Synthetic organ modeling

Abstract

A method for generating synthetic data from a virtual three-dimensional (3D) model includes receiving a parameter value for an adjustable parameter. The parameter value is associated with a severity of a condition of an organ. The method includes generating, using the parameter value, a virtual 3D model of the organ. The method also includes generating, from a virtual point of view within the virtual 3D model of the organ, an image capturing at least a portion of the virtual 3D model of the organ. The method also includes estimating, using an evaluation model, the severity of the condition of the organ based on the image.

Description

Synthetic Organ ModelingTECHNICAL FIELD

[0001] This disclosure relates to synthetic organ modeling.BACKGROUND

[0002] Video capsule endoscopy (VCE) is a procedure used to capture images by a camera located within the human body. For example, a patient may swallow a “pill camera” and the camera will capture images as the camera passes through the gastrointestinal tract. These images provide valuable data for a variety of uses (e.g., diagnostic, training, etc.) but are difficult and expensive to acquire in any significant quantities.SUMMARY

[0003] One aspect of the disclosure provides a computer-implemented method that when executed on data processing hardware causes the data processing hardware to perform operations for generating synthetic or virtual images of an organ. The operations include receiving a parameter value for an adjustable parameter. The parameter value may be associated with a severity of a condition of an organ. Alternatively, the parameter value is associated with a positive characteristic of an organ. The operations also include generating, using the parameter value, a virtual three-dimensional (3D) model of the organ. The operations also include generating, from a virtual point of view within the virtual 3D model of the organ, an image capturing at least a portion of the virtual 3D model of the organ. The operations also include estimating, using an evaluation model, the severity of the condition of the organ based on the image.

[0004] Implementations of the disclosure may include one or more of the following optional features. In some implementations, the organ includes an intestine. In these implementations, the adjustable parameter includes one of a height of villi within the intestine, a model of villi within the intestine, a density of villi within the intestine, a color of villi within the intestine, or a crypt depth severity of the intestine. Generating thevirtual 3D model of the organ may include transmitting an application programming interface (API) request to a 3D modeling application.

[0005] In some examples, the operations further include generating a plurality of images each capturing a respective portion of the virtual 3D model of the organ and estimating, using the evaluation model, the severity of the condition of the organ based on each image in the plurality of images. Here, the operations may further include generating a video file based on a concatenation of the plurality of images. In these examples, each image in the plurality of images may be generated from a different virtual point of view within the virtual 3D model of the organ. Here, the different virtual point of view of each image in the plurality of images simulates movement through the virtual 3D model of the organ.

[0006] The model may include a machine learning model. In some implementations, the operations further include associating metadata with the image. The metadata includes the parameter value. In these implementations, the operation may further include storing the image and the associated metadata in a catalog indexed by the metadata. Generating the virtual 3D model of the organ may be based on a second adjustable parameter that includes one of a video capsule endoscopy (VCE) camera parameter, a lighting parameter, a type of the organ, or a position parameter representing a position within the 3D model of the organ. In some examples, the operations further include validating the evaluation model based on the estimation of the severity of the condition and the parameter value. Generating the virtual 3D model of the organ may include selecting a villi model from among a plurality of villi models based on the parameter value.

[0007] Another aspect of the disclosure provides a system that includes data processing hardware and memory hardware storing instructions that when executed on the data processing hardware causes the data processing hardware to perform operations. The operations include receiving a parameter value for an adjustable parameter. The parameter value is associated with a severity of a condition of an organ. The operations also include generating, using the parameter value, a virtual three-dimensional (3D) model of the organ. The operations also include generating, from a virtual point of viewwithin the virtual 3D model of the organ, an image capturing at least a portion of the virtual 3D model of the organ. The operations also include estimating, using an evaluation model, the severity of the condition of the organ based on the image.

[0008] Implementations of the disclosure may include one or more of the following optional features. In some implementations, the organ includes an intestine. In these implementations, the adjustable parameter includes one of a height of villi within the intestine, a model of villi within the intestine, a density of villi within the intestine, a color of villi within the intestine, or a crypt depth severity of the intestine. Generating the virtual 3D model of the organ may include transmitting an application programming interface (API) request to a 3D modeling application.

[0009] In some examples, the operations further include generating a plurality of images each capturing a respective portion of the virtual 3D model of the organ and estimating, using the evaluation model, the severity of the condition of the organ based on each image in the plurality of images. Here, the operations may further include generating a video file based on a concatenation of the plurality of images. In these examples, each image in the plurality of images may be generated from a different virtual point of view within the virtual 3D model of the organ. Here, the different virtual point of view of each image in the plurality of images simulates movement through the virtual 3D model of the organ.

[0010] The model may include a machine learning model. In some implementations, the operations further include associating metadata with the image. The metadata includes the parameter value. In these implementations, the operation may further include storing the image and the associated metadata in a catalog indexed by the metadata. Generating the virtual 3D model of the organ may be based on a second adjustable parameter that includes one of a video capsule endoscopy (VCE) camera parameter, a lighting parameter, a type of the organ, or a position parameter representing a position within the 3D model of the organ. In some examples, the operations further include validating the evaluation model based on the estimation of the severity of the condition and the parameter value. Generating the virtual 3D model of the organ mayinclude selecting a villi model from among a plurality of villi models based on the parameter value.

[0011] Another aspect of the disclosure provides a system that includes a reservoir containing a solution. The system also includes a physical phantom of an organ having a hollow portion that has a proximal end and a distal end. The physical phantom is immersed within the solution of the reservoir and represents a condition of the organ. The system also includes an ingestible data capture device operable to capture sensor data and at least one actuator that controls movement of the ingestible data capture device through the hollow portion of the physical phantom. The system also includes an evaluation model.

[0012] Implementations of the disclosure may include one or more of the following optional features. In some implementations, the system further includes data processing hardware and memory hardware in communication with the data processing hardware. The memory hardware stores instructions that when executed on the data processing hardware causes the data processing hardware to perform operations. The operations include instructing the ingestible data capture device to traverse through the hollow portion of the physical phantom according to a predetermined path and receiving, from the ingestible data capture device, sensor data captured as the ingestible data capture device traverses through the hollow portion of the physical phantom. The operations also include estimating a severity of the condition based on the received sensor data using the evaluation model and validating at least one of the physical phantom, the ingestible data capture device, or the evaluation model based on the severity of the condition. In some implementations, the system further includes a second physical phantom of the organ including a second hollow portion that has a second proximal end and a second distal end. The second physical phantom is immersed within the solution of the reservoir and represents a second condition of the organ. The physical phantom is different from the second physical phantom. In these implementations, the operations further include: instructing the ingestible data capture device to traverse through the second hollow portion of the second physical phantom according to the predetermined path; receiving, from the ingestible data capture device, second sensor data captured as the ingestible datacapture device moves through the second hollow portion of the second physical phantom; and estimating, using the evaluation model, a second severity of the second condition based on the received second sensor data. Here, validating the at least one of the physical phantom, the ingestible data capture device, or the evaluation model includes comparing the severity of the condition for the physical phantom with the second severity of the condition for the second physical phantom. In these implementations, the condition of the organ represented by the physical phantom may represent a different condition than the second condition of the organ represented by the second physical phantom.

[0013] In some examples, the system further includes a second ingestible data capture device operable to capture sensor data. The ingestible data capture device is different from the second ingestible data capture device. In these examples, the operations further include: instructing the second ingestible data capture device to traverse through the hollow portion of the physical phantom according to the predetermined path; receiving, from the second ingestible data capture device, second sensor data captured as the second ingestible data capture device moves through the hollow portion of the physical phantom; and estimating, using the evaluation model, a second severity of the condition based on the received second sensor data. Here, validating the at least one of the physical phantom, the ingestible data capture device, or the evaluation model includes comparing the severity of the condition for the physical phantom with the second severity of the condition for the physical phantom. In these examples, the ingestible data capture device includes at least one different optical characteristic than the second ingestible data capture device.

[0014] In some implementations, the system further includes a second evaluation model different from the evaluation model. In these implementations, the operations further include: instructing the ingestible data capture device to traverse through the hollow portion of the physical phantom according to the predetermined path; receiving, from the ingestible data capture device, second sensor data captured as the ingestible data capture device moves through the hollow portion of the physical phantom; and estimating, using the second evaluation model, a second severity of the second condition based on the received second sensor data. Here, validating the at least one of the physicalphantom, the ingestible data capture device, or the evaluation model includes comparing the severity of the condition for the physical phantom with the second severity of the condition for the physical phantom. In these examples, the evaluation model includes one or more parameters where the evaluation model estimates the severity of the condition using the one or more parameters and the second estimation model includes one or more second parameters where the second estimation model estimates the severity of the condition using the one or more second parameters. The one or more parameters are different from the one or more second parameters.

[0015] The operations may further include: instructing the ingestible data capture device to traverse through the hollow portion of the physical phantom according to a second predetermined path different from the predetermined path; receiving, from the ingestible data capture device, second sensor data captured as the ingestible data capture device moves through the hollow portion of the physical phantom according to the second predetermined path; and estimating, using the evaluation model, a second severity of the condition based on the received second sensor data. Here, validating the at least one of the physical phantom, the ingestible data capture device, or the estimation model includes comparing the severity of the condition for the physical phantom with the second severity of the condition for the physical phantom. In some implementations, the operations further include obtaining example path data samples and determining, based on the example data samples, one or more predetermined paths traversing through the hollow portion of the physical phantom. Each respective example path data sample includes corresponding movement data and corresponding angle data of an example ingestible data capture device traversing through an example organ. Each respective predetermined path of the one or more predetermined determined paths defining a corresponding movement and a corresponding angle of the ingestible data capture device traversing the each position of the hollow portion of the physical phantom. The one or more predetermined paths includes at least the predetermined path.

[0016] In some examples, the system further includes a three-dimensional (3D) printer configured to print one or more physical phantoms of the organ and a 3D printing model including a set of adjustable printing parameters each influencing a respectiveattribute of a respective physical phantom printed by the 3D printer. In these examples, the operations further include instructing the 3D printer to print one or more physical phantoms of the organ according to the 3D printing model. The one or more physical phantoms include at least the physical phantom. In some implementations, the system further includes: a wire including a proximal end and a distal end, the ingestible data capture device coupled to the distal end of the wire; a first actuator that engages the wire, the first actuator operable to control movement of the ingestible capture device from the proximal end to the distal end of the hollow portion of the physical phantom; and a second actuator operable to control an angle of the ingestible data capture device relative to the distal end of the wire. In these implementations, instructing the ingestible data capture device to traverse through the hollow portion of the physical phantom includes controlling at least one of the movement or the angle of the ingestible data capture device. The ingestible data capture device may include an ingestible image capture device and the sensor data may include images. The ingestible data capture device includes an image endoscopy capsule.

[0017] Another aspect of the disclosure provides a computer-implemented method that when executed on data processing hardware causes the data processing hardware to perform operations for validating an estimation model that estimates a severity of a condition of a three-dimensional printed physical phantom of an organ. The operations include instructing an ingestible data capture device to traverse through a hollow portion of a physical phantom according to a predetermined path. The physical phantom is immersed within a solution of a reservoir and represents the condition of an organ. The operations also include receiving, from the ingestible data capture device, sensor data captured as the ingestible data capture device traverses through the hollow portion of the physical phantom. The operations also include estimating, using an evaluation model, a severity of the condition based on the received sensor data. The operations also include validating, based on the severity of the condition, at least one of the physical phantom, the ingestible data capture device, or the evaluation model.

[0018] Implementations of the disclosure may include one or more of the following optional features. In some implementations, the operations further include: instructingthe ingestible data capture device to traverse through a second hollow portion of a second physical phantom according to the predetermined path, the second physical phantom immersed within the solution of the reservoir and representing a second condition of the organ, the physical phantom different from the second physical phantom; receiving, from the ingestible data capture device, second sensor data captured as the ingestible data capture device moves through the second hollow portion of the second physical phantom; and estimating, using the evaluation model, a second severity of the second condition based on the received second sensor data. Here, validating the at least one of the physical phantom, the ingestible data capture device, or the evaluation model includes comparing the severity of the condition for the physical phantom with the second severity of the condition for the second physical phantom. In these implementations, the condition of the organ represented by the physical phantom may represent a different condition than the second condition of the organ represented by the second physical phantom.

[0019] In some examples, the operations further include: instructing a second ingestible data capture device to traverse through the hollow portion of the physical phantom according to the predetermined path, the ingestible data capture device different from the second ingestible data capture device; receiving, from the second ingestible data capture device, second sensor data captured as the second ingestible data capture device moves through the hollow portion of the physical phantom; and estimating, using the evaluation model, a second severity of the condition based on the received second sensor data. Here, validating the at least one of the physical phantom, the ingestible data capture device, or the evaluation model includes comparing the severity of the condition for the physical phantom with the second severity of the condition for the physical phantom. In these examples, the ingestible data capture device may include at least one different optical characteristic than the second ingestible data capture device.

[0020] In some implementations, the operations further include: instructing the ingestible data capture device to traverse through the hollow portion of the physical phantom according to the predetermined path; receiving, from the ingestible data capture device, second sensor data captured as the ingestible data capture device moves through the hollow portion of the physical phantom; and estimating, using a second evaluationmodel, a second severity of the second condition based on the received second sensor data. Here, validating the at least one of the physical phantom, the ingestible data capture device, or the evaluation model includes comparing the severity of the condition for the physical phantom with the second severity of the condition for the physical phantom. In these implementations, the evaluation model may include one or more parameters where the evaluation model estimates the severity of the condition using the one or more parameters and the second estimation model includes one or more second parameters where the second estimation model estimates the severity of the condition using the one or more second parameters. The one or more parameters are different from the one or more second parameters.

[0021] In some examples, the operations further include: instructing the ingestible data capture device to traverse through the hollow portion of the physical phantom according to a second predetermined path, the predetermined path different from the second predetermined path; receiving, from the ingestible data capture device, second sensor data captured as the ingestible data capture device moves through the hollow portion of the physical phantom according to the second predetermined path; and estimating, using the evaluation model, a second severity of the condition based on the received second sensor data. In these examples, validating the at least one of the physical phantom, the ingestible data capture device, or the estimation model includes comparing the severity of the condition for the physical phantom with the second severity of the condition for the physical phantom. The operations may further include obtaining example path data samples and determining, based on the example path data samples, one or more predetermined paths traversing through the hollow portion of the physical phantom. Each respective example path data sample including corresponding movement data and corresponding angle data of an example ingestible data capture device traversing through an example organ. Each respective predetermined path of the one or more predetermined determined paths defining a corresponding movement and a corresponding angle of the ingestible data capture device traversing the each position of the hollow portion of the physical phantom. The one or more predetermined paths including at least the predetermined path.

[0022] In some implementations, the operations further include instructing a three- dimensional (3D) printer including a 3D printing model to print one or more physical phantoms of the organ according to the 3D printing model with the one or more physical phantoms that includes at least the physical phantom. The 3D printing model includes a set of adjustable printing parameters each influencing a respective attribute of a respective physical phantom printed by the 3D printer. Instructing the ingestible data capture device to traverse through the hollow portion of the physical phantom according to the predetermined path may include instructing at least one actuator to control a movement of the ingestible data capture device and an angle of the ingestible data capture device. In some examples, the ingestible data capture device includes an ingestible image capture device and the sensor data includes images. The ingestible data capture device may include an image endoscopy capsule.

[0023] Another aspect of the disclosure includes receiving a parameter value for an adjustable parameter that is associated with a severity of a condition of an organ. The method includes generating, using the parameter value, a virtual three-dimensional (3D) model of the organ and generating, from a virtual point of view within the virtual 3D model of the organ, an image capturing at least a portion of the virtual 3D model of the organ. The method includes estimating, using an evaluation model, a first severity of the condition of the organ based on the image. The method also includes printing, using the generated virtual 3D model of the organ, a physical phantom comprising a hollow portion and instructing an ingestible data capture device to traverse through the hollow portion of the physical phantom according to a predetermined path. The physical phantom is immersed within a solution of a reservoir and represents a condition of an organ. The method includes receiving, from the ingestible data capture device, sensor data captured as the ingestible data capture device traverses through the hollow portion of the physical phantom. The method also includes estimating, using an evaluation model, a second severity of the condition based on the received sensor data and comparing the first severity and second severity.

[0024] The details of one or more implementations of the disclosure are set forth in the accompanying drawings and the description below. Other aspects, features, and advantages will be apparent from the description and drawings, and from the claims.DESCRIPTION OF DRAWINGS

[0025] FIG. 1 is a schematic view of an example system for generating synthetic data based on a parameterized virtual 3D model of an organ.

[0026] FIGS. 2A and 2B illustrate exemplary options for a modeling tool for implementing adjustable parameters of a virtual 3D model of an organ.

[0027] FIGS. 3A and 3B illustrate villi models with different variations.

[0028] FIG. 4 illustrates exemplary images of aspects of a generated virtual 3D model.

[0029] FIG. 5 illustrates additional exemplary images of aspects of a generated virtual 3D model.

[0030] FIG. 6 is a flowchart of an exemplary arrangement of operations for a computer-implemented method for validating an estimation model based on synthetic data generated from a parameterized virtual 3D model.

[0031] FIG. 7 is a schematic view of an example system for validating an estimation model that estimates a severity of a condition of a three-dimensional printed physical phantom of an organ.

[0032] FIG. 8 is a schematic view of an example test environment for validating the estimation model.

[0033] FIG. 9 is a schematic view of an example actuator for controlling an ingestible data capture device.

[0034] FIG. 10 is a schematic view of an example controller for operating one or more actuators.

[0035] FIG. 11 illustrates an example physical phantom.

[0036] FIG. 12 illustrates another example physical phantom.

[0037] FIG. 13 illustrates two example sensor data points captured by the ingestible data capture device.

[0038] FIG. 14 illustrates a retention element of a wire that secures the ingestible data capture device.

[0039] FIG. 15, illustrates an exploded view of the retention element.

[0040] FIG. 16 is a flowchart of an exemplary arrangement of operations for a computer-implemented method for validating an estimation model that estimates a severity of a condition of a three-dimensional printed physical phantom of an organ.

[0041] FIG. 17 is a schematic view of an example computing device that may be used to implement the systems and methods described herein.

[0042] Like reference symbols in the various drawings indicate like elements.DETAILED DESCRIPTION

[0043] A video capsule endoscopy (VCE) is a noninvasive technique for diagnosing digestive issues where a patient swallows a small camera that captures images as the camera passes through the patient’s gastrointestinal tract. Image data captured via VCE or other procedures is valuable in many medical applications. However, acquiring this data faces many challenges. For example, one of the challenges is the difficulty in generating a large quantity of data from actual patients due to expenses, patient regulatory compliance, and other complications. More specifically, VCE data is inherently stochastic, meaning it exhibits randomness due to various factors such as camera movement, lighting conditions, and physiological variations. Additionally, artifacts (such as motion blur or interference) can affect the quality of captured images. Furthermore, VCE data may be imbalanced, with certain conditions occurring less frequently. These data issues may therefore affect the performance of machine learning models trained on such data that could otherwise be beneficial to gastrointestinal health analysis (e.g., disease detection based on the gastrointestinal state).

[0044] Implementations herein include a synthetic organ modeler for generating virtual three-dimensional (3D) models of organs (e.g., intestines). The synthetic organ modeler may generate the virtual 3D model based on one or more adjustable parameters. After generating the virtual 3D model, the synthetic organ modeler may generate, from a virtual point of view within the virtual 3D model, an image that captures at least a portionof the virtual 3D model. In some implementations, a model or algorithm processes the generated image to estimate or characterize one or more characteristics of the organ (e.g., estimate a severity of a condition of the organ, estimate a healthiness of an organ, or otherwise estimate or quantify a health state of the organ). For example, a model (e.g., a machine learning model) may score the generated image to estimate or predict a severity of a disease. As used herein, severity may represent a healthy and / or diseased organ. For instance, a relatively high severity may indicate negative characteristics of a diseased organ while a relatively low severity may indicate positive characteristics of a healthy organ.

[0045] Referring to FIG. 1 , in some implementations, an example system 100 includes a processing system 10. The processing system 10 may be a single computer, multiple computers, or a distributed system (e.g., a cloud environment) having fixed or scalable elastic computing resources 12 (e.g., data processing hardware) and / or storage resources 14 (e.g., memory hardware). The processing system 10 may be in communication with a user device 16 associated with a respective user 22 via, for example, the network 24. The user device 16 may correspond to any computing device, such as a desktop workstation, a laptop workstation, or a mobile device (i.e., a smart phone). The user device 16 includes computing resources 18 (e.g., data processing hardware) and / or storage resources 20 (e.g., memory hardware).

[0046] The processing system 10 executes a virtual synthetic organ modeler 110. The virtual synthetic organ modeler 110 (also referred to herein as the modeler 110), in some implementations, receives a virtual model request 30 requesting that the virtual synthetic organ modeler 110 generate a virtual 3D model 115 (also referred to herein as the model 115) of an organ or other collection of tissues. The virtual model request 30 may originate from the user 22 via the user device 16 (e.g., transferred via the network 24). In some examples, the user 22 interacts directly with the processing system 10 to provide the virtual model request 30. Optionally, the virtual model request 30 defines or selects the organ or tissue to model.

[0047] The virtual model request 30 for the virtual 3D model 115 may be for any organ or tissue. In some examples, the organ is an intestine. Optionally, the 3D model115 is of a particular portion of an organ. For example, the 3D model 115 is of only the duodenum, only the jejunum, or only the ileum of the small intestine. In another example, the model 115 includes multiple portions of the organ (e.g., the duodenum and the jejunum). In yet another example, the model 115 is of the organ. Other organs or tissues may also be modeled, such as the large intestine, the esophagus, the stomach, etc.

[0048] The virtual model request 30 includes one or more respective parameter values 34 for one or more adjustable parameters 32. Each adjustable parameter 32 controls or adjusts an aspect of the model 115. For example, when the model 115 is of a portion or a surface of an intestine, the adjustable parameter 32 may be for villi height. Each adjustable parameter 32 has a range of possible parameter values 34 between a minimum parameter value 34 and a maximum parameter value 34. The virtual model request 30 may include any number of parameter values 34 for any number of adjustable parameters 32 simultaneously. In some examples, the modeler 110 sets any adjustable parameters 32 not set to a particular parameter value 34 via the request to a default or initial parameter value 34. The modeler 110 may prompt the user 22 for a missing parameter value 34 when necessary or helpful for generating the model 115.

[0049] The modeler 110, in some implementations, includes a modeler controller 112. The modeler controller 112 receives the virtual model request 30 and determines the parameter values 34 for all relevant adjustable parameters 32 based on the virtual model request 30. In some examples, the modeler controller 112 adjusts a default parameter value 34 for one or more adjustable parameters 32 based on the parameter values 34 received in the virtual model request 30. For example, a particular parameter value 34 of a first adjustable parameter 32 may dictate a particular parameter value 34 of a second adjustable parameter 32. In another example, a parameter value 34 provided by the virtual model request 30 may not be feasible and the modeler controller 112 may select the closest feasible parameter value 34.

[0050] The modeler controller 112 generates, based on the adjustable parameters 32, the parameter values 34, and the virtual model request 30, a model generate command 113 (also referred to herein as the command 113). The model generate command 113 instructs a 3D modeler 114 to generate the model 115 requested by the virtual modelrequest 30. In some examples, the modeler controller 112 parameterizes the parameter values 34 into the command 113 and communicates the command 113 to the 3D modeler 114 via an application programming interface (API) or the like (e.g., using a PYTHON® script or the like). The 3D modeler 114 may be any applicable 3D modeling application (e.g., BLENDER®). The 3D modeler 114 receives the model generate command 113 and generates the model 115 based on the command 113. The 3D modeler 114 may generate the model 115 using any suitable file type.

[0051] In some implementations, the modeler controller 112, based on the adjustable parameters 32, adjusts any number of options or attributes of the model 115. That is, in some examples, the adjustable parameter 32 and parameter value 34 refer directly to an option to be adjusted by the 3D modeler 114. For example, the adjustable parameter 32 may be for a scale (e.g., the scale or size of the organ or a particular feature of the organ, such as villi), the parameter value 34 may represent the scale value, and the 3D modeler 114 may directly adjust a scale option based on the parameter value 34. In other examples, the adjustable parameter 32 is associated with one or more options in the 3D modeler 114. The modeler controller 112 may translate or otherwise convert the parameter value(s) 34 into the necessary options or attributes of the 3D modeler. For example, the adjustable parameter 32 may represent a color of the organ, and the modeler controller 112 may translate the adjustable parameter 32 into a number of options for the 3D modeler 114, such as a base color, a subsurface color, a sheen, a tint, a texture, etc.

[0052] As shown in FIGS. 2A and 2B, the adjustable parameters 32 may be translated into any number of configurable options for the 3D modeler 114. Here, FIG. 2A includes an image 122 A of configurable options for a healthy organ, while FIG. 2B includes an image 122B of configurable options for a diseased organ. Based on the adjustable parameters 32 and the corresponding parameter values 34, the modeler controller 112 generates a command 113 that configures the 3D modeler 114 with, for example, some or all of these options.

[0053] The adjustable parameters 32 may adjust any number of features of the organ or even the type, orientation, or position of the organ. For example, when the organ is an intestine, the adjustable parameters 32 control or define a height of villi, a density of villi,a color of villi or other tissue, a crypt depth severity, a position within the intestine, etc. Moreover, the adjustable parameters 32 may control or define scalloping and / or flattening of duodenal folds, fissuring over the folds, and a mosaic pattern of mucosa of folds, cancerous lesions, ulcers, and / or wall thinning. Thus, the adjustable parameters 32 may be adjusted to model disease and non-disease attributes of organs. As shown in FIGS. 3A and 3B, in some implementations, the adjustable parameters 32 include instructing the 3D modeler 114 to select a particular model from a set of models for one or more features of the organ. For example, an adjustable parameter 32 may include a value associated with one or more particular models from a set of available models, and the value instructs the 3D modeler 114 to use or select the models associated with the value. Additionally or alternatively, the adjustable parameters 32 may cause selection of a distribution or density of different models or variations in the different models. In some examples, the adjustable parameters 32 define “macro-models” of the organ (i.e., “tissue subassemblies”) that serve as libraries of building blocks with different organ characteristics to ease assembly of a larger model and to smooth across assembled boundaries.

[0054] FIG. 3 A includes an image 130A of four villi models. In some examples, the 3D modeler 114 uses multiple different models (e.g., each of the four models in FIG. 3 A) to provide variety for the generated model 115. When modeling an organ, for instance, having a variety of virtual models for components / subcomponents of the virtual organ can more accurately correspond to tissue variation that is present in a real organ; allowing a virtual model of an organ to represent a real organ more accurately. The adjustable parameter 32 may adjust a resolution, a polygon count, or otherwise a quality of the models. For example, FIG. 3B includes an image 13 OB of the villi models of FIG. 3 A with a lower polygon count.

[0055] In some implementations, the adjustable parameters 32 include parameters that mimic different ways physical cameras capture images. More specifically, the adjustable parameters 32 may include, for example, a video capsule endoscopy (VCE) camera parameter (e.g., selecting a brand, type, size, etc. of camera), a lighting parameter to simulate the type of light produced by the camera, or other parameters that define the different means for capturing images of the organ. In this way, the adjustable parameters32 may be able to simulate different types of procedures using different types or brands or models of equipment. For example, a first type of pill camera may use lighting of a first temperature, a first lumens output, etc., while a second type of pill camera may use lighting of a second temperature, a second lumens output, etc.

[0056] Optionally, the adjustable parameters 32 may more broadly indicate a desired severity of a particular condition, and the modeler controller 112 translates this (e.g., via a lookup table or the like) into the options and the command 113 for the 3D modeler 114. For example, the adjustable parameter 32 may indicate a severity of villous atrophy, crypt hyperplasia, tissue discoloration, etc. The options for the 3D modeler may include selections of particular textures from a set or group of textures and options that configure the textures. In some examples, the adjustable parameters 32 indicate a positive characteristic of the organ. That is, the adjustable parameters 32 may be used to represent a range of organ states from healthy to diseased.

[0057] In some examples, the 3D modeler 114 (or a separate application capable of viewing or rendering the generated 3D model 115) generates, from a virtual point of view, an image 116 that captures at least a portion of the model 115. The virtual point of view, in some examples, is within the model 115 (e.g., when the model 115 represents an intestine, the virtual point of view is within the intestine). In other examples, the virtual point of view is outside of the model 115. The parameters of the virtual point of view (e.g., the location, the orientation, etc.) may be based on the virtual model request 30. That is, the virtual model request 30 may define, to some extent, the virtual point of view. In some implementations, the 3D modeler 114 generates multiple images 116 that each include at least a portion of the model 115. In these implementations, each image 116 may be captured from a different virtual point of view.

[0058] Referring back to FIG. 1, optionally, the virtual synthetic organ modeler 110 generates metadata 117 based on the adjustable parameters 32 and / or parameter values 34 used to generate the model 115 and associates the generated metadata 117 with the 3D model 115 and / or images 116. For example, the metadata 117 includes a record of the parameter values 34 used when generating the model 115. In some examples, the metadata 117 includes other information, such as a model creation date, informationregarding the 3D modeler 114 (e.g., version), and any other information useful in describing and / or recreating the model 115.

[0059] After the 3D modeler 114 has generated the model 115 and, optionally, generated one or more images 116, the virtual synthetic organ modeler 110 may store the model 115 and / or the image(s) 116 at a catalog 118. The catalog 118 may be indexed by the metadata 117. That is, a user may search the catalog 118 for models 115 and / or images 116 associated with the metadata 117. For example, when the user 22 desires an image 116 captured from a model 115 with a specific villi height, the user 22 searches the catalog 118 using the specific villi height and the catalog 118, based on the metadata 117, returns to the user 22 each model 115 and / or image 116 (or records of such) to the user 22.

[0060] In some implementations, using one or more images 116 generated by the modeler 110, an evaluation model 119 estimates or predicts a severity 120 or characteristic (e.g., a health estimate that estimates a healthiness or unhealthiness of the organ) of a condition of the organ represented by the model 115. That is, the evaluation model 119 may process one or more images 116 and, based on features within the images 116 (e.g., features based on the parameter values 34 of one or more adjustable parameters 32), estimate a condition or a severity 120 of a condition (e.g., a disease) or any other characteristic of the organ. For example, when the evaluation model 119 evaluates an image 116 generated from a model 115 with an adjustable parameter 32 at a first parameter value 34, the evaluation model 119 estimates or predicts a first score or value. When the evaluation model 119 evaluates another image 116 generated from another model 115 with the adjustable parameter 32 at a second parameter value 34 different from the first parameter value 34, the evaluation model 119 estimates or predicts a second score or value that may be different from the first score or value because of the differences between the first parameter value 34 and the second parameter value 34. The evaluation model 119 may receive the images 116 directly from the modeler 110.Additionally, or alternatively, the evaluation model 119 may receive the images 116 from the catalog 118.

[0061] In some implementations, the evaluation model 119 is a Convolutional Neural Network (CNN). The CNN architecture is designed to process an input image 116 as a two-dimensional array of pixel values. The network may include a series of convolutional layers, each applying a set of learnable filters or kernels to the input to create feature maps that detect low-level features such as edges, colors, and textures. Each convolutional layer is typically followed by a non-linear activation function, such as a Rectified Linear Unit (ReLU), and a pooling layer, such as a max-pooling layer, which down-samples the feature maps to reduce dimensionality and create translational invariance. After several such convolutional and pooling blocks, the resulting high-level feature maps may be flattened into a one-dimensional vector. This vector is provided to one or more fully- connected dense layers, which perform classification or regression based on the extracted features. The final output layer of the CNN may be configured to produce the severity 120. For example, if the severity 120 is a continuous score, the output layer may consist of a single neuron with a linear activation function.Alternatively, if the severity 120 corresponds to discrete classes (e.g., a Marsh score for celiac disease), the output layer may use a softmax activation function to produce a probability distribution across the defined classes.

[0062] The training process for the evaluation model 119 may employ a hybrid approach that leverages both real-world and synthetic data to overcome the limitations of relying on either alone. Initially, the evaluation model 119 may be pre-trained on a limited dataset of authentic images captured from real patient procedures, where each image has been manually annotated with a ground-truth severity by an expert clinician. Due to the scarcity of real-world data, particularly for rare or extreme conditions, this initial model may lack robustness. To address this, the virtual synthetic organ modeler 110 may be used to generate a large and diverse set of synthetic images 116. Because these synthetic images 116 are generated from a model 115 with known parameter values 34, they are perfectly and automatically annotated with ground-truth metadata 117. This synthetic dataset, optionally, is used to augment the real-world data, and the evaluation model 119 is further trained or fine-tuned on this combined dataset. During training, a loss function (e.g., mean squared error for regression or cross-entropy for classification)quantifies the difference between the model’s predicted severity 120 and the ground- truth severity. An optimization algorithm, such as stochastic gradient descent, iteratively adjusts the weights of the CNN to minimize this loss, thereby improving the model’s accuracy, generalization, and reliability.

[0063] In some examples, the 3D modeler 114 generates multiple images 116 (each from the same or from a different virtual point of view) and the virtual synthetic organ modeler 110 generates a video file by concatenating the images 116 together using any applicable means. When the images 116 are captured from different virtual points of view, the video file may simulate movement through the virtual 3D model 115 of the organ. For example, the video file may simulate a camera, such as a pill camera or the like, that captures a video of images as the camera moves through the organ (e.g., an internal cavity or chamber of an organ). Optionally, the evaluation model 119 scores or evaluates multiple images 116 in the video file. The evaluation model 119 may score or evaluate each image 116 individually (e.g., estimate a severity 120 of a condition present in each image 116) and / or determine an overall estimate of the severity 120 of the condition based on the entirety of the video file. In some examples, the condition is a disease (e.g., celiac disease). In other examples, the condition is inflammation, scarring, or any other anomalous or pathogenic state.

[0064] In this way, the virtual synthetic organ modeler 110 generates synthetic data to train, test, validate, explain, or otherwise exercise the evaluation model 119 (e.g., based on a comparison of the output of the evaluation model 119 and a ground truth evaluation based on the parameter values 34). When, for example, the evaluation model 119 is a machine learning model, the features that the evaluation model 119 uses to estimate the severity 120 of the condition may be unknown. Controlling the adjustable parameters 32 and correlating the adjustable parameters 32 with the estimates provided by the evaluation model 119 may provide some explanation or insight as to how the evaluation model 119 functions.

[0065] A user 22 (via the adjustable parameters 32) may use the modeler 110 to generate any number of models 115 with any combination of parameter values 34. These models 115 in turn may be used to generate images 116 for the evaluation model 119 toprocess. These images 116 mimic images that may be captured during, for example, medical procedures (e.g., via a video capsule endoscopy or the like). Authentic images captured from real patients are expensive and difficult to obtain, and finding a patient that exhibits a particular combination of symptoms may be difficult if not impossible. Thus, the virtual synthetic organ modeler 110 allows valuable data to be generated virtually without the expenses and difficulties of obtaining a similar quantity of data from actual patients.

[0066] Referring now to FIG. 4, an image 140A includes the model 115 as a distal small intestine that includes sculpted folds (e.g., shallow folds) in an elongated tube with finger-like villi. The villi may be scaled or rescaled to a variety of different heights (e.g., mean heights) prior to frame rendering. The images 140B-D show the villi scaled to different heights (with image 140B having the longest villi, image 140C having an intermediate villi, and image 140D having the shortest villi). As shown in image 140E, the evaluation model 119 may show a correlation with a change in an adjustable parameter 32 (which is the villous height in this example). That is, in this example, a change in the adjustable parameter 32 of villous height corresponds to a correlation between the predicted severity and villous height (e.g., a negative correlation such that a taller villi distribution results in a lesser injury severity value), which can be indicative of villous height being a meaningful factor for characterizing injury severity with respect to villous atrophy. Accordingly, since villous atrophy may be considered a hallmark of celiac disease, the evaluation model 119 suggests that villous height may be a key performance indicator of a patient’s severity of celiac’s disease.

[0067] Generally speaking, the parameterization of the system 100 can isolate specific anatomical attributes; allowing the modeler 115 to evaluate a specific anatomic attribute’s impact on injury severity as inferred by a trained model. This parameterized evaluation approach can enable feed-forward capability that can benefit several aspects of medical device design, implementation, and / or evaluation. For example, by understanding the impact of specific anatomical attributes for organ / tissue health, design teams can be forward thinking to curate data that focuses on influential anatomical attributes for organ / tissue health. For instance, by identifying that villous height is aninfluential anatomical attribute for the gastrointestinal tract, data collection techniques, such as image capture, can focus or tune the capture resolution for determining villous height; potentially leading to improved image data being fed to the model to predict tissue / organ injury. In a similar vein, it may allow healthcare providers to recommend that a patient uses one VCE camera over another to improve the accuracy of a model to predict that patient’s organ / tissue health state. Additionally or alternatively, it could lead someone to design an improved instrument to extract a particular parameter. As an example, someone could design a specialty VCE camera that has a particular image capture resolution, image capture field of view, or other capability to focus on capturing an influential parameter (e.g., villous height) as the camera traverses some internal portion of the gastrointestinal tract.

[0068] For example, in one implementation, the system 100 is configured to generate technical design specifications for a new or improved medical imaging device, such as a specialized VCE camera. In this implementation, the virtual synthetic organ modeler 110 performs a systematic parameter sweep, generating a plurality of virtual 3D models 115 that represent a specific, influential anatomical attribute (e.g., villous height) across its full clinical range. For each model, the system further simulates the image acquisition process using a range of virtual camera parameters, such as sensor resolution, lens focal length, and illumination wavelength. The resulting matrix of synthetic images 116 is processed by the evaluation model 119. The system analyzes the correlation between the evaluation model’s output severity 120 and the ground truth parameters to identify a set of virtual camera parameters that yields the highest accuracy or sensitivity for that specific anatomical attribute. The system 100 generates, as a machine-readable output, a technical specification data structure that defines the optimal hardware parameters (e.g., a minimum sensor resolution of 1024x1024 pixels, a required focal length, a specific illumination wavelength of 470 nm) for an imaging device specifically designed to measure that attribute. This output provides a concrete technical blueprint for manufacturing an improved medical device.

[0069] In another implementation, the system 100 functions as a clinical procedure optimization engine that automatically selects an optimal imaging device for a specificpatient. In this aspect, the system 100 first receives input data identifying a suspected condition for a patient (e.g., celiac disease). The system accesses a device library stored in storage resources 14, where each entry in the library corresponds to a commercially available VCE camera and includes its known technical specifications. The system 100 generates a virtual 3D model 115 tailored to the suspected condition (e.g., modeling a range of villous atrophy). It simulates the image capture process for each available VCE camera from the device library, generating a distinct set of synthetic images 116 for each camera. By processing these sets of images with the evaluation model 119, the system calculates a predicted diagnostic confidence score for each available camera. The system 100 generates a device selection directive, which is a data object that identifies the VCE camera model that yields the highest confidence score. This directive may be transmitted via the network 24 to an electronic health record system or a clinical workflow management system to automatically guide the selection of equipment for the patient’s endoscopic procedure, thereby improving the technical process of medical diagnosis.

[0070] Referring now to FIG. 5, here, an image 150A includes another model 115 of the proximal small bowel / duodenum, characterized by deeper folds of the intestinal walls and tongue-like, flattened villous morphology. The adjustable parameters 32 for this particular model 115 may include villous morphology (e.g., selected from a collection of villi mimicking the morphology of healthy tissue and various levels of unhealthy tissue) as shown in image 150B, crypt hyperplasia (e.g., modeled via the depth of a surface texture which leads to the appearance of cracking), and / or blood flow (e.g., modeled via the color saturation and / or subsurface scattering of walls and villi). An image 150C shows a model 115 with adjustable parameters 32 that mimic healthy tissue (e.g., richer blood flow color and distinctive villi shape / morphology across the surface distribution) while an image 150D shows a model 115 with adjustable parameters 32 that mimic unhealthy tissue (e.g., pale blood flow color and blunted / less distinctive villi shape / morphology across the surface distribution).

[0071] FIG. 6 is a flowchart of an exemplary arrangement of operations for a method 160 for generating synthetic data to test, validate, or explain a computer- implemented evaluation model. The method 160, executed for example by the virtual synthetic organmodeler 110 operating on the processing system 10, begins at operation 162, which includes receiving a parameter value 34 for an adjustable parameter 32. The adjustable parameter 32 is associated with a specific, quantifiable attribute of an organ that correlates with a condition, such as a disease state. For example, the adjustable parameter 32 may be “villous height” and the parameter value 34 may be a specific height in millimeters, which is an attribute associated with celiac disease severity. This may be received as part of a virtual model request 30 originating from a user device 16. The modeler controller 112 processes the request 30 and its associated parameters 32, 34.

[0072] At operation 164, the method 160 includes generating, using the parameter value 34, a virtual three-dimensional (3D) model 115 of the organ. The modeler controller 112 translates the received parameter value 34 into a model generate command 113, which instructs a 3D modeler 114 to construct the virtual 3D model 115 with the specified attribute. This process allows for the deterministic creation of an anatomically precise model representing a specific health state. At operation 166, the method 160 includes generating, from a virtual point of view within the virtual 3D model 115, an image 116 that captures at least a portion of the virtual 3D model 115. This step effectively simulates the data acquisition process of an endoscopic camera inside the organ. Subsequently, at operation 168, the method 160 includes estimating, using an evaluation model 119, the severity 120 of the condition of the organ based on the generated image 116. The evaluation model 119, which may be a pre-trained machine learning model, processes the synthetic image 116 to produce a score or classification representing the severity 120.

[0073] This arrangement, where a virtual 3D model 115 is parametrically generated to produce a synthetic image 116 for analysis by an evaluation model 119, provides numerous technical advantages and represents a significant improvement over conventional methods for developing and validating medical diagnostic models. The method 160 directly addresses the core technical problem of data scarcity in the medical field, where obtaining large, diverse, and precisely annotated datasets from real patients is prohibitively expensive, time-consuming, and fraught with regulatory and ethical challenges. This process contrasts sharply with prior art reliance on limited, imbalanced,and often artifact-laden patient data, which leads to evaluation models with poor generalization, unknown biases, and an inability to accurately diagnose rare or extreme conditions. The method 160, by using the virtual synthetic organ modeler 110 to systematically control adjustable parameters 32, enables the generation of vast, perfectly- annotated datasets that can cover the entire pathophysiological spectrum of a disease, thereby improving the robustness and accuracy of the evaluation model 119.

[0074] Moreover, as compared with other techniques, the method 160 offers technical benefits that directly improve the functioning of the computer system itself, particularly in the context of machine learning development and medical device design. By isolating a single adjustable parameter 32 (e.g., villous height) and observing its effect on the output severity 120, the system provides a technical means for model explainability, allowing developers to understand and debug the complex inner workings of the evaluation model 119. Furthermore, this method transforms the computer from a passive analysis tool into an active design tool. As described previously, the system may generate a technical specification data structure for an improved imaging device or a device selection directive for a clinical workflow. This represents an improvement in the use of computing resources, enabling the automated design and optimization of other technologies. This systematic, computer-controlled generation of synthetic data is more efficient in terms of time and computational resources than the logistics and expense of large-scale clinical trials, ultimately leading to faster development cycles for more reliable and accurate computer-implemented diagnostic tools.

[0075] Implementations herein further include a system and method for validating an evaluation model that characterizes a physical phantom. Generally speaking, a physical phantom refers to a real object designed to simulate a part of the human body. In this sense, a physical phantom can be designed to simulate a health state of the portion of the human body that it represents such that the evaluation model can characterize the health state (e.g., an injury severity 120) of the represented portion of the human body via the physical phantom.

[0076] In some examples, the system includes a reservoir containing a solution and a physical phantom of an organ immersed within the solution of the reservoir andrepresenting a condition of the organ. The system also includes an ingestible data capture device, at least one actuator that controls movement of the ingestible data capture device through a hollow portion of the physical phantom, and an evaluation model. The system also includes data processing hardware and memory hardware in communication with the data processing hardware. The memory hardware stores instructions that when executed on the data processing hardware cause the data processing hardware to perform operations. The operations include instructing the ingestible data capture device to traverse through the hollow portion of the physical phantom according to a predetermined path and receiving sensor data captured as the ingestible data capture device traverses through the hollow portion of the physical phantom. The operations also include estimating a severity 120 of the condition based on the received sensor data using the evaluation model and validating at least one of the physical phantom, the ingestible data capture device, or the evaluation model based on the severity 120 of the condition.

[0077] In other examples, implementations herein include a system that includes a reservoir containing a solution and a physical phantom of an organ that has a hollow portion that has a proximal end and a distal end. The physical phantom is immersed within the solution of the reservoir and represents a condition of the organ. The system also includes an ingestible data capture device operable to capture sensor data, at least one actuator that controls movement of the ingestible data capture device through the hollow portion of the physical phantom, and an evaluation model. Notably, the system may optionally include data processing hardware and memory hardware. That is, in some configurations, the system does not include data processing hardware and memory hardware while other configurations do include the data processing hardware and the memory hardware (e.g., to instruct a data capture device to travel through the physical phantom).

[0078] Referring to FIG. 7, in some implementations, an example system 300 includes the processing system 10. The processing system 10 may be a single computer, multiple computers, or a distributed system (e.g., a cloud environment) having fixed or scalable / elastic computing resources 12 (e.g., data processing hardware) and / or storage resources 14 (e.g., memory hardware). The processing system 10 may be incommunication with the user device 16 associated with a respective user 22 via, for example, the network 24. The processing system 10, the user device 16, and the user 22 may be the same or different as the example system 100 (FIG. 1). The user device 16 may correspond to any computing device such as a desktop workstation, a laptop workstation, or a mobile device (i.e., a smart phone). The user device 16 includes computing resources 18 (e.g., data processing hardware) and / or storage resources 20 (e.g., memory hardware).

[0079] The processing system 10 executes a test sample generator 310 (also referred to herein as the generator 310), a test environment 400, and a validation module 340. The generator 310 may include a 3D printer 320 and a path model 330. The 3D printer 320 is configured to generate a physical phantom 325 of an organ using a 3D printing model 322. For example, the 3D printer may generate a respective physical phantom 325 of a small intestine. In some examples, the 3D printer 320 generates one or more physical phantoms 325. For instance, the 3D printer may generate physical phantoms 325 of multiple different organs or generate multiple physical phantoms 325 of the same organ representing different conditions of the organ. The different conditions may include a healthy representation of the organ and a diseased representation of the organ (e.g., where the diseased representation includes anatomical attributes that characterize the organ in a diseased state). The generator 310, in some implementations, receives a generate physical phantom request 40 requesting that the generator 310 generate the physical phantom 325 of an organ or other collection of tissues. The request may originate from the user 22 via the user device 16 (e.g., transferred via the network 24). In some examples, the user 22 interacts directly with the processing system 10 to provide the generate physical phantom request 40. Optionally, the generate physical phantom request 40 defines or selects the organ or tissue to generate.

[0080] The generate physical phantom request 40 may include one or more respective parameter values 44 for one or more adjustable printing parameters 42. Each adjustable printing parameter 42 controls or adjusts an aspect of the 3D printing model 322. The adjustable printing parameters 42 may include parameters that control dimensions and / or geometric features (i.e., size shape) of each region of the physical phantom 325 andspecify a type of organ (e.g., small intestine, large intestine, etc.) to generate. Moreover, the adjustable printing parameters 42 may include at least one of a height of villi within the physical phantom 325, a density of villi within the physical phantom 325, a color of villi within the physical phantom 325, or a crypt depth severity of the physical phantom 325. In this manner, the adjustable printing parameters 42 may cause the 3D printer 320 to generate the physical phantom 325 that has one or more attributes specified by the adjustable printing parameters 42. As a result, the 3D printer 320 generates the physical phantom 325 such that the physical phantom 325 resembles an organ with a particular condition (e.g., particular health condition). For example, when the physical phantom 325 is a portion of an intestine, the adjustable printing parameter 42 may be for villi height associated with a particular condition or disease of the intestine (e.g., villous atrophy).

[0081] Each adjustable printing parameter 42 has a range of possible parameter values 44 between a minimum parameter value 44 and a maximum parameter value 44. The generate physical phantom request 40 may include any number of parameter values 44 for any number of adjustable printing parameters 42 simultaneously. In some examples, the 3D printing model 322 sets any adjustable printing parameters 42 not specified to a particular parameter value 44 via the request to a default or initial parameter value 44. In other examples, the 3D printing model 322 receives imaging data from a Micro-computed tomography (CT) scanner that scans an actual organ of a test subject and generates the adjustable printing parameters 42 based on the received imaging data. The generator 310 may prompt the user 22 for a missing parameter value 44 when necessary or helpful for generating the physical phantom 325.

[0082] Based on the adjustable printing parameters 42, the parameter values 44, and the generate physical phantom request 40, the 3D printer 320 generates (i.e., prints) the physical phantom 325 using the 3D printing model 322. In some examples, the 3D printer 320 generates the physical phantom 325 by printing each villi structure independently of other villi structures that form the physical phantom 325. As such, the 3D printer 320 may use different adjustable printing parameters 42 for each respective villi structure. The 3D printer 320 may generate the physical phantom 325 as a planarphysical phantom 325, 325A as shown in FIG. 11. In some examples, the 3D printer 320 may form or shape the planar physical phantom 325 according to a corresponding shape of the organ. For instance, FIG. 12 shows the physical phantom 325 as a cylindrical physical phantom 325, 325B. The user 22 or administrator of the system 300 may also manually form or shape the physical phantom 325 to the corresponding shape of the organ.

[0083] In some implementations, the generate physical phantom request 40 may include one of the models 115 generated by the virtual synthetic organ modeler 110 (FIG. 1). That is, in some scenarios, one or more of the virtually generated models 115 may be selected by the user 22 for 3D printing. As such, the generate physical phantom request 40 may include the one of the models 115 and the corresponding adjustable parameters 32 and parameter values 34 associated with the one of the models 115 (FIG. 1). The 3D printer 320 generates a corresponding physical phantom 325 that replicates the one of the models 115 using the 3D printing model 322.

[0084] In some implementations, the 3D printer 320, or the user 22 or an administrator of the system 100, applies one or more post processing steps on the physical phantom 325 output from the 3D printer 320. That is, the villi structure of the physical phantom 325 may not accurately resemble the villi structure of an actual human organ. Put another way, some physical phantoms 325 generated by the 3D printer 320 may not have certain imperfections that real organs have. To that end, the one or more post processing steps may include mechanical and / or chemical post processing steps that distort the physical phantom 325 to more closely resemble an actual human organ. The one or more post processing steps create the imperfections on the physical phantoms 325 generated by the 3D printer 320 to resemble real organs. The 3D printer 320, or the user 22 or the administrator of the testing environment 400, may provide the physical phantom 325 printed by the 3D printer 320 to the testing environment 400. As the 3D printer 320 generates the one or more physical phantoms 325, the user 22 or the administrator may install each physical phantom 325 in the testing environment 400. Installation of each physical phantom 325 may include replacing a currently installed physical phantom 325 with another physical phantom 325.

[0085] The path model 330 of the generator 310 is configured to generate one or more predetermined paths 332, where each path 332 defines a time-series of movement and orientation data for controlling the ingestible data capture device 440 as it traverses the physical phantom 325. Obtaining realistic path data from actual in vivo procedures presents a significant technical challenge, as an ingestible capsule’s trajectory is a complex result of peristalsis and fluid dynamics. To overcome this, the path model 330 is developed using path data samples derived from real-world VCE video recordings. Specifically, a computer vision algorithm, such as a Simultaneous Localization and Mapping (SLAM) or Structure-from-Motion (SfM) algorithm, is applied to the video streams from actual patient procedures. This algorithm identifies and tracks stable feature points on the organ wall across consecutive video frames to simultaneously reconstruct a 3D map of the traversed organ segment and estimate the six-degree-of- freedom (6-DOL) pose (i.e., x, y, z position and pitch, yaw, roll orientation) of the VCE camera for each frame. The output of this process is a set of high-fidelity, empirical path data samples, each representing a real trajectory through an organ.

[0086] These empirical path data samples are then used to train the path model 330, which may be implemented as a generative model, such as a Recurrent Neural Network (RNN) or a Long Short-Term Memory (LSTM) network, capable of learning the temporal dynamics of capsule movement. By training on numerous real-world path samples, the path model 330 learns the underlying statistical patterns of velocity, acceleration, and angular rotation characteristic of different organ regions (e.g., rapid tumbling in the duodenum versus slower, more linear motion in the ileum). When a test is initiated, the path model 330 can then generate a new, realistic predetermined path 332 that is statistically consistent with real in vivo movement but is not a direct copy of any single training sample. In alternative implementations, the predetermined paths 332 may be derived from physics-based simulations that model peristaltic wave propagation and fluid forces or may be defined by mathematical splines based on heuristics provided by clinical experts. Based on the generate physical phantom request 40, which may specify an organ type or condition, the system selects or generates an appropriate predetermined path 332 and sends the path 332 to the controller 460 of the test environment 400.

[0087] In some examples, the systems and methods described herein include a test environment 400. The test environment 400 may be designed to integrate the physical phantom 325 (e.g., an artificial small intestine) with a motion control system to allow the passage of the ingestible data capture device 440 through the physical phantom 325 in a manner that mimics in vivo peristalsis. That is, the test environment 400 may be considered to be a peristalsis simulator. Referring now to FIG. 8, the test environment 400 may include a reservoir (i.e., organ bath) 402 that contains a solution 404. The reservoir 402 may include one or more sidewalls each coupled to a base. Here, the one or more sidewalls and the base cooperatively define a volume that contains the solution 404. The reservoir 402 may also include a proximal end 406 and a distal end 408. The physical phantom 325 generated (i.e., printed) by the 3D printer 320 (FIG. 7) includes a hollow portion 326 that extends from a proximal end 327 to a distal end 328 of the physical phantom 325. The hollow portion 326 may extend the entire length of the physical phantom 325 or a portion thereof. The physical phantom 325 is immersed within the solution 404 of the reservoir 402. For instance, the user 22 or an administrator of the test environment 400 may manually place the physical phantom 325 into the reservoir 402 such that the physical phantom 325 is immersed within the solution 404. In the example shown, the physical phantom 325 has a length of 1-2 feet.

[0088] In some examples, the solution 404 includes a Krebs solution. For example, the solution is Krebs solution with a composition, in millimole (mM), of 118NaCL, 4.8 KC1, 25 NaHCCh, 1.0 NaH2PO4, 1.2 MgSC , 11.1 glucose, and 2.5 CaCh bubbled with carbon gas that includes 95 percent O2 and 5% CO2. Moreover, the solution 404 may include a temperature of 35-37 degrees Celsius. However, the solution 404 may be configured to include any suitable composition or temperature. In some implementations, the proximal end 406 of the reservoir 402 includes an inlet 410 and the distal end 408 of the reservoir 402 includes an outlet 412. In these implementations, one or more pumps (not shown) may be coupled to the inlet 410 and / or the outlet 412 to pump the solution 404 from the inlet 410 to the outlet 412 at a specified rate, for example, at about 5 milliliters (ml) per minute. The one or more pumps may circulate thesolution 404 from the inlet 410 out through the outlet 412 and back through the inlet 410 in a continuous loop.

[0089] The test environment 400 may further include a first tube 416 coupled to the proximal end 327 of the physical phantom 325 and a second tube 418 coupled to the distal end 328 of the physical phantom 325. The first tube 416 includes a connector 420, such as a T-shaped connector, which couples a third tube 422 to a midpoint of the first tube 416. The third tube 422 is coupled to a flow device 424 containing liquid. The liquid may include any liquid that flows through an organ. As such, the flow device 424 is configured to provide a constant flow of the liquid through the third tube 422, the hollow portion 326 of the physical phantom 325, and the second tube 418 into a waste container 426. The flow of the liquid through the hollow portion 326 of the physical phantom 325 simulates fluid flow of liquid through an organ. The second tube 418 may include a pressure sensor (i.e., pressure gauge) 428 that monitors the pressure of the liquid traveling through the second tube 418. The pressure sensor 428 may communicate the pressure of the liquid to the flow device 424 whereby the flow device 424 controls the constant flow of the liquid according to a specified pressure, for example, about 3 hectopascals (hpa).

[0090] The motion control system of the test environment 400 may include a set of actuators (e.g., first actuator 430 and a second actuator 450) that actuate a transport mechanism 432 configured to transport the ingestible data capture device 440 to traverse the physical phantom 325. In some examples, such as FIG. 8, the transport mechanism 432 is a wire capable of being fed by the motion control system through the physical phantom 325 within the test environment 400. In some examples, the transport mechanism 432 is configured with motion control attributes that enable the transport mechanism 432 to move in multiple directions as the transport mechanism 432 traverses the physical phantom 325. For instance, in the case of a wire as the transport mechanism 432, the wire may be an articulating wire capable of adjusting the shape of the wire along an entire length of the wire. The transport mechanism 432 may traverse through the first tube 416, the hollow portion 326 of the physical phantom 325, and the second tube 418. The transport mechanism 432 generally includes a proximal end 434 and a distal end 436.In some implementations, in the case of a wire as the transport mechanism 432, the wire may be an articulating wire capable of adjusting the shape of the wire at the distal end 436 of the wire. An ingestible data capture device 440 may be coupled to the distal end 436 of the transport mechanism 432. More specifically, the distal end 436 of the transport mechanism 432 may include a retention element (i.e., holder) 438 that secures the ingestible data capture device 440 to the distal end 436 of the transport mechanism 432.

[0091] FIGS. 14 and 15 show an example retention element 438 at the distal end 436 of the transport mechanism 432. As shown in FIG. 14, the transport mechanism 432 includes a wire which has one or more articulating wires that attach to the retention element 438. Thus, the retention element 438 moves as the transport mechanism 432 articulates. For instance, one or more of the articulating wires of the transport mechanism 432 may move so as to control the articulation of the retention element 438. As shown in FIG. 15, the retention element 438 may include a first element 470, a second element 472, and a third element 474. The first element 470 interfaces with and secures the ingestible data capture device 440. The first element 470 may include a cylindrical shape and size corresponding to the ingestible data capture device 440. The second element 472 may be a collar that clamps the articulating wires of the transport mechanism 432 to the first element 470. Thus, the second element 472 may be coupled to the first element 470 with a force fit interface. Finally, the third element 474 secures any excess portion of the articulating wires of the transport mechanism 432. Put another way, the third element 474 conceals the articulating wires from the surrounding environment.

[0092] Referring again to FIG. 8, the motion control system (e.g., via the first actuator 430) is operable to control a movement of the ingestible data capture device 440 through the first tube 416, the hollow portion 326 of the physical phantom 325, and the second tube 418. Notably, the first actuator 430 controls the movement of the ingestible data capture device 440 from the proximal end 327 to the distal end 328 of the hollow portion 326 of the physical phantom 325 to simulate how an actual ingestible data capture device 440 travels through an organ (i.e., mimicking in vivo peristalsis).

[0093] The motion control system (e.g., via the first actuator 430) may control a velocity of the ingestible data capture device 440 as the ingestible data capture device 440 traverses through each region of the physical phantom 325. For instance, as shown in FIG. 8, the first actuator 430 may control the ingestible data capture device 440 to move at a first velocity through a first region of the physical phantom 325 and move at a second velocity through a second region of the physical phantom 325. Here, the first velocity may be greater than or less than the second velocity. In some examples, the first actuator 430 includes a stepper motor and two drive wheels to control the movement (i.e., translation) of the ingestible data capture device 440 through the physical phantom 325 by pushing and / or pulling the transport mechanism 432.

[0094] In some examples, the control velocity of the ingestible data capture device 440 is configured to realistically reflect the dynamics of an ingestible data capture device (e.g., a VCE camera) moving through the gastrointestinal tract of a patient (e.g., the patient’s duodenum, jejunum, and ileum). More particularly, the peristaltic dynamics of each region of a gastrointestinal tract may differ, for example, due to the anatomical attributes of the respective region. Meaning that an ingestible data capture device may for instance travel at a different velocity in the duodenum than the ileum. In light of this, the motion control system may be programmed to control the ingestible data capture device at a variable velocity or velocity profile. In some configurations, the programmed velocity or velocity profile may correspond to a measured velocity derived from real- patient data captured by ingestible data capturing devices. For example, when the captured data are image frames, the velocity of a patient’s ingestible data capturing device may be derived using metadata of the images frames such as the frame and time data.

[0095] The motion control system of the test environment 400 may have the capability to control an angle of the ingestible data capture device 440 relative to the retention element 438. Here, controlling the angle of the ingestible data capture device 440 may, similar to the velocity control, enable the data captured in the test environment 400 for the physical phantom 325 to simulate realistic angle parameters of an ingestible data capture device (e.g., a VCE camera) moving through the gastrointestinal tract of apatient (e.g., the patient’s duodenum, jejunum, and ileum). In some configurations, to control the angle of the ingestible data capture device 440 relative to the retention element 438, the motion control system of the test environment 400 includes a second actuator 450. For instance, the second actuator 450 may include a set of servo motors (e.g., two servo motors) that articulate the angle of the ingestible data capture device 440 in a first plane and a second plane such that the second actuator may provide multi-way articulation (e.g., 4-way articulation) of the angle of the ingestible data capture device 440. In some implementations, the second actuator 450 includes a directional controller (e.g., joystick controller) that may be manually operated by the user 22 or administrator of the system 300 or operated by the data processing hardware 12 (FIGS. 7A-7C). That is, the joystick controller may control the angle of the ingestible data capture device 440 by controlling or moving the joystick. In some configurations, the joystick controller is automatically operated such that no manual input from the user 22 is required to control the angle of the ingestible data capture device 440.

[0096] For example, FIG. 9 illustrates an example second actuator 450, 450A whereby a first motor 452 is capable of communicating with the joystick controller to control the angle of the ingestible data capture device 440 in a first plane and a second motor 454 is capable of communicating with the joystick controller to control the angle of the ingestible data capture device 440 in a second plane. Simply put, the motors are configured to enable the movement of the joystick to cause the angle of the ingestible data capture device 440 to correspondingly move. For example, moving the joystick controller left or right signals the first motor 452 to move the angle of the ingestible data capture device 440 left or right (e.g., in a yaw direction relative to the longitudinal axis of the physical phantom 325) while moving the joystick controller up or down signals the second motor 454 to correspondingly move the angle of the ingestible data capture device 440 up or down (e.g., in a pitch direction relative to the longitudinal axis of the physical phantom 325). In particular, movement by the joystick controller may cause the transport mechanism 432 to move about an axis of rotation relative to the direction of travel of the transport mechanism 432 along the longitudinal axis of the physical phantom 325. In some examples, the transport mechanism 432 includes one or more articulating members(e.g., articulating wires) whereby each articulating member is controllable by the joystick controller. For example, when the transport mechanism 432 is a wire, the wire may include four articulating wires to enable four-way articulation. FIG. 9 further illustrates a controller 460 for controlling the first actuator 430 and / or the second actuator 450. The controller 460 may be in communication with the data processing hardware 12 (FIGS. 7A-7C). To that end, the controller 460 may instruct the first actuator 430 and / or the second actuator 450 to traverse the ingestible data capture device 440 according to the predetermined path 332 without any manual input at the first actuator 430 or the second actuator 450.

[0097] Referring now to FIG. 10, in some implementations, the controller 460 includes a graphical user interface (GUI) 462. The GUI 462 may display the current movement (e.g., velocity) and / or angle of the ingestible data capture device 440. Moreover, the controller 460 may include one or more user interface buttons 464. In particular, a first user interface button 464, 464a and a second user interface button 464, 464b may be selectable by the user 22 to start or stop the ingestible data capture device 440 from traversing along the predetermined path 332. Moreover, a third user interface button 464, 464c may include a button to control the movement (i.e., translation) of the ingestible data capture device 440 via the first actuator 430. A fourth user interface button 464, 464d may include a manual joystick for the user to manually control or override the translation or angle of the ingestible data capture device 440.

[0098] Referring again to FIG. 8, the first actuator 430 and the second actuator 450 control the movement (i.e., translation) of the ingestible data capture device 440 and the angle of the ingestible data capture device 440 through the physical phantom 325. The first actuator 430 and the second actuator 450 may control the movement and the angle of the ingestible data capture device 440 at each region of the physical phantom 325. In some examples, the first actuator 430 and the second actuator 450 instruct the ingestible data capture device 440 to traverse through the hollow portion 326 of the physical phantom 325 according to one of the predetermined paths 332. Here, the predetermined path 332 may define the movement (i.e., translation) and the angle of the ingestible image data capture device 440 at each region of the physical phantom 325.

[0099] The ingestible data capture device 440 is configured to capture sensor data 442. In particular, the ingestible data capture device 440 may capture the sensor data 442 as the ingestible data capture device 440 moves through the hollow portion 326 of the physical phantom 325. In some examples, the ingestible data capture device 440 includes an ingestible image capture device or an image endoscopy capsule that captures images as sensor data 442. For instance, the ingestible image capture device may include a PillCam that has one or more cameras 444 that capture the images. In the example shown, the ingestible data capture device 440 includes three cameras 444. In other examples, the ingestible data capture device 440 includes a dissolvable pill that captures sensor data as it moves through the physical phantom 325. The ingestible data capture device 440 may include any suitable ingestible data capture device 440. The ingestible data capture device 440 transmits the captured sensor data 442 to a receiver 446 in wireless communication with the ingestible data capture device 440. The receiver 446 may store the sensor data 442 received from the ingestible data capture device 440 and transmit the sensor data 442 to the validation module 340.

[0100] Referring back to FIG. 7, the validation module 340 includes the evaluation model 119. The evaluation model 119 estimates a severity 722 of the condition of the physical phantom 325 based on the received sensor data 442. In particular, the test sample generator 310 may generate a respective physical phantom 325 representing a particular condition of an organ and the test environment 400 may produce sensor data 442 corresponding to the respective physical phantom 325. Thus, the evaluation model 119 estimates the severity 722 of the condition of the physical phantom 325 based on the received sensor data 442. For instance, the estimated severity corresponds to a value that represents a health state or degree of injury for an organ or tissue represented by the physical phantom. In that respect, the severity 722 estimated by the evaluation model 119 may aim to predict the severity of the particular condition of the organ generated by the 3D printer 320. In this manner, the test sample generator 310 may generate the physical phantom 325 with a known condition whereby the test environment 400 produces sensor data 442 that the evaluation model 119 aims to estimate or predict the known condition of the physical phantom 325. For example, FIG. 13 shows first sensordata 442, 442A and second sensor data 442, 442B. Here, the first sensor data 442A corresponds to an image captured by an ingestible image capture device 440 from an actual physical organ and the second sensor data 442B corresponds to an image captured by the ingestible image capture device 440 from a corresponding physical phantom 325 that aims to replicate the actual physical organ. As such, the first sensor data 442A and the second sensor data 442B shows the similarities between data captured from the actual organ and the physical phantom 325 of the organ. Accordingly, the validation module 340 may validate one or more components associated with producing the severity 722 of the condition. The one or more components may include, the evaluation model 119, the physical phantom 325, or the ingestible data capture device 440.

[0101] The test environment 400, by integrating a physically-realized phantom 325 with a precisely controlled motion system, provides a significant technical improvement over prior methods for testing and validating ingestible medical devices and associated diagnostic algorithms. This system directly overcomes the trade-off between realism and repeatability that plagues conventional approaches. Testing in actual patients, while realistic, is inherently non-repeatable due to stochastic peristalsis and changing patient conditions, making it impossible to perform controlled, comparative experiments. Conversely, simple benchtop tests lack the anatomical, fluidic, and motional realism of the in vivo environment. The disclosed test environment 400 solves this technical problem by providing a platform that is both highly realistic and repeatable. The combination of the 3D printed physical phantom 325, the temperature-controlled solution 404, and the fluid flow simulation creates a high-fidelity anatomical and physiological analog. Notably, the motion control system, guided by a predetermined path 332 from the path model 330, allows the exact same complex, multi-axis trajectory to be executed repeatedly. This enables the rigorous, isolated testing of variables (e.g., comparing the performance of two different ingestible data capture devices 440 on the identical phantom and path, or evaluating an algorithm’s response to different physical phantoms 325 under identical motion profiles), a level of controlled experimentation that was previously difficult if not impossible. This improves the process of medical device design andalgorithm validation, enabling faster, more reliable, and more cost-effective development cycles.

[0102] In some examples, the user 22 may want to validate that the evaluation model 119 produces different severities 722 for two physical phantoms 325 representing different conditions. For instance, the severity 722 produced by the evaluation model 119 for a first physical phantom 325 with a healthy or disease-free condition should not be the same as the severity produced by the evaluation model 119 for a second physical phantom 325 with a disease condition. In contrast, the severities 722 for two physical phantoms 325 representing the same or similar conditions should be relatively the same or similar.

[0103] To that end, the 3D printer 320 may generate a first physical phantom 325 of an organ representing a first condition of the organ and a second physical phantom 325 of the organ representing a second condition of the organ. Here, the first condition may represent a same or different condition as the second condition. Thereafter, the first physical phantom 325 may be installed (e.g., by the user 22) in the test environment 400 such that the ingestible data capture device 440 captures first sensor data 442 as the ingestible data capture device 440 traverses through the physical phantom 325 according to the predetermined path 332. The evaluation model 119 receives the first sensor data 442 and estimates a corresponding severity 722 of the condition of the first physical phantom 325.

[0104] Subsequently, the user 22 may replace the first physical phantom 325 in the test environment 400 with the second physical phantom 325. The ingestible data capture device 440 captures second sensor data 442 as the ingestible data capture device 440 traverses through the second physical phantom 325. The evaluation model 119 receives the second sensor data 442 and estimates a corresponding severity 722 of the condition of the second physical phantom 325. The validation module 340 may compare the corresponding severity 722 of the condition of the first physical phantom 325 with the corresponding severity 722 of the condition of the second physical phantom 325 and generate a validation output 342. The validation output 342 may indicate whether the severities 722 satisfy or fail to satisfy a similarity threshold. The similarity thresholdmay define a similarity or likeness between two or more severities 722. Accordingly, two or more severities 722 that satisfy the similarity threshold may indicate that the severities 722 are relatively the same or similar. On the other hand, two or more severities that fail to satisfy the threshold may indicate that the severities are relatively distinct from one another. Comparison or validation of the severity scores 722 with reference to the similarity threshold may correlate with the condition(s) of the organ. For example, when the similarity threshold is satisfied, the first condition and the second condition may be associated with similar severities.

[0105] In other scenarios, the user 22 may want to validate different versions of the evaluation model 119. The evaluation model 119 may include one or more parameters such that the evaluation model 119 estimates the severity 722 using the one or more parameters. The one or more parameters may be adjustable such that the user 22 adjusts one or more parameters to generate a second evaluation model 119 that has one or more second parameters. Here, the second evaluation model 119 estimates the severity 722 using the one or more second parameters. To that end, the user 22 may validate whether any of the parameter updates causes a change in the severity produced by the evaluation model 119.

[0106] Accordingly, after receiving the first sensor data 442 from the ingestible data capture device 440 traversing through the physical phantom 325, the first evaluation model 119 may estimate a corresponding severity 722 of the condition of the physical phantom 325. Moreover, the second evaluation model 119 may receive second sensor data 442 from the ingestible data capture device 440 traversing through the same physical phantom 325 and estimate a corresponding severity 722 of the condition of the physical phantom 325. Here, the first sensor data 442 and the second sensor data 442 may be the same or substantially similar. The validation module 340 may compare the corresponding severity 722 estimated by the first evaluation model 119 and the corresponding severity 722 estimated by the second evaluation model 119 and generate the validation output 342. The validation output 342 may indicate whether the severities 722 satisfy or fail to satisfy a similarity threshold. Optionally, the validation module 340 may further compare the severity 722 estimated by the evaluation model 119 for thephysical phantom and the severity 120 estimated by the evaluation model 119 for the model 115 when generating the validation output 342.

[0107] In yet other scenarios, the user 22 may want to validate different ingestible data capture devices 440 (FIG. 4). For instance, a first ingestible data capture device 440 may include at least one different optical characteristic than a second ingestible data capture device 440. Thus, the user 22 may validate whether the first or second ingestible data capture device 440 impacts the severity 722 estimated by the evaluation model 119. In particular, the test environment 400 may produce first sensor data 442 captured by the first ingestible data capture device 440 as the first ingestible data capture device 440 traverses through the physical phantom 325 and the evaluation model 119 estimates a corresponding severity 722 based on the first sensor data 442. Thereafter, the test environment 400 may produce second sensor data 442 captured by the second ingestible data capture device 440 as the second ingestible data capture device 440 traverses through the same physical phantom 325 and the evaluation model 119 estimates a corresponding severity 722 based on the second sensor data 442. The validation module 340 may compare the corresponding severity 722 estimated by the evaluation model 119 using the first sensor data 442 and the corresponding severity 722 estimated by the evaluation model 119 using the second sensor data 442 and generate the validation output 342. The validation output 342 may indicate whether the severities 722 satisfy or fail to satisfy a similarity threshold.

[0108] Although the valuation models are described herein as estimating the severity of a subject object (e.g., the model 115 or the physical phantom 325), the “severity” predicted by the valuation models are indicative of a state (e.g., a health state) for the subject object with respect to a theoretical reference object. In other words, the severity can be a value (e.g., numerical value) representing the degree to which a state of the subject object (e.g., an injured / unhealthy organ / tissue or pathophysiological state) deviates from the theoretical reference object (e.g., a healthy organ / tissue). From this perspective, the severity estimation can therefore characterize (e.g., quantify) the injury state of an organ or tissue that may be undergoing a disease condition, an abnormality, or a pathogenic occurrence. This can in turn enable a valuation model by transitiveprinciples to characterize the undergoing disease condition, abnormality, or pathogenic occurrence in some manner.

[0109] Moreover, these synthetic organ modeling techniques can permit a designer / developer to exaggerate a pathophysiological state beyond known patent states to enable health predictive tools, such as the valuation model, to have greater predictive accuracy at the boundaries of a particular medical condition. In other words, developing synthetic data on rare, extreme, and potentially non-feasible pathophysiological states enables a tool that predicts a health state (e.g., a severity for a tissue / organ) to have an acceptable threshold of accuracy across an entire pathophysiological spectrum of a disease or health condition.

[0110] FIG. 16 is a flowchart of an exemplary arrangement of operations for a computer-implemented method 1600 for validating an evaluation model 119 that estimates a severity 722 of a condition of a three-dimensional printed physical phantom 325 of an organ. At operation 1602, the method 1600 includes instructing an ingestible data capture device 440 to traverse through a hollow portion 326 of a physical phantom 325 according to a predetermined path 332. The physical phantom 325 is immersed within a solution 404 of a reservoir 402 and represents a condition of an organ. At operation 1604, the method 1600 includes receiving, from the ingestible data capture device 440, sensor data 442 captured as the ingestible data capture device 440 traverses through the hollow portion 326 of the physical phantom 325. At operation 1606, the method 1600 includes estimating, using an evaluation model 119, a severity 722 of the condition based on the received sensor data 442. At operation 1608, the method 1600 includes validating at least one of the physical phantom 325, the ingestible data capture device 440, or the evaluation model 119 based on the severity 722 of the condition.

[0111] The virtual modeling system 100 and the physical phantom testing system 300 are not merely independent tools but can be synergistically integrated into a comprehensive validation and calibration feedback loop. This integrated system addresses the problem that while virtual models are fast and flexible, they often rely on idealized assumptions about physics (e.g., light interaction, fluid dynamics). Conversely, physical testing is grounded in reality but is slow and expensive. Implementationsdescribed herein uses the physical system 300 to systematically identify and correct the inaccuracies of the virtual system 100, resulting in a physically-calibrated virtual modeler that generates synthetic data with realism and reliability.

[0112] In one implementation of this integrated process, the virtual synthetic organ modeler 110 first generates a virtual 3D model 115 based on a selected set of anatomical parameters 32 (e.g., villous height, crypt depth). A baseline synthetic image 116 is generated from this model 115, and the evaluation model 119 produces a predicted severity score 120. This same virtual 3D model 115 is then used as the direct input to the 3D printer 320 to create a physical phantom 325 with a one-to-one correspondence in anatomical features. This physical phantom 325 is subjected to testing in the test environment 400, where a real ingestible data capture device 440 captures sensor data 442. The same evaluation model 119 processes this real sensor data 442 to produce a physically-grounded severity score 722.

[0113] A validation module 340 then compares the predicted severity score 120 with the physically-grounded severity score 722. Any discrepancy between these scores represents a “reality gap,” that is an error attributable to the virtual modeler’s imperfect simulation of non-anatomical factors, such as the lighting model, the camera physics parameters, or the subsurface scattering and texture rendering. This error may be used to generate a calibration signal that is fed back to the virtual synthetic organ modeler 110. The modeler 110 adjusts its internal simulation parameters (e.g., lighting temperature, sheen, virtual camera lens distortion) to minimize this error. This iterative process is repeated until the predicted virtual severity converges with the measured physical severity, yielding a physically-calibrated virtual modeler capable of generating large- scale synthetic datasets that are verifiably representative of real-world data acquisition physics.

[0114] The implementations described above provide methods and systems for generating synthetic or virtual images 116 of an organ to estimate the severity 120 of a condition of the organ. The 3D modeler 114 creates a virtual 3D model 115 of the organ using adjustable parameters 32 related to the condition of the organ. Images 116 are then generated from within the virtual 3D model 115, and an evaluation model 119 estimatesthe severity 120 of the condition. Advantageously, the system 100 may simulate different viewpoints and movements through the organ, and may include features like machine learning, metadata association, and cataloging of images 116 indexed by metadata 117. Additionally, the system 300 validates the evaluation model 119 using physical phantoms 325 of the organ and an ingestible data capture device 440 in a test environment 400. The systems 100, 300 aim to provide valuable data for training, testing, and validating evaluation models 119 without the need for patient-derived images.

[0115] FIG. 17 is a schematic view of an example computing device 1700 that may be used to implement the systems and methods described in this document. The computing device 1700 is intended to represent various forms of digital computers, such as laptops, desktops, workstations, personal digital assistants, servers, blade servers, mainframes, and other appropriate computers. The components shown here, their connections and relationships, and their functions, are meant to be exemplary only, and are not meant to limit implementations of the inventions described and / or claimed in this document.

[0116] The computing device 1700 includes a processor 1710, memory 1720, a storage device 1730, a high-speed interface / controller 1740 connecting to the memory 1720 and high-speed expansion ports 1750, and a low speed interface / controller 1760 connecting to a low speed bus 1770 and a storage device 1730. Each of the components 1710, 1720, 1730, 1740, 1750, and 1760, are interconnected using various busses, and may be mounted on a common motherboard or in other manners as appropriate. The processor 1710 can process instructions for execution within the computing device 1700, including instructions stored in the memory 1720 or on the storage device 1730 to display graphical information for a graphical user interface (GUI) on an external input / output device, such as display 1780 coupled to high speed interface 1740. In other implementations, multiple processors and / or multiple buses may be used, as appropriate, along with multiple memories and types of memory. Also, multiple computing devices 1700 may be connected, with each device providing portions of the necessary operations (e.g., as a server bank, a group of blade servers, or a multi-processor system).

[0117] The memory 1720 stores information non-transitorily within the computing device 1700. The memory 1720 may be a computer-readable medium, a volatile memory unit(s), or non-volatile memory unit(s). The non-transitory memory 1720 may be physical devices used to store programs (e.g., sequences of instructions) or data (e.g., program state information) on a temporary or permanent basis for use by the computing device 1700. Examples of non-volatile memory include, but are not limited to, flash memory and read-only memory (ROM) / programmable read-only memory (PROM) / erasable programmable read-only memory (EPROM) / electronically erasable programmable read-only memory (EEPROM) (e.g., typically used for firmware, such as boot programs). Examples of volatile memory include, but are not limited to, random access memory (RAM), dynamic random access memory (DRAM), static random access memory (SRAM), phase change memory (PCM) as well as disks or tapes.

[0118] The storage device 1730 is capable of providing mass storage for the computing device 1700. In some implementations, the storage device 1730 is a computer-readable medium. In various different implementations, the storage device 1730 may be a floppy disk device, a hard disk device, an optical disk device, or a tape device, a flash memory or other similar solid state memory device, or an array of devices, including devices in a storage area network or other configurations. In additional implementations, a computer program product is tangibly embodied in an information carrier. The computer program product contains instructions that, when executed, perform one or more methods, such as those described above. The information carrier is a computer- or machine-readable medium, such as the memory 1720, the storage device 1730, or memory on processor 1710.

[0119] The high speed controller 1740 manages bandwidth-intensive operations for the computing device 1700, while the low speed controller 1760 manages lower bandwidth-intensive operations. Such allocation of duties is exemplary only. In some implementations, the high-speed controller 1740 is coupled to the memory 1720, the display 1780 (e.g., through a graphics processor or accelerator), and to the high-speed expansion ports 1750, which may accept various expansion cards (not shown). In some implementations, the low-speed controller 1760 is coupled to the storage device 1730 anda low-speed expansion port 1790. The low-speed expansion port 1790, which may include various communication ports (e.g., USB, Bluetooth, Ethernet, wireless Ethernet), may be coupled to one or more input / output devices, such as a keyboard, a pointing device, a scanner, or a networking device such as a switch or router, e.g., through a network adapter.

[0120] The computing device 1700 may be implemented in a number of different forms, as shown in the figure. For example, it may be implemented as a standard server 1700a or multiple times in a group of such servers 1700a, as a laptop computer 1700b, or as part of a rack server system 1700c.

[0121] Various implementations of the systems and techniques described herein can be realized in digital electronic and / or optical circuitry, integrated circuitry, specially designed ASICs (application specific integrated circuits), computer hardware, firmware, software, and / or combinations thereof. These various implementations can include implementation in one or more computer programs that are executable and / or interpretable on a programmable system including at least one programmable processor, which may be special or general purpose, coupled to receive data and instructions from, and to transmit data and instructions to, a storage system, at least one input device, and at least one output device.

[0122] These computer programs (also known as programs, software, software applications or code) include machine instructions for a programmable processor, and can be implemented in a high-level procedural and / or object-oriented programming language, and / or in assembly / machine language. As used herein, the terms “machine-readable medium” and “computer-readable medium” refer to any computer program product, non- transitory computer readable medium, apparatus and / or device (e.g., magnetic discs, optical disks, memory, Programmable Logic Devices (PLDs)) used to provide machine instructions and / or data to a programmable processor, including a machine-readable medium that receives machine instructions as a machine-readable signal. The term “machine-readable signal” refers to any signal used to provide machine instructions and / or data to a programmable processor.

[0123] The processes and logic flows described in this specification can be performed by one or more programmable processors, also referred to as data processing hardware, executing one or more computer programs to perform functions by operating on input data and generating output. The processes and logic flows can also be performed by special purpose logic circuitry, e.g., an FPGA (field programmable gate array) or an ASIC (application specific integrated circuit). Processors suitable for the execution of a computer program include, by way of example, both general and special purpose microprocessors, and any one or more processors of any kind of digital computer. Generally, a processor will receive instructions and data from a read only memory or a random access memory or both. The essential elements of a computer are a processor for performing instructions and one or more memory devices for storing instructions and data. Generally, a computer will also include, or be operatively coupled to receive data from or transfer data to, or both, one or more mass storage devices for storing data, e.g., magnetic, magneto optical disks, or optical disks. However, a computer need not have such devices. Computer readable media suitable for storing computer program instructions and data include all forms of non-volatile memory, media and memory devices, including by way of example semiconductor memory devices, e.g., EPROM, EEPROM, and flash memory devices; magnetic disks, e.g., internal hard disks or removable disks; magneto optical disks; and CD ROM and DVD-ROM disks. The processor and the memory can be supplemented by, or incorporated in, special purpose logic circuitry.

[0124] To provide for interaction with a user, one or more aspects of the disclosure can be implemented on a computer having a display device, e.g., a CRT (cathode ray tube), LCD (liquid crystal display) monitor, or touch screen for displaying information to the user and optionally a keyboard and a pointing device, e.g., a mouse or a trackball, by which the user can provide input to the computer. Other kinds of devices can be used to provide interaction with a user as well; for example, feedback provided to the user can be any form of sensory feedback, e.g., visual feedback, auditory feedback, or tactile feedback; and input from the user can be received in any form, including acoustic, speech, or tactile input. In addition, a computer can interact with a user by sendingdocuments to and receiving documents from a device that is used by the user; for example, by sending web pages to a web browser on a user’s client device in response to requests received from the web browser.

[0125] A number of implementations have been described. Nevertheless, it will be understood that various modifications may be made without departing from the spirit and scope of the disclosure. Accordingly, other implementations are within the scope of the following claims.

Claims

WHAT IS CLAIMED:

1. A computer-implemented method executed by data processing hardware that causes the data processing hardware to perform operations comprising: receiving a parameter value for an adjustable parameter, the parameter value associated with a severity of a condition of an organ; generating, using the parameter value, a virtual three-dimensional (3D) model of the organ; generating, from a virtual point of view within the virtual 3D model of the organ, an image capturing at least a portion of the virtual 3D model of the organ; and estimating, using an evaluation model, the severity of the condition of the organ based on the image.

2. The method of claim 1 , wherein the organ comprises an intestine.

3. The method of claim 2, wherein the adjustable parameter comprises one of: a height of villi within the intestine; a model of villi within the intestine; a density of villi within the intestine; a color of villi within the intestine; or a crypt depth severity of the intestine.

4. The method of any of claims 1-3, wherein generating the virtual 3D model of the organ comprises transmitting an application programming interface (API) request to a 3D modeling application.

5. The method of any of claims 1-4, wherein the operations further comprise: generating a plurality of images, each image in the plurality of images capturing a respective portion of the virtual 3D model of the organ; and estimating, using the evaluation model, the severity of the condition of the organ based on each image in the plurality of images.

6. The method of claim 5, wherein the operations further comprise generating a video file based on a concatenation of the plurality of images.

7. The method of claims 5 or 6, wherein each image in the plurality of images is generated from a different virtual point of view within the virtual 3D model of the organ.

8. The method of claim 7, wherein the different virtual point of view of each image in the plurality of images simulates movement through the virtual 3D model of the organ.

9. The method of any of claims 1-8, wherein the model comprises a machine learning model.

10. The method of any of claims 1-9, wherein the operations further comprise associating metadata with the image, the metadata comprising the parameter value.

11. The method of claim 10, wherein the operations further comprise storing the image and the associated metadata in a catalog, the catalog indexed by the metadata.

12. The method of any of claims 1-11, wherein generating the virtual 3D model of the organ is based on a second adjustable parameter, the second adjustable parameter comprising one of : a video capsule endoscopy (VCE) camera parameter; a lighting parameter; a type of the organ; or a position parameter representing a position within the 3D model of the organ.

13. The method of any of claims 1-12, wherein the operations further comprise validating the evaluation model based on the estimation of the severity of the condition and the parameter value.

14. The method of any of claims 1-13, wherein generating the virtual 3D model of the organ comprises selecting, based on the parameter value, a villi model from among a plurality of villi models.

15. A system comprising: data processing hardware; and memory hardware in communication with the data processing hardware, the memory hardware storing instructions that when executed on the data processing hardware cause the data processing hardware to perform operations comprising: receiving a parameter value for an adjustable parameter, the parameter value associated with a severity of a condition of an organ; generating, using the parameter value, a virtual three-dimensional (3D) model of the organ; generating, from a virtual point of view within the virtual 3D model of the organ, an image capturing at least a portion of the virtual 3D model of the organ; and estimating, using an evaluation model, the severity of the condition of the organ based on the image.

16. The system of claim 15, wherein the organ comprises an intestine.

17. The system of claim 16, wherein the adjustable parameter comprises one of: a height of villi within the intestine; a model of villi within the intestine; a density of villi within the intestine; a color of villi within the intestine; or a crypt depth severity of the intestine.

18. The system of any of claims 15-17, wherein generating the virtual 3D model of the organ comprises transmitting an application programming interface (API) request to a 3D modeling application.

19. The system of any of claims 15-18, wherein the operations further comprise: generating a plurality of images, each image in the plurality of images capturing a respective portion of the virtual 3D model of the organ; and estimating, using the evaluation model, the severity of the condition of the organ based on each image in the plurality of images.

20. The system of claim 19, wherein the operations further comprise generating a video file based on a concatenation of the plurality of images.

21. The system of claims 19 or 20, wherein each image in the plurality of images is generated from a different virtual point of view within the virtual 3D model of the organ.

22. The system of claim 21, wherein the different virtual point of view of each image in the plurality of images simulates movement through the virtual 3D model of the organ.

23. The system of any of claims 15-22, wherein the model comprises a machine learning model.

24. The system of any of claims 15-23, wherein the operations further comprise associating metadata with the image, the metadata comprising the parameter value.

25. The system of claim 24, wherein the operations further comprise storing the image and the associated metadata in a catalog, the catalog indexed by the metadata.

26. The system of any of claims 15-25, wherein generating the virtual 3D model of the organ is based on a second adjustable parameter, the second adjustable parameter comprising one of : a video capsule endoscopy (VCE) camera parameter; a lighting parameter; a type of the organ; or a position parameter representing a position within the 3D model of the organ.

27. The system of any of claims 15-26, wherein the operations further comprise validating the evaluation model based on the estimation of the severity of the condition and the parameter value.

28. The system of any of claims 15-27, wherein generating the virtual 3D model of the organ comprises selecting, based on the parameter value, a villi model from among a plurality of villi models.

29. A system comprising: a reservoir containing a solution; a physical phantom of an organ comprising a hollow portion that has a proximal end and a distal end, the physical phantom immersed within the solution of the reservoir and representing a condition of the organ; an ingestible data capture device operable to capture sensor data; at least one actuator that controls movement of the ingestible data capture device through the hollow portion of the physical phantom; and an evaluation model.

30. The system of claim 29, further comprising: data processing hardware; andmemory hardware in communication with the data processing hardware, the memory hardware storing instructions that when executed on the data processing hardware cause the data processing hardware to perform operations comprising: instructing the ingestible data capture device to traverse through the hollow portion of the physical phantom according to a predetermined path; receiving, from the ingestible data capture device, sensor data captured as the ingestible data capture device traverses through the hollow portion of the physical phantom; estimating, using the evaluation model, a severity of the condition based on the received sensor data; and validating, based on the severity of the condition, at least one of the physical phantom, the ingestible data capture device, or the evaluation model.

31. The system of claim 30, further comprising: a second physical phantom of the organ comprising a second hollow portion that has a second proximal end and a second distal end, the second physical phantom immersed within the solution of the reservoir and representing a second condition of the organ, the physical phantom different from the second physical phantom, wherein the operations further comprise: instructing the ingestible data capture device to traverse through the second hollow portion of the second physical phantom according to the predetermined path; receiving, from the ingestible data capture device, second sensor data captured as the ingestible data capture device moves through the second hollow portion of the second physical phantom; and estimating, using the evaluation model, a second severity of the second condition based on the received second sensor data, and wherein validating the at least one of the physical phantom, the ingestible data capture device, or the evaluation model comprises comparing the severity of thecondition for the physical phantom with the second severity of the condition for the second physical phantom.

32. The system of claim 31 , wherein the condition of the organ represented by the physical phantom represents a different condition than the second condition of the organ represented by the second physical phantom.

33. The system of any of claims 30-32, further comprising: a second ingestible data capture device operable to capture sensor data, the ingestible data capture device different from the second ingestible data capture device, wherein the operations further comprise: instructing the second ingestible data capture device to traverse through the hollow portion of the physical phantom according to the predetermined path; receiving, from the second ingestible data capture device, second sensor data captured as the second ingestible data capture device moves through the hollow portion of the physical phantom; and estimating, using the evaluation model, a second severity of the condition based on the received second sensor data, and wherein validating the at least one of the physical phantom, the ingestible data capture device, or the evaluation model comprises comparing the severity of the condition for the physical phantom with the second severity of the condition for the physical phantom.

34. The system of claim 33, wherein the ingestible data capture device comprises at least one different optical characteristic than the second ingestible data capture device.

35. The system of any of claims 30-34, further comprising: a second evaluation model different from the evaluation model, wherein the operations further comprise:instructing the ingestible data capture device to traverse through the hollow portion of the physical phantom according to the predetermined path; receiving, from the ingestible data capture device, second sensor data captured as the ingestible data capture device moves through the hollow portion of the physical phantom; and estimating, using the second evaluation model, a second severity of the second condition based on the received second sensor data, and wherein validating the at least one of the physical phantom, the ingestible data capture device, or the evaluation model comprises comparing the severity of the condition for the physical phantom with the second severity of the condition for the physical phantom.

36. The system of claim 35, wherein: the evaluation model comprises one or more parameters, wherein the evaluation model estimates the severity of the condition using the one or more parameters; and the second estimation model comprises one or more second parameters, the one or more parameters different from the one or more second parameters, wherein the second estimation model estimates the severity of the condition using the one or more second parameters.

37. The system of any of claims 30-36, wherein the operations further comprise: instructing the ingestible data capture device to traverse through the hollow portion of the physical phantom according to a second predetermined path, the predetermined path different from the second predetermined path; receiving, from the ingestible data capture device, second sensor data captured as the ingestible data capture device moves through the hollow portion of the physical phantom according to the second predetermined path; and estimating, using the evaluation model, a second severity of the condition based on the received second sensor data,wherein validating the at least one of the physical phantom, the ingestible data capture device, or the estimation model comprises comparing the severity of the condition for the physical phantom with the second severity of the condition for the physical phantom.

38. The system of any of claims 30-37, wherein the operations further comprise: obtaining example path data samples, each respective example path data sample comprising corresponding movement data and corresponding angle data of an example ingestible data capture device traversing through an example organ; and determining, based on the example data samples, one or more predetermined paths traversing through the hollow portion of the physical phantom, each respective predetermined path of the one or more predetermined determined paths defining a corresponding movement and a corresponding angle of the ingestible data capture device traversing the each position of the hollow portion of the physical phantom, the one or more predetermined paths comprising at least the predetermined path.

39. The system of any of claims 30-38, further comprising: a three-dimensional (3D) printer configured to print one or more physical phantoms of the organ; and a 3D printing model comprising a set of adjustable printing parameters, each adjustable printing parameter influencing a respective attribute of a respective physical phantom printed by the 3D printer, wherein the operations further comprise instructing the 3D printer to print one or more physical phantoms of the organ according to the 3D printing model, the one or more physical phantoms comprising at least the physical phantom.

40. The system of any of claims 30-39, further comprising: a wire comprising a proximal end and a distal end, the ingestible data capture device coupled to the distal end of the wire;a first actuator that engages the wire, the first actuator operable to control movement of the ingestible capture device from the proximal end to the distal end of the hollow portion of the physical phantom; and a second actuator operable to control an angle of the ingestible data capture device relative to the distal end of the wire, wherein instructing the ingestible data capture device to traverse through the hollow portion of the physical phantom comprises controlling at least one of the movement or the angle of the ingestible data capture device.

41. The system of any of claims 30-40, wherein: the ingestible data capture device comprises an ingestible image capture device; and the sensor data comprises images.

42. The system of any of claims 30-41, wherein the ingestible data capture device comprises an image endoscopy capsule.

43. A computer-implemented method executed by data processing hardware that causes the data processing hardware to perform operations comprising: instructing an ingestible data capture device to traverse through a hollow portion of a physical phantom according to a predetermined path, the physical phantom immersed within a solution of a reservoir and representing a condition of an organ; receiving, from the ingestible data capture device, sensor data captured as the ingestible data capture device traverses through the hollow portion of the physical phantom; estimating, using an evaluation model, a severity of the condition based on the received sensor data; and validating, based on the severity of the condition, at least one of the physical phantom, the ingestible data capture device, or the evaluation model.

44. The method of claim 43, wherein the operations further comprise: instructing the ingestible data capture device to traverse through a second hollow portion of a second physical phantom according to the predetermined path, the second physical phantom immersed within the solution of the reservoir and representing a second condition of the organ, the physical phantom different from the second physical phantom; receiving, from the ingestible data capture device, second sensor data captured as the ingestible data capture device moves through the second hollow portion of the second physical phantom; and estimating, using the evaluation model, a second severity of the second condition based on the received second sensor data, wherein validating the at least one of the physical phantom, the ingestible data capture device, or the evaluation model comprises comparing the severity of the condition for the physical phantom with the second severity of the condition for the second physical phantom.

45. The method of claim 44, wherein the condition of the organ represented by the physical phantom represents a different condition than the second condition of the organ represented by the second physical phantom.

46. The method of any of claims 43-45, wherein the operations further comprise: instructing a second ingestible data capture device to traverse through the hollow portion of the physical phantom according to the predetermined path, the ingestible data capture device different from the second ingestible data capture device; receiving, from the second ingestible data capture device, second sensor data captured as the second ingestible data capture device moves through the hollow portion of the physical phantom; and estimating, using the evaluation model, a second severity of the condition based on the received second sensor data, wherein validating the at least one of the physical phantom, the ingestible data capture device, or the evaluation model comprises comparing the severity of thecondition for the physical phantom with the second severity of the condition for the physical phantom.

47. The method of claim 46, wherein the ingestible data capture device comprises at least one different optical characteristic than the second ingestible data capture device.

48. The method of any of claims 43-47, wherein the operations further comprise: instructing the ingestible data capture device to traverse through the hollow portion of the physical phantom according to the predetermined path; receiving, from the ingestible data capture device, second sensor data captured as the ingestible data capture device moves through the hollow portion of the physical phantom; and estimating, using a second evaluation model, a second severity of the second condition based on the received second sensor data, wherein validating the at least one of the physical phantom, the ingestible data capture device, or the evaluation model comprises comparing the severity of the condition for the physical phantom with the second severity of the condition for the physical phantom.

49. The method of claim 48, wherein: the evaluation model comprises one or more parameters, wherein the evaluation model estimates the severity of the condition using the one or more parameters; and the second estimation model comprises one or more second parameters, the one or more parameters different from the one or more second parameters, wherein the second estimation model estimates the severity of the condition using the one or more second parameters.

50. The method of any of claims 43-49, wherein the operations further comprise:instructing the ingestible data capture device to traverse through the hollow portion of the physical phantom according to a second predetermined path, the predetermined path different from the second predetermined path; receiving, from the ingestible data capture device, second sensor data captured as the ingestible data capture device moves through the hollow portion of the physical phantom according to the second predetermined path; and estimating, using the evaluation model, a second severity of the condition based on the received second sensor data, wherein validating the at least one of the physical phantom, the ingestible data capture device, or the estimation model comprises comparing the severity of the condition for the physical phantom with the second severity of the condition for the physical phantom.

51. The method of any of claims 43-50, wherein the operations further comprise: obtaining example path data samples, each respective example path data sample comprising corresponding movement data and corresponding angle data of an example ingestible data capture device traversing through an example organ; and determining, based on the example path data samples, one or more predetermined paths traversing through the hollow portion of the physical phantom, each respective predetermined path of the one or more predetermined determined paths defining a corresponding movement and a corresponding angle of the ingestible data capture device traversing the each position of the hollow portion of the physical phantom, the one or more predetermined paths comprising at least the predetermined path.

52. The method of any of claims 43-51 , wherein the operations further comprise instructing a three-dimensional (3D) printer comprising a 3D printing model to print one or more physical phantoms of the organ according to the 3D printing model, the one or more physical phantoms comprising at least the physical phantom, the 3D printing model comprising a set of adjustable printing parameters, each adjustable printing parameterinfluencing a respective attribute of a respective physical phantom printed by the 3D printer.

53. The method of any of claims 43-52, wherein instructing the ingestible data capture device to traverse through the hollow portion of the physical phantom according to the predetermined path comprises instructing at least one actuator to control a movement of the ingestible data capture device and an angle of the ingestible data capture device.

54. The method of any of claims 43-53, wherein: the ingestible data capture device comprises an ingestible image capture device; and the sensor data comprises images.

55. The method of any of claims 43-54, wherein the ingestible data capture device comprises an image endoscopy capsule.

56. A method executed by data processing hardware that causes the data processing hardware to perform operations comprising: receiving a parameter value for an adjustable parameter, the parameter value associated with a severity of a condition of an organ; generating, using the parameter value, a virtual three-dimensional (3D) model of the organ; generating, from a virtual point of view within the virtual 3D model of the organ, an image capturing at least a portion of the virtual 3D model of the organ; estimating, using an evaluation model, a first severity of the condition of the organ based on the image; printing, using the generated virtual 3D model of the organ, a physical phantom comprising a hollow portion;instructing an ingestible data capture device to traverse through the hollow portion of the physical phantom according to a predetermined path, the physical phantom immersed within a solution of a reservoir and representing a condition of an organ; receiving, from the ingestible data capture device, sensor data captured as the ingestible data capture device traverses through the hollow portion of the physical phantom; estimating, using an evaluation model, a second severity of the condition based on the received sensor data; and comparing the first severity and second severity.

Images