling the percutaneous entry point on the display screen.

10. The method of any preceding claim, wherein the trained model comprises a supervised machine learning model trained on labeled ultrasound images.

11. The method of any preceding claim, wherein the one or more boundaries include one or more of: a first boundary indicating dermal tissue, a second boundary indicating thyroid tissue, a third boundary indicating a trachea, and a fourth boundary indicating a blood vessel.Attorney Docket No. A0013140W00112. The method of any preceding claim, wherein the output further comprises an identification of a tracheal tube in the ultrasound image, and the method further comprises displaying an indicator of the identified tracheal tube.

13. An ultrasound guidance system comprising: a probe generating an ultrasound signal; a display screen coupled to the probe and displaying an ultrasound image from the ultrasound signal; a processor communicating with the display screen and programmed to perform a set of instructions including: providing the ultrasound image as an input to a trained machine learning model; receiving, as an output from the trained machine learning model, a segmentation of tissue structure in the ultrasound image; displaying the segmentation on the display screen by overlaying one or more boundaries onto the displayed ultrasound image; prompting a user confirmation of the displayed segmentation; and upon receiving the user confirmation, identifying first and second candidate tracheal tubes and displaying the identified first and second candidate tracheal tubes on the display screen.

14. The system of claim 13, wherein the display screen and the processor are housed in a portable tablet computer.

15. The system of claim 13, wherein the display screen and the processor are housed in a smartphone.

Images