Automated estimation of pelvic measurements from medical images using machine learning

Abstract

A computer-implemented method and system automatically infers pelvimetry measurements from volumetric medical images. A processing system receives a volumetric medical image of a pelvic region of a patient and applies machine learning model trained to infer a set of volumetric probability maps for a plurality of respective anatomical landmarks of the pelvic region. Each of the volumetric probability maps represents a per-pixel array of probabilities for a location of an anatomical landmark. The processing system selects respective locations for the respective anatomical landmarks based on the set of volumetric probability maps. The processing computes a set of pelvimetry measurements characterizing the respective locations for the respective anatomical landmarks and outputs the set of measurements. The measurements may be outputted as visual overlay on the captured images for review by a medical practitioner, and / or may be provided as control inputs to medical instrumentation utilized in pelvic-related medical procedures.

Description

UTOMATED ESTIMATION OF PELVIC MEASUREMENTS FROM MEDICAL IMAGES USING MACHINE LEARNINGCROSS-REFERENCE TO RELATED APPLICATIONS

[0001] This application claims the benefit of U.S. Provisional Patent Application No. 63 / 561,548 filed on March 5, 2024, which is incorporated by reference herein.BACKGROUND

[0002] Measurements of pelvic dimensions can be predictive of the degree of difficulty and outcome of medical procedures such as total mesorectal excision (TME) for treatment of rectal cancer. Conventionally, pelvic measurements are manually measured by a medical provider, which is time consuming and prone to human error.DETAILED DESCRIPTION

[0003] The Figures (FIGS.) and the following description describe certain embodiments by way of illustration only. One skilled in the art will readily recognize from the following description that alternative embodiments of the structures and methods illustrated herein may be employed without departing from the principles described herein. Reference will now be made to several embodiments, examples of which are illustrated in the accompanying figures. Wherever practicable, similar or like reference numbers may be used in the figures and may indicate similar or like functionality.

[0004] A computer-implemented method and system automatically infers pelvimetry measurements from volumetric medical images. A processing system receives a volumetric medical image of a pelvic region of a patient and applies machine learning model trained to infer a set of volumetric probability maps for a plurality of respective anatomical landmarks of the pelvic region. Each of the volumetric probability maps represents a per-pixel array of probabilities for a location of an anatomical landmark. The processing system selects respective locations for the respective anatomical landmarks based on the set of volumetric probability maps. The processing system computes a set of pelvimetry measurements characterizing the respective locations for the respective anatomical landmarks and outputs the set of measurements. The measurements may be outputted as visual overlay on the captured images for review by a medical practitioner, and / or may be provided as control inputs to medical instrumentation utilized in pelvic-related medical procedures.

[0005] The estimated pelvic measurements may be applied in various medical contexts. For example, having accurate pelvic measurements is beneficial in procedures such as a TME for treatment of rectal cancer. This procedure involves the precise dissection of the mesorectum to achieve optimal oncological outcomes. There is a large variation in the difficulty of performing this procedure. The surgical difficulty depends on both patient-related clinical and anatomical factors. For example, BMI, sex, tumor height and size, and dimensions of the pelvic cavity can be predictive factors of surgical difficulty in patients undergoing rectal surgery. The anatomical constraints of the bony pelvis directly impact surgical access to the rectum and the ability to achieve precise mesorectal dissection. In particular, TME is challenging in patients with a narrow and deep pelvis because the bony structure surrounding the rectum disturbs the surgical maneuver. Accurate pelvimetry can therefore provide significant predictive value for TME outcomes.

[0006] Pelvimetry derived using the techniques herein may also be implemented for other use cases. For example, the estimated pelvic measurement may predict early postoperative morbidity and oncological outcomes. Automation of these measurements enables pre-operative assessment of the expected difficulty of TME or other procedures, and ultimately improves surgical outcomes.

[0007] FIG. 1 illustrates an example embodiment of a computing environment 100 for automatically detecting pelvic measurements based on medical images. The computing environment 100 includes one or more medical imaging device 102, a processing system 104, one or more client devices 106, an image store 110, and medical instrumentation 112 coupled by a network 108. In alternative embodiments, two or more of the imaging device 102, processing system 104, client devices 106, image store 110, and / or medical instrumentation 112 may be directly coupled to each other and are not necessarily all directly coupled to the network 108. Furthermore, in some embodiments, one or more components may be isolated from the network 108 and instead allow for data transfer via physical storage devices that can be manually transported. For example, images from the medical imaging device 102 could be transported and loaded to the image store 110 (or a client device 106) via a physical storage medium such as a disk drive, solid state drive, or other storage medium. In some embodiments, two more components (such as the processing system 104 and image store 110) may be integrated into a single device.

[0008] The medical imaging device 102 captures medical images. In an embodiment, the medical imaging device 102 comprises a magnetic resonance imaging (MRI) device that captures volumetric images comprising a set of image slices. In other embodiments, the medical imagingdevice 102 may comprise a computed tomography (CT) imaging device, an X-ray device, an ultrasound imaging device, or other imaging device suitable for capturing volumetric images of the pelvic anatomy. While FIG. 1 shows only a single medical imaging device 102, the computing environment 100 may include multiple imaging devices 102 that may be operated by different institutions and may be physically distributed in different medical or research facilities.

[0009] At least some subset of volumetric images from the medical imaging device(s) 102 may be stored to the image store 110. The image store 110 can furthermore store historic volumetric medical images that may have originated from medical imaging devices 102 that are no longer in operation. The image store 110 may store the images themselves and may store various metadata associated with the images such as information about the patient, type of machine used to capture the images, time of capture, etc. The metadata may furthermore include annotations (labels) associated with the images as described further below. The image store 110 may include any local storage (e.g., one or more on-site storage devices at one or more medical facilities) that may be co-located with one or more other components of the computing environment 100, cloud storage that may include remote and / or distributed network-based storage, or a combination thereof.

[0010] The processing system 104 performs processing of medical images from the image store 110 and / or images that may be received directly from the medical imaging device 102 or client device 106. The processing system 104 may utilize various machine learning techniques to infer pelvic measurements from the medical images as will be further described below. The processing system 104 may be implemented using on-site computing or storage systems, cloud computing or storage systems, or a combination thereof and may be implemented utilizing local or cloud-based servers, which may include physical or virtual machines, or a combination thereof. Cloud-based servers may include private cloud systems, public cloud systems, hybrid public / private cloud systems, or a combination thereof. Accordingly, the processing system 104 may be local, remote, and / or distributed relative to the medical environments where the medical images are obtained and where the client device 106 operates. Furthermore, different portions of the processing system 104 may execute on different remote servers and various system elements of the processing system 104 may be communicatively coupled over a network 108.

[0011] The client device 106 may include any computing devices for inputting information and / or control commands, viewing various information, and / or interacting with various data. The client device 106 may execute one or more applications that include a user interface for performing various functions described herein. The application may comprise a web-based application accessible by a web browser or a locally installed application. The application mayallow the client device 106 to facilitate functions such as controlling the medical imaging device 102, viewing, adding, removing, editing, or otherwise interacting with images in the image store 110, initiating processes performed by the processing system 104, inputting information to the processing system 104, viewing outputs from the processing system 104, etc.

[0012] The client device 106 may comprise, for example, a mobile phone, a tablet, a laptop or desktop computer, or other computing device. The client device 106 may include conventional computer hardware such as a display, input device (e.g., touch screen), memory, a processor, and a non-transitory computer-readable storage medium that stores instructions for execution by the processor in order to carry out functions described herein.

[0013] In an embodiment, aspects of the processing system 104 that perform inferences of pelvimetry measurements may be integrated with a medical imaging device 102 and / or the client device 106. For example, in one implementation, a trained model may be edge-deployed to execute locally on a medical imaging device 102 that captures the volumetric images, such as an MRI machine, a CT scanning machine, or other imaging system. In this example, the pelvimetry measurements may be automatically inferred upon capturing the volumetric images and may be outputted directly as visual overlays on the image slices or as metadata associated with the images output from the medical imaging device 102.

[0014] The medical instrumentation 112 may comprise a robotically-assisted surgery platform or other medical instrument that may be utilized in a medical procedure. For example, a robotically-assisted surgery platform may include one or more robot arms that may be equipped with specialized surgical instruments, high-definition cameras for visualization, and a control console where the surgeon operates. This platform allows for minimally invasive procedures with increased dexterity, stability, and 3D visualization. The robotic system can translate the surgeon's hand movements into precise micro-movements of the instruments inside the patient's body. Advanced versions may incorporate real-time imaging data and pre-operative plans to guide instrument positioning and surgical navigation. In some embodiments, the pelvimetry measurements inferred by the processing system 104 may be applied to the robotic system as a control parameter during performance of a particular procedure. In other embodiments, the medical instrumentation 112 could include non-robotic medical instruments such as an imaging system, radiation system, medicament delivery system, patient monitoring system, or other medical instrument that may utilize pelvimetry measurements as an operational control input.

[0015] The network 108 comprises communication pathways for communication between the medical imaging device 102, the processing system 104, and the client device 106. The network 108 may include one or more local area networks and / or one or more wide area networks(including the Internet). The network 108 may also include one or more direct wired or wireless connections (e.g., Ethernet, WiFi, cellular protocols, WiFi direct, Bluetooth, Universal Serial Bus (USB), or other communication link).

[0016] The above-described system may be utilized to automate estimation of pelvic measurements from volumetric medical images. In a training process, the processing system 104 learns one or more machine learning models based on a training set of annotated pelvic images from the image store 110. The annotated pelvic images may include labels indicating three- dimensional coordinates of the locations of anatomical landmarks of the pelvic anatomy. The processing system 104 applies a supervised training algorithm to train the model to identify these landmark coordinates in the images. In an inference process, the processing system 104 applies the trained model to an input MRI image (without annotations) and infers the locations of the landmarks. The processing system 104 may furthermore compute and output distances between the landmarks using the inferred coordinates, as will be described in further detail below.

[0017] The estimated pelvic measurements may be applied in various medical contexts. For example, having accurate pelvic measurements is beneficial in procedures such as a TME for treatment of rectal cancer. This procedure involves the precise dissection of the mesorectum to achieve optimal oncological outcomes. There is a large variation in the difficulty of performing this procedure. The surgical difficulty depends on both patient-related clinical and anatomical factors. For example, BMI, sex, tumor height and size, and dimensions of the pelvic cavity can be predictive factors of surgical difficulty in patients undergoing rectal surgery. The anatomical constraints of the bony pelvis directly impact surgical access to the rectum and the ability to achieve precise mesorectal dissection. In particular, TME is challenging in patients with a narrow and deep pelvis because the bony structure surrounding the rectum disturbs the surgical maneuver. Accurate pelvimetry can therefore provide significant predictive value for TME outcomes.

[0018] Pelvimetry derived using the techniques herein may also be implemented for other use cases. For example, the estimated pelvic measurement may predict early postoperative morbidity and oncological outcomes. Automation of these measurements enables pre-operative assessment of the expected difficulty of TME or other procedures, and ultimately improves surgical outcomes.

[0019] FIG. 2 illustrates an example embodiment of a training process for training a machine learning model capable of estimating anatomical landmarks from volumetric images. The processing system 104 obtains 202 training volumetric images annotated with locations (e.g., three-dimensional coordinates) of a set of anatomical landmarks in the volumetric images. Theimages may be preprocessed to normalize the images. For example, the intensity of pixels in the training volumetric images may be normalized using a percentile-based normalization. The size of the images may furthermore be normalized to the mean size of the whole dataset. Annotations may be obtained manually by one or more trained physicians or other experts.

[0020] Examples of annotations in an image slice for a volumetric image are shown in FIG. 4. In this example, the labeled landmarks may comprise the promotorium (A) 402, S3-vertebrae (B) 404, coccyx (C) 406, dorsal part of os pubis (D) 408, and cranial part of os pubis (E) 410. The landmarks provide the basis for measuring various relevant pelvic dimensions such as the pelvic inlet (W) 412 (representing a distance from the promotorium (A) 402 to the cranial part of the os pubis (E) 410), pelvic outlet (X) 414 (representing a distance from the coccyx (C) 406 to the dorsal part of the os pubis (D) 408), pelvic depth (Y) 416 (representing a distance from the promotorium (A) 402 to the coccyx (C) 406), and the sacral angulation (Z) 418 (representing an angle between a first line from the promotorium (A) 402 to the S3-vertebrae (B) 404 and a second line from the S3-vertebrae (B) 404 to the coccyx (C) 406). These dimensions may represent predictors for outcomes of TME or other medical procedures. For purposes of annotating the landmarks, the midline on the sagittal MRI may be used as the point of reference for annotation, which may be identified visually by the most comprehensive display of the sacrum.

[0021] In the example of FIG. 4, the annotations are shown with labeled coordinates of the landmarks and derived dimensions (e.g., distances and angles) projected into a single image slice. However, the labeled coordinates and dimensions may in practice be located in three- dimensional space. Thus in one embodiment, the annotator may be presented with a set of image slices making up the volume and the annotator may label the landmarks in the appropriate image slice. In another embodiment, a computer system may render a three-dimensional model that may be rotated or otherwise manipulated to enable an annotator to mark a three-dimensional coordinate.

[0022] Returning to FIG. 2, for each annotated volumetric image in the training set, a set of multiple probability maps are generated 204, one for each of the anatomical landmarks. A probability map associated with a landmark may be derived by applying a filter (e.g., Gaussian filter) to a three-dimensional matrix encoding the annotated location of the landmark. In one embodiment ,the matrix representation may comprise a matrix of the same size as the volumetric image in which the three-dimensional coordinates of the landmark is encoded as a 1 (or other predefined value) and the remaining values are encoded as 0s (or other predefined value). The Gaussian filter is applied to the this matrix representation to generate the probability map. Theprobability map has the same size as the original volumetric matrix and the original matrix representation, and has values between 0 and 1 for each position in the probability map. In this representation, the landmark location is represented as a Gaussian distribution around the labeled location identified by the annotator.

[0023] In an embodiment, a separate probability map is generated for each landmark. For example, in the case of five anatomical landmarks labeled in each volumetric training image (e.g., landmarks (A)-(E) 402-410 in FIG. 4), a set of five such probability maps are generated from each volumetric training image in the training set.

[0024] A learning algorithm is then applied 206 to learn a mapping between the original volumetric images in the training set and their corresponding sets of probability maps. The model may be trained end-to-end on the training dataset using a 3D U-Net training algorithm in one embodiment. In an embodiment, the training process may apply mean squared error (MSE)- loss as a loss function. The model may be trained over a sufficient number of epochs (e.g., 500 or more epochs), with an early stopping if the validation loss does not improve (e.g., no improvement for 50 epochs). The epoch with the lowest validation loss may be chosen for model selection. The model is outputted 208 and may be stored for application in the inference process described below.

[0025] In alternative embodiments, the learning process may employ different types of learning algorithms and corresponding models such as other types of convolutional neural networks, other neural networks, or other supervised learning techniques.

[0026] FIG. 3 illustrates an example embodiment of an inference process for automatically generating pelvic measurements from a volumetric image. An input volumetric image is obtained 302. The learned model is applied 304 to the input volumetric image to generate a set of probability maps (one per anatomical landmark). Each probability map indicates the predicted probability distribution of the corresponding landmark location on a per pixel basis. The landmark locations are then determined 306 from the probability maps. For example, in an embodiment, the landmark location may be determined from each probability map by selecting the location of the maximum probability value. Alternatively, an averaging function or other function may be applied to determine a single landmark location per probability map.

[0027] Pelvic dimensions may then be computed 308 based on distances between pairs of landmarks, angles formed by three or more landmarks, or other measurements. An output representation may be generated 310 indicative of the estimated dimensions and / or the landmark locations.

[0028] In an example embodiment, the landmark locations and / or dimensions may be overlaidon one or more slices of the original input image to provide a visual representation of the estimates that may be presented via a client device 106. The visual representation may be similar to the labeled image shown in FIG. 4 and described above. In FIG. 4 a single image slice is shown with the three-dimensional landmark locations projected into the plane of the image slice. In other embodiments, the visual representation may comprise a rendering of a three- dimensional model of the pelvic anatomy with labels for the inferred target locations and dimensions. In other embodiments, a text-based representation may provide the inferred coordinate for landmarks and / or the derived pelvic dimensions without necessarily showing them on an image overlay.

[0029] In other embodiments, the output representation may be formatted and transmitted for input to a medical instrumentation device 112 as described above. For example, the landmark locations and / or inferred pelvimetry measurements may be direct input as control signals to a robotically-assisted surgical platform or other medical instruments. In such embodiments, the robotically-assisted surgical platform may automatically infer the pelvimetry measurements (and subsequently the resulting robotic control parameters) from MRI or other preprocedural images accessible by the platform.

[0030] In the context of robotically-assisted surgery platform, the landmark locations and / or inferred pelvimetry measurements may be utilized by the platform for purposes of calibrating a robot arm, control arm positioning, preplanning a surgical path, selection or calibration of a surgical instrument controlled by the robotically-assisted surgery platform, controlling approach angles, or other control aspects, thereby enabling precise and safe navigation within a confined pelvic space.

[0031] In image-guided surgical systems, the landmark coordinates and derived measurements can enhance the accuracy of registration between pre-operative imaging and intra-operative navigation. Furthermore, the pelvimetry data may be integrated as overlays into augmented reality surgical displays, providing surgeons with real-time, patient-specific anatomical overlays to guide their actions during procedures. By incorporating these automated pelvimetry measurements, robotically-assisted surgical platforms and other advanced medical instruments can potentially achieve higher levels of precision, safety, and personalization in pelvic surgeries.

[0032] The inferred pelvimetry landmarks and measurements may also serve as control inputs for other non-robotic medical systems and devices. For instance, these measurements can be used to automatically adjust the positioning and configuration of radiation therapy equipment, ensuring optimal beam targeting while minimizing exposure to surrounding healthy tissue during pelvic cancer treatments. For diagnostic imaging, these measurements can be used toautomatically optimize MRI or CT scan parameters, such as field of view, slice thickness, and patient positioning, to capture the most relevant anatomical details while minimizing scan time and radiation exposure. In the context of virtual reality surgical simulators, the pelvimetry data can be used to generate patient-specific anatomical models, allowing for more realistic and personalized training scenarios. Additionally, the measurements may be integrated into computer-aided design systems for creating custom pelvic implants or prosthetics, with the system automatically adjusting implant dimensions and contours to match the patient's anatomy.

[0033] The disclosed system and method provide significant technological improvements in the field of pelvimetry, medical imaging analysis robot-assisted surgical platforms, and various medical instrumentation technologies. By leveraging machine learning techniques to automatically identify and localize key anatomical landmarks from volumetric medical images, the invention enables rapid and accurate estimation of critical pelvic measurements. This automated approach represents a substantial advancement over conventional manual measurement techniques, which are time-consuming, prone to human error, and subject to interobserver variability. The improved accuracy and consistency of pelvic measurements obtained through this technological solution directly translates to enhanced surgical planning and risk assessment for procedures such as total mesorectal excision (TME). By providing surgeons with more reliable pre-operative pelvimetry data, the disclosed embodiments can lead to improved patient outcomes, reduced surgical complications, and more efficient utilization of healthcare resources. Furthermore, the automated nature of the system allows for standardized measurements across different medical institutions, potentially enabling larger-scale studies and more robust predictive models for surgical outcomes. Thus, the disclosed embodiments also contributes to advancements in surgical planning and patient care in the broader field of colorectal surgery or other medical procedures involving pelvimetry.

[0034] Furthermore, the disclosed embodiments provide improvements in robotically-assisted surgical platforms and various medical instrumentation platforms in which the inferred landmarks locations and / or pelvimetry measurements can be utilized as control parameters in association with a variety of operations. For example, as explained above, having these parameters automatically and rapidly made available from preoperative images enables a robotically-assisted surgical platform or other medical instrumentation systems to operate with higher precision of movement relative to the target anatomy, operate with higher safety, and achieve better patient outcomes.

[0035] The presented system and method may also be applied to automated measurement of other pelvic dimensions. For example, the disclosed embodiments can be used in transverseMRI volumes. This would enable the automated measurement of the intraspinal and intratubular distances, which can be predictive for TME outcomes. The disclosed embodiments can also be used for automated pelvimetry in CT acquisitions. Further still, the disclosed embodiments n be applied for the detection of landmarks that are relevant to other clinical fields. For instance, certain pelvic dimensions may be predictive for the radicality of the prostatectomy.Examples

[0036] An example application of the above-described processes and associated performance results are now described in relation to a conducted study. The study was performed on patients included in the Minimally Invasive Rectal Carcinoma (MIRECA) cohort. This cohort was part of a retrospective study performed in eight institutes with a large comparison of L-TME versus R-TME versus TaTME for primary rectal cancer. Patients 18 years or older with rectal cancer (based on the sigmoidal take-off definition) that underwent tumor resection between January 2015 and December 2021 were included. Patients were enrolled for this study based on the availability, the presence of artifacts, and the field of view of MRI scans, which had to include the entire bony pelvis from the coccyx up to the sacral promontory.

[0037] In this study, the training medical images were derived from pre-operative MRI volumes using 1.5T and 3.0T MR scanners. 3D T2 -weighted TSE sequences from standardized pelvic MR imaging protocols may be clinically used for pre-operative staging of the rectal tumor. Parameters varied across equipment and differed for the TR, TE , flip-angle, slice thickness, image matrix, field of view, acquisition time.

[0038] A stratified five-fold cross-validation was used to assess the performance of the model. Each fold comprised twenty percent of the training data obtained from each of a set of eight different training data sources from different institutes. This cross-validation enabled the assessment of the performance of all patients and the performance per source. The Euclidean distance between the ground truth and model output on the test set was used to quantify the performance of the landmark localization. This localization error was bundled by the Mean Absolute Error (MAE). The difference in the measured Euclidean distance and angulations between the ground truth and the predicted landmarks was also used for the performance assessment. This measurement error was also bundled through the MAE. The coefficient of determination (R2) was also used to quantify how well the predicted measurements approximate the manual data. In addition, the inter-observer variability was computed based on the blinded annotation of 500 patients by two annotators. The variability was used to put the performance into perspective with the human capabilities.

[0039] In the example study, a total of 2292 patients were initially included in the MIRECAcohort. However, a subset of these patients did not have a pre-operative T2 -weighted sagittal MRI acquisition. In addition, based on visual inspection, another subset of patients were excluded due to the presence of imaging artifacts or incomplete depiction of the pelvic bones. After these exclusions, 1707 patients were used for the development of the model and the fivefold cross validation.

[0040] The results of the five-fold cross-validation for the example study are shown in Table 1, which states the landmark localization error per fold. The different folds show comparable performance with an MAE ranging between 5.0 and 6.7 millimeters. The interobserver variability showed an MAE of 3.7 mm based on 500 patients, while the average MAE of the automatic landmark localization is 5.9.Table 1. Localization performance for the five-fold cross-validation.

[0041] The mean error for the individual landmarks is given in Table 2. The table shows that the landmarks are localized with a comparable MAE. The localization of landmark B, the sacral vertebra 3, has both the largest mean error and standard deviation.Table 2. Localization error for the individual landmarks.

[0042] Table 3 provides an overview of the landmark localization performance for the eight different institutes and the number of patients. The number of patients per institute varies greatly, ranging from 65 to 498 patients. However, the performance per institute shows an MAE that only varies between 4.7 and 6.2 millimeters.Table 3. Localization error for the individual institutes.

[0043] Table 4 shows example statistics associated with performance of the model for pelvic dimensions computed from landmark locations predicted by the model using the above-described techniques.Table 4. Measurement error of the pelvic dimensions.

[0044] The performance results show that the model can accurately predict pelvic dimensions with a mean error of 5.6 millimeters or less. In addition, the automatically localized landmarks led to an accurate measurement of various clinically relevant dimensions. Tables 1, 2, and 3 showed that the presented method had a comparable MAE across the different data distributions, the different landmarks, and the different institutions. Therefore, the approach for landmark detection provides consistent and generalizable performance. Comparing these results with the measured interobserver variability shows that the performance of the model at least approaches or exceeds human capabilities. The MAE in Table 4 shows that the automatically localizedlandmarks can be used to measure the various pelvic dimensions accurately.Additional Considerations

[0045] The foregoing description of the embodiments has been presented for the purpose of illustration; it is not intended to be exhaustive or to limit the embodiments to the precise forms disclosed. Persons skilled in the relevant art can appreciate that many modifications and variations are possible in light of the above disclosure.

[0046] Some portions of this description describe the embodiments in terms of algorithms and symbolic representations of operations on information. These operations, while described functionally, computationally, or logically, are understood to be implemented by computer programs or equivalent electrical circuits, microcode, or the like. Furthermore, it has also proven convenient at times, to refer to these arrangements of operations as modules, without loss of generality. The described operations and their associated modules may be embodied in software, firmware, hardware, or any combinations thereof.

[0047] Any of the steps, operations, or processes described herein may be performed or implemented with one or more hardware or software modules, alone or in combination with other devices. Embodiments may also relate to an apparatus for performing the operations herein. This apparatus may be specially constructed for the required purposes, and / or it may include a general-purpose computing device selectively activated or reconfigured by a computer program stored in the computer. Such a computer program may be stored in a tangible non- transitory computer readable storage medium or any type of media suitable for storing electronic instructions and coupled to a computer system bus. Furthermore, any computing systems referred to in the specification may include a single processor or may include architectures employing multiple processor designs for increased computing capability.

[0048] Finally, the language used in the specification has been principally selected for readability and instructional purposes, and it may not have been selected to delineate or circumscribe the inventive subject matter. It is therefore intended that the scope is not limited by this detailed description, but rather by any claims that issue on an application based hereon. Accordingly, the disclosure of the embodiments is intended to be illustrative, but not limiting, of the scope of the invention, which is set forth in the following claims.

Claims

CLAIMS1. A computer-implemented method for estimating pelvic measurements from medical images comprising: obtaining a volumetric medical image of a pelvic region of a patient; applying, by a processor, a machine learning model to the volumetric medical image that is trained to infer a set of volumetric probability maps for a plurality of respective anatomical landmarks of the pelvic region, each of the volumetric probability maps representing a per-pixel array of probabilities for a location of an anatomical landmark; selecting respective locations for the respective anatomical landmarks based on the set of volumetric probability maps; computing a set of pelvimetry measurements characterizing the respective locations for the respective anatomical landmarks; and outputting the set of measurements.

2. The computer-implemented method of claim 1, wherein selecting the respective locations for the respective anatomical landmarks comprises identifying a maximum probability value in each of the volumetric probability maps.

3. The computer-implemented method of claim 1, wherein the respective anatomical landmarks include one or more of: a promotorium, an S3-vertebrae, a coccyx, a dorsal part of os pubis, and a cranial part of os pubis.

4. The computer-implemented method of claim 1, wherein the set of pelvimetry measurements comprises at least one of: a pelvic inlet measurement, a pelvic outlet measurement, a pelvic depth measurement, and a sacral angulation measurement.

5. The computer-implemented method of claim 1, wherein computing the set of pelvimetry measurements comprises computing a distance in three-dimensional space between two of the respective anatomical landmarks.

6. The computer-implemented method of claim 1, wherein computing the set of pelvimetry measurements comprises computing an angle in three-dimensional space that relates three of the respective anatomical landmarks.

7. The computer-implemented method of claim 1, wherein outputting the set of measurements comprises generating a visual representation of the pelvic region with the set of measurements overlaid on the visual representation.

8. The computer-implemented method of claim 7, wherein the visual representation comprises at least one of: a two-dimensional slice of the volumetric medical image and a three- dimensional rendering of the pelvic region.

9. The computer-implemented method of claim 1, wherein outputting the set of measurements comprises: outputting the set of measurements as a control parameter that affects operation of a robotically-assisted surgical platform during performance of a medical procedure.

10. The computer-implemented method of claim 1, wherein outputting the set of measurements comprises: outputting the set of measurements as a control parameter that affects operation of a non- robotic medical instrument during performance of a medical procedure.

11. The method of claim 1, wherein the machine learning model is trained according to a process comprising: obtaining a training image set comprising a set of volumetric medical images depicting pelvic regions of patients; obtaining for each of the volumetric medical images, annotations indicative of locations of a plurality of anatomical landmarks in the volumetric medical images; for each of the volumetric medical images, deriving a set of probability maps that each correspond to one of the anatomical landmarks and comprise an array of probabilities derived from applying a Gaussian fdter to respective arrays encoding the respective locations of the plurality of anatomical landmarks; and applying a machine learning algorithm to learn model parameters for the machine learning model for mappings the training image set to the set of probability maps.

12. A non -transitory computer-readable storage medium storing instructions for estimating pelvic measurements from medical images, the instructions when executed by one or more processors causes the one or more processors to perform steps including: obtaining a volumetric medical image of a pelvic region of a patient; applying a machine learning model to the volumetric medical image that is trained to infer a set of volumetric probability maps for a plurality of respective anatomical landmarks of the pelvic region, each of the volumetric probability maps representing a per-pixel array of probabilities for a location of an anatomical landmark;selecting respective locations for the respective anatomical landmarks based on the set of volumetric probability maps; computing a set of pelvimetry measurements characterizing the respective locations for the respective anatomical landmarks; and outputting the set of measurements.

13. The non-transitory computer-readable storage medium of claim 12, wherein selecting the respective locations for the respective anatomical landmarks comprises identifying a maximum probability value in each of the volumetric probability maps.

14. The non-transitory computer-readable storage medium of claim 12, wherein outputting the set of measurements comprises generating a visual representation of the pelvic region with the set of measurements overlaid on the visual representation.

15. The non-transitory computer-readable storage medium of claim 14, wherein the visual representation comprises at least one of: a two-dimensional slice of the volumetric medical image and a three-dimensional rendering of the pelvic region.

16. The non-transitory computer-readable storage medium of claim 12, wherein outputting the set of measurements comprises: outputting the set of measurements as a control parameter that affects operation of a robotically-assisted surgical platform during performance of a medical procedure.

17. The non-transitory computer-readable storage medium of claim 12, wherein outputting the set of measurements comprises: outputting the set of measurements as a control parameter that affects operation of a non- robotic medical instrument during performance of a medical procedure.

18. The non-transitory computer-readable storage medium of claim 12, wherein the machine learning model is trained according to a process comprising: obtaining a training image set comprising a set of volumetric medical images depicting pelvic regions of patients; obtaining for each of the volumetric medical images, annotations indicative of locations of a plurality of anatomical landmarks in the volumetric medical images; for each of the volumetric medical images, deriving a set of probability maps that each correspond to one of the anatomical landmarks and comprise an array of probabilities derived from applying a Gaussian fdter to respective arrays encoding the respective locations of the plurality of anatomical landmarks; andapplying a machine learning algorithm to learn model parameters for the machine learning model for mappings the training image set to the set of probability maps.

19. A computer system, comprising: one or more processors; and a non-transitory computer-readable storage medium storing instructions for estimating pelvic measurements from medical images, the instructions when executed by the one or more processors causes the one or more processors to perform steps including: obtaining a volumetric medical image of a pelvic region of a patient; applying a machine learning model to the volumetric medical image that is trained to infer a set of volumetric probability maps for a plurality of respective anatomical landmarks of the pelvic region, each of the volumetric probability maps representing a per-pixel array of probabilities for a location of an anatomical landmark; selecting respective locations for the respective anatomical landmarks based on the set of volumetric probability maps; computing a set of pelvimetry measurements characterizing the respective locations for the respective anatomical landmarks; and outputting the set of measurements.

20. The computer system of claim 19, wherein the machine learning model is trained according to a process comprising: obtaining a training image set comprising a set of volumetric medical images depicting pelvic regions of patients; obtaining for each of the volumetric medical images, annotations indicative of locations of a plurality of anatomical landmarks in the volumetric medical images; for each of the volumetric medical images, deriving a set of probability maps that each correspond to one of the anatomical landmarks and comprise an array of probabilities derived from applying a Gaussian fdter to respective arrays encoding the respective locations of the plurality of anatomical landmarks; and applying a machine learning algorithm to learn model parameters for the machine learning model for mappings the training image set to the set of probability maps.

Images