1. A computer-implemented method, the method comprising:

applying a training image of the environment divided into regions to a neural network and performing classification to label test images based on a closest one of the regions;

extracting features from the retrieved training images and pose information of the test image that match the closest region;

performing a bundle adjustment on the extracted features by triangularizing map points of the closest region to generate a reprojection error, and minimizing the reprojection error to determine an optimal pose of the test image; and

for the optimal pose, providing an output indicative of a location or a location probability within the environment of the test image at the optimal pose.

2. The computer-implemented method of claim 1, wherein applying the training image comprises: training images associated with gestures in a region of the environment are received as historical or simulation data, and the received training images are provided to a neural network.

3. The computer-implemented method of claim 2, wherein the neural network is a deep learning neural network that learns regions associated with the poses and determines the closest region for the test image.

4. The computer-implemented method of claim 1, wherein the bundle adjustment comprises: the 3D points associated with the measured pose and the triangulated map points are re-projected into 2D image space to generate a result, and the result is compared to the registered 2D observations to determine the re-projection error.

5. The computer-implemented method of claim 4, wherein the pose of the test image is determined to be the optimal pose for reprojection errors below or equal to a threshold.

6. The computer-implemented method of claim 4, wherein the pose of the test image is determined to be incorrect and the calculation of the pose of the test image is determined to be correct for reprojection errors above a threshold.

7. The computer-implemented method of claim 1, wherein minimizing the reprojection error comprises adjusting a pose of the test image to minimize the reprojection error.

8. A non-transitory computer-readable medium having a storage device storing instructions for execution by a processor, the instructions comprising:

applying a training image of the environment divided into regions to a neural network and performing classification to label test images based on a closest one of the regions;

extracting features from the retrieved training images and pose information of the test image that match the closest region;

performing a bundle adjustment on the extracted features by triangularizing map points of the closest region to generate a reprojection error, and minimizing the reprojection error to determine an optimal pose of the test image; and

for the optimal pose, providing an output indicative of a location or a location probability within the environment of the test image at the optimal pose.

9. The non-transitory computer-readable medium of claim 8, wherein applying the training image comprises: training images associated with gestures in a region of the environment are received as historical or simulation data, and the received training images are provided to a neural network.

10. The non-transitory computer-readable medium of claim 9, wherein the neural network is a deep learning neural network that learns regions associated with the pose and determines a closest region for the test image.

11. The non-transitory computer-readable medium of claim 8, wherein the bundle adjustment comprises: the 3D points associated with the measured pose and the triangulated map points are re-projected into 2D image space to generate a result, and the result is compared to the registered 2D observations to determine the re-projection error.

12. The non-transitory computer-readable medium of claim 11, wherein the pose of the test image is identified as the optimal pose for a reprojection error that is less than or equal to a threshold.

13. The non-transitory computer-readable medium of claim 11, wherein the pose of the test image is determined to be incorrect and the calculation of the pose of the test image is determined to be correct for reprojection errors above a threshold.

14. The non-transitory computer-readable medium of claim 8, wherein minimizing the reprojection error comprises adjusting a pose of the test image to minimize the reprojection error.

15. A computer-implemented system for locating and tracking a viewing device in an environment to identify a target, the computer-implemented system configured to:

applying a training image of the environment associated with the viewing device divided into regions to a neural network and performing classification to label test images generated by the viewing device based on a closest region in the region of the environment associated with the viewing device;

extracting features from the retrieved training images and pose information of the test image that match the closest region;

performing a bundle adjustment on the extracted features by triangularizing map points of the closest region to generate a reprojection error, and minimizing the reprojection error to determine an optimal pose of the test image; and

for the optimal pose, providing an output indicative of a location or a location probability within the environment of the test image generated by the viewing device at the optimal pose.

16. The computer-implemented system of claim 15, wherein the environment comprises a gastrointestinal tract, or bronchial tracts of one or more lungs.

17. The computer-implemented system of claim 15, wherein the viewing device is configured to provide a location of one or more targets including at least one of polyps, lesions, and cancerous tissue.

18. The computer-implemented system of claim 15, wherein the viewing device comprises one or more sensors configured to receive the test image associated with the environment, and the test image is a visual image.

19. The computer-implemented system of claim 15, wherein the viewing device is an endoscope or a bronchoscope.

20. The computer-implemented system of claim 15, wherein the environment is a piping system, a subterranean environment, or an industrial facility.

Background

The related art endoscope systems can provide a minimally invasive approach to examining internal body structures. More specifically, related art Minimally Invasive Surgical (MIS) approaches may provide medical practitioners with tools to examine internal body structures and may be used for accurate therapeutic intervention.

For example, a viewing device (scope), such as an endoscope or bronchoscope, may be placed in the patient's environment, such as the bowel or lungs, to examine its structure. Means such as sensors or cameras on the viewing device may sense information, such as images, video, etc. of the environment, and provide the information to the user. A medical professional, such as a surgeon, may analyze the video. Based on the analysis, the surgeon may provide a recommendation or perform an action.

Various related art Gastrointestinal (GI) tract observation device solutions have been developed using related art robot and sensor technologies. For such related art GI tract solutions, accurate localization and tracking may enable medical practitioners to localize and track the progress of various pathological findings, such as polyps, cancerous tissues, lesions, and the like. Such related art endoscopic systems may meet the need for accurate therapeutic intervention, and therefore must be able to be accurately located and tracked in a given Gastrointestinal (GI) tract and/or bronchial tract.

Related art approaches to tracking the GI tract may include image similarity comparisons, such as comparing image similarity using related art image descriptors, also referred to as image classification. Furthermore, correlation techniques may use geometry-based pose regression, such as correlation techniques' geometry techniques, e.g., SLAM or recovering shapes from shadows for image-to-model registration, also known as geometric optimization. The correlation technique may also use images based on deep learning to perform pose regression.

The related art deep learning schemes have various problems and disadvantages unique to tracking applications such as colonoscopy or bronchoscopy, such as small annotated training data sets and lack of recognizable texture, unlike other indoor or outdoor environments where deep learning has been used in the related art. For example, there are no angular points to define texture, while the nature of body tissue is that it has blood flow, smooth curves and tube structures, without angular points. Thus, there are volume surface angle-like, and mixtures of solids and liquids.

For example, but not by way of limitation, related art deep learning and regression schemes suffer from having data set deficiencies as explained above and lack of corners and textures; in these respects, the surgical situation is different and distinguishable from related technical solutions used in other environments (e.g., autopilot, etc.). For example, there are many tubular structures without corners due to the characteristic physiological properties of the GI tract in the lung.

In addition, because the related art schemes of deep learning and regression attempt to find the location of the viewing device, additional problems and/or disadvantages may occur. For example, but not by way of limitation, there is another problem associated with outliers (outliers) that are completely outside the environment due to the lack of sufficient quality and quantity of data sets for training. The consequences of these outliers are quite significant in the medical field, where determining that the viewing device is completely outside of an environment such as the lung or GI tract can make it difficult for a medical professional to rely on the information and make the appropriate analysis and processing.

A related approach to localization in the GI tract uses monocular images, utilizing related art computer vision techniques (e.g., SIFT and SURF). However, such related art solutions may have various problems and disadvantages, such as deformation, strength, and various obstacles. For example, related art systems may lack depth perception or may be poorly positioned within the limited field of view provided by related art RGB/monocular images. For example, the viewing device has a small field of view due to the proximity of soft tissue in the environment of the patient's body.

Since 3D depth information is not provided and the only available data is RGB video, the related art depth/stereo based viewing device positioning system cannot be directly adapted to a monocular endoscope.

Furthermore, the large amount of data that needs to be used in order to generalize deep learning based localization and tracking is not satisfied. Such data is difficult to obtain, especially in the medical field due to privacy concerns. Furthermore, the geometry-based approach of the related art is not suitable for GI tract observation device tracking because of the small number of features and the loss of registration. It is also impractical or unhealthy to increase the number of data sets by continuing to forcefully insert a viewing device into the patient.

Thus, practitioners may find it difficult to determine the position of the viewing apparatus in the environment of the human body, such as the endoscope position in the GI tract. This problem becomes more severe in certain tissues such as the lungs, due to the branched physiological nature of the lungs.

Disclosure of Invention

According to an aspect of an example implementation, there is provided a computer-implemented method comprising: applying a training image of the environment divided into regions to a neural network, and performing classification to label test images based on a closest region in the regions; extracting features from the retrieved training images and test images that match the closest region; performing bundle adjustment (bundle adjustment) on the extracted features by triangularizing the map points of the closest region to generate a reprojection error, and minimizing the reprojection error to determine an optimal pose of the test image; and for an optimal pose, providing an output indicative of a location or probability of location within the environment of the test image at the optimal pose.

Example implementations may also include a non-transitory computer-readable medium having a memory and a processor capable of executing instructions for image-based localization in a target tissue that fuse depth learning and geometric constraints for image-based localization.

Drawings

This patent or application file contains at least one drawing executed in color. Copies of this patent or patent application publication with color drawing(s) will be provided by the office upon request and payment of the necessary fee.

FIG. 1 illustrates various aspects of a framework for training and testing according to an example implementation.

FIG. 2 illustrates an example representation and data generated by a simulator in accordance with an example implementation.

FIG. 3 illustrates a training process according to an example implementation.

Fig. 4 illustrates a training scheme according to an example implementation.

Fig. 5 illustrates a prediction scheme according to an example implementation.

Fig. 6 illustrates bundle adjustment according to an example implementation.

Fig. 7 illustrates results according to example implementations.

FIG. 8 illustrates results according to example implementations.

FIG. 9 illustrates an example process for some example implementations.

FIG. 10 illustrates an example computing environment having an example computer apparatus suitable for use in some example implementations.

FIG. 11 illustrates an example environment suitable for use in some example implementations.

Detailed Description

The following detailed description provides further details of the figures and example implementations of the present application. For clarity, reference numerals and descriptions of redundant elements between figures are omitted. The terminology used throughout the description is provided by way of example only and is not intended to be limiting.

Aspects of the example implementations are directed to combining deep learning approaches with geometric constraints for use in various fields, including but not limited to Minimally Invasive Surgery (MIS) approaches (e.g., endoscopic approaches).

MIS, in contrast to open surgery, narrows the surgical field. Thus, the surgeon receives less information than an open surgical plan. Thus, MIS solutions require the use of elongated tools to perform the procedure in narrow spaces without direct 3D vision. Furthermore, the training data set may be small and limited.

Example implementations are directed to utilizing tissue (e.g., gastrointestinal tract, lung, etc.) to provide image-based localization for MIS techniques by constraining localization within the tissue.

More specifically, example embodiments classify a test image into one of the training images based on similarity. The closest training image and its neighboring images, as well as its pose information, are used to generate the best pose (e.g., position and orientation) of the test image using feature registration and bundle adjustment. By locating the position and orientation of the viewing device, the surgeon can know the position of the viewing device in the body. Although the present example implementation relates to a viewing apparatus, the example implementation is not so limited and other MIS structures, apparatus, systems and/or methods may be substituted therefor without departing from the scope of the present invention. For example, but not by way of limitation, a probe may be used in place of the viewing device.

For example, but not by way of limitation, example implementations relate to hybrid systems that fuse deep learning with traditional geometry-based techniques. Using this fusion scheme, a smaller data set can be used to train the system. Thus, example implementations may optionally provide a solution for localization using monocular RGB images with fewer samples of training data and texture.

Further, example implementations use geometric methods with deep learning techniques, which may provide robustness of the estimated pose. More specifically, during the reprojection error minimization process, poses with large reprojection errors can be directly rejected.

Training images may be obtained and labeled such that at least one image is assigned to each region. The labeled images are used to train a neural network. When the neural network is trained, test images may be provided and classified into regions. Further, a training data set tree and images from the training data set are obtained. Key features are obtained and adjusted to recover the points of interest and minimize any projection errors.

The example implementations described above relate to a hybrid system that fuses deep learning with geometry-based positioning and tracking. More specifically, the deep learning component according to an example implementation provides a high level of region classification that may be used by geometry-based refinements to optimize the pose of a given test image.

In example implementations, applying geometry to perform refinement may assist in constraining the prediction of the deep learning model, and may optionally better perform pose estimation. Furthermore, with the fused deep learning and geometry techniques described herein, a small training data set may be used to obtain accurate results, and problems with related techniques, such as outliers and the like, may be avoided.

The present example implementation provides a simulation dataset that provides ground truth (ground truth). In a training aspect, the images are input to a neural network and provided as output as image markers associated with regions of the environment. More specifically, the environment is divided into regions. The segmentation may be performed automatically, e.g. by dividing the region into equal lengths, or may be performed by expert knowledge in the medical field, such as based on input of a surgeon regarding a suitable segmentation of the region. Thus, each image is labeled for a region, and the images are classified into regions.

After the training phase, test images are input and classified into regions. The test images are fed into the neural network and compared to the training data set to extract corners from the training images and the test images to establish the global position of the map points. In other words, a comparison is made with the training data set, and key features are obtained and determined and identified as corners.

For the training image, 3D points are projected to the 2D image, and an operation of minimizing a distance between the projected 3D points and the 2D image is performed. Thus, the corner points are recovered in a way that minimizes the reprojection error.

Fig. 1-5 illustrate various aspects of example implementations. Fig. 1 illustrates an overall view of an example implementation, including training and derivation.

Example implementations may be divided into two main blocks: PoseNet107 (e.g., prediction) and pose refinement 109. In the prediction phase of 107, the example implementation utilizes PoseNet, which is a deep learning framework (e.g., GoogleNet). The system consists of a prescribed number (e.g., 23) of convolutional layers and one or more fully-connected layers (e.g., 1). At 107, the model learns region level classifications instead of learning the actual pose. During the derivation, PoseNet may classify the closest region that has a degree of match to a given test image.

In the refinement stage of 109, the regions classified by PoseNet in step 107 and the images and pose information retrieved from the training images are applied to determine the closest match. For pose optimization, streams of adjacent poses are adopted to the closest matching training images. The image and its corresponding stream of pose information are used for pose estimation.

More specifically, according to one example implementation, Unity3D may be used to generate image-pose pairs from a phantom (phantom). These training sets from 101 are used to train the PoseNet model 101. For example, but not limited to, gesture regression may be replaced with region classification. Thus, images in adjacent poses are classified as regions, and labeling is performed at 105.

With respect to the training data, at 101, the training images are provided to a deep learning neural network 103. As shown in fig. 2, at 200, the large intestine 201 may be divided into a plurality of regions identified by lines that process an image of the large intestine 201. For example, but not by way of limitation, the first image 203 may represent a first one of the regions, while the second image 205 may represent a second one of the regions.

Fig. 3 illustrates the foregoing example implementation applied to the training phase at 300. As described above, the training image 301 is provided to the deep learning neural network 303 to generate image labels 305 associated with the region classification of the location of the image. This is further indicated as image at 313. The plurality of images 307 are correspondingly used for training at 309 and labeled at 311.

In PoseNet107, the test image 111 is provided to a deep learning neural network 113, and a marker 115 is generated. This is also denoted 401 in fig. 4. More specifically, for the test image, the most similar regions in the training set are predicted using a deep neural network.

In the pose refinement at 109, the training database 117 receives input from the PoseNet 107. This is also denoted 501 in fig. 5. For example, but not by way of limitation, the training database may provide image IDs that are associated with poses and markers. The pose indicates an image condition and the label indicates a classification associated with the pose.

This information is fed to feature extractors which receive output images at 119, 121 and 123 associated with poses n-k 133, n 129 and n + k 125, respectively. For example, but not by way of limitation, regions and adjacent regions are included to avoid the potential risk of misclassification before beamleveling and reprojection errors are minimized.

Thus, at 135, 131 and 127, features are extracted from each of the images 123, 121 and 119, respectively. More specifically, a feature extractor is employed to extract from the stream of images (e.g., SURF). These extracted features will be further used for bundle adjustment and the features from each image are registered based on their attributes.

More specifically, and as shown in FIG. 6, the feature extractor involves using the output image 601 (which are images 119, 121, and 123). The foregoing feature extraction operation is performed on the output image 601 for a plurality of adjacent poses n-k 603, n 605, and n + k 607, which may indicate respective regions. Triangularization of map points may be performed based on the predicted regions, as shown at 609 and 611.

At 139, in the bundle adjustment, local bundle adjustment is performed using features (e.g., 135, 131, and 127) extracted from images 123, 121, and 119 and pose information 133, 129, and 125 for these images to map the pose. Since the pose of the image concerned is the ground truth, mapping a plurality of corner feature points is a multi-image triangularization process.

At 141, and also represented at 503 in fig. 5, the reprojection error, which may be defined in equation (1), may be re-optimized.

P (position) and R (orientation) are the attitude of the observation apparatus, and v_iAre triangulated map points. Ii () re-projects the 3D point into the 2D image space, and O_iIs a registered 2D observation. At 137, key features of the test image 111 may also be fed into the reprojection error minimization of 141.

If the optimized average reprojection level is below or equal to the threshold, then the best global pose is found at 143. Otherwise, the initial pose is assumed to be incorrect and is due to a failure of the PoseNet. Because the output of the PoseNet can be fully measured, example implementations provide a robust way to identify the validity of the output.

In addition, the reprojection error is minimized. More specifically, registration is established between the key features and the test image, which further serves to optimize the pose of the test image by minimizing the reprojection error of the registered key feature points.

The foregoing example implementations may be implemented in various applications. For example, a viewing device may be used in a medical environment to provide information associated with temporal changes related to features. In one example application, the growth of polyps over time can be tracked, and with the ability to determine the exact location of the viewing device and correctly identify the polyps and their size, a medical professional can more accurately track the polyps. As a result, the medical professional may be able to provide a more accurate risk analysis and provide relevant recommendations and guidelines for action in a more accurate manner.

Furthermore, the viewing apparatus may also comprise means or tools for performing actions in the human environment. For example, the viewing device may include a tool that is capable of modifying the target in the environment. In an example embodiment, the tool may be a cutting tool, such as a laser or heat or knife, or other cutting structure as understood by those skilled in the art. The cutting tool may perform an action, such as cutting a polyp in real time if the polyp is larger than a certain size.

Depending on the medical protocol, polyps are typically cut, or only when a medical professional determines that they are too large or harmful to the patient; according to a more conservative approach, the growth of the target in the environment can be tracked. In addition, the viewing device may be used according to example implementations to more accurately perform subsequent screening after action is taken by the apparatus or tool.

Although an example of a polyp is shown herein, the present example implementation is not so limited and it may be replaced with other environments or targets without departing from the scope of the present invention. For example, but not by way of limitation, the environment may be the bronchi of the lungs rather than the GI tract. Similarly, the target may be a lesion or tumor rather than a polyp.

Additionally, the example actual implementation may feed the results into a prediction tool. According to such an example approach, based on demographic information, growth rate of the tissue, and historical data, an analysis may be performed to generate a predictive risk assessment. The predictive risk assessment may be reviewed by a medical professional, and the medical professional may verify or confirm the results of the predictive tool. Confirmation or verification by the medical professional may be fed back into the predictive tool to improve its accuracy. Alternatively, the predictive risk assessment may be entered into the decision support system with or without validation or confirmation by a medical professional.

In this case, the decision support system may provide recommendations to the medical professional in real time or after removal of the viewing device. In the option where recommendations are provided to the medical professional in real time, real-time operations may be performed based on the recommendations of the decision support system, as the viewing device may also carry a cutting tool.

Additionally, although the foregoing example implementations may define an environment as an environment within the human body that does not have well-defined corners, such as the lungs or bowel, example implementations are not so limited, and other environments with similar characteristics may also be within the scope of the present invention.

For example, but not by way of limitation, piping systems such as waste or water lines may be difficult to inspect for damage, wear and shut-off, or to replace, due to the difficulty in accurately determining in which pipe section the inspection tool is located. By employing the present example implementation, wastewater and water supply lines can be more accurately inspected over time, and line maintenance, replacement, etc. can be performed with fewer accuracy issues. Similar approaches may be employed in industrial safety, such as in a factory environment, underwater, underground environment (e.g., a cave), or other similar environment that satisfies the conditions associated with the present example implementations.

FIG. 7 illustrates results associated with an example implementation at 700. At 701, a related art scheme involving only deep learning is shown. More specifically, a scheme employing regression is shown, and it can be seen that outliers outside the ground truth values are significant in both size and number. As explained above, this is due to the related art problem of a small field of view of the camera and the associated risk of misclassification.

At 703, a scheme is shown for employing test image information using only classification. However, according to this scheme, the data is limited to strictly available data from the video.

At 705, a scheme according to an example implementation is shown that includes classification and bundle adjustment. Although there are small amounts of errors, these errors are mainly due to image texture.

FIG. 8 illustrates validation of an example implementation showing the difference in error over time. The X-axis illustrates keyframes that vary with time and the Y-axis illustrates errors. At 801, the position error is shown and at 803, the angle error is shown. The blue line represents the error using the technique of the example implementation, and the red line represents the error calculated using only the classification technique, and it corresponds to 703 as described above and shown in fig. 7.

More specifically, a simulation dataset is generated based on an off-the-shelf model of the male's digestive system. A virtual colonoscope is placed inside the colon and the observations are simulated. Unity3D (https:// Unity. com /) was used to simulate and generate continuous 2D RGB images using a rigid pinhole camera model. The frame rate and size of the simulated in vivo digester (e.g., as shown in fig. 2) was 30 frames per second and 640 x 480. At the same time, the overall posture of the colonoscope was recorded.

As illustrated, and as described above, the red plots are directed to results with only classification (e.g., related art), and the blue plots are directed to results with pose refinement according to example embodiments. It can be seen that in general, the results of having performed pose refinement have better accuracy, both in terms of position difference and angle difference.

More specifically, and as explained above, FIG. 8 illustrates a comparison 800 of position difference errors relative to key frame IDs at 801 and a comparison of angle difference errors for key frame IDs at 803. Table 1 shows an error comparison between the related art (i.e., ContextualNet) and the example implementations described herein.

TABLE 1

Example implementations may be integrated with other sensors or schemes. For example, but not by way of limitation, other sensors may be integrated on the viewing device, such as an inertial measurement unit, a temperature sensor, an acidity sensor, or other sensors associated with sensing a parameter associated with the environment.

Similarly, multiple sensors of a given type of sensor may be employed; the related art solution may not employ such a plurality of sensors; the related art focuses on providing precise location, as opposed to the approaches described herein that use markers, feature extraction, bundle adjustment, and reprojection error minimization.

Because the present example implementation does not require a high degree of sensor or camera precision or additional training data sets, existing devices may be used with the example implementation to achieve more accurate results. Thus, the need to upgrade hardware to obtain a more accurate camera or sensor may be reduced.

In addition, the increased accuracy may also allow different types of cameras and viewing devices to be interchanged, and allow different medical institutions to more easily exchange results and data and engage more and different medical professionals without sacrificing accuracy, and with the ability to properly analyze and suggest recommendations and take action.

FIG. 9 illustrates an example process 900 according to an example implementation. As illustrated herein, the example process 900 may be performed on one or more devices.

At 901, a neural network receives input and labels training images. For example, but not by way of limitation, training images may be generated from simulations, as explained above. Alternatively, historical data associated with one or more patients may be provided. Training data is used in the model and the pose regression is replaced with region classification. For example, but not by way of limitation, images in adjacent poses may be classified as regions.

At 903, feature extraction is performed. More specifically, the images are provided to a training database 117. Based on the key features, a classification determination is provided as to whether the features of the image can be classified as being in a particular pose.

At 905, bundle adjustment is performed. More specifically, as described above, the map points are triangulated using the predicted regions.

At 907, an operation is performed to minimize the reprojection error of the map points on the test image by adjusting the pose. Based on the results of this operation, an optimal pose is determined.

At 909, an output is provided. For example, but not by way of limitation, the output may be an indication of the image or a region or location within a region of the viewing device associated with the image. Thus, the medical professional may be assisted in determining the location of the image in a target tissue, such as the GI tract, lungs, or other tissue.

FIG. 10 illustrates an example computing environment 1000 with an example computer apparatus 1005 suitable for use in some example implementations. Computing device 1005 in computing environment 1000 may include one or more processing units, cores, or processors 1010, memory 1015 (e.g., RAM, ROM, etc.), internal storage 1020 (e.g., magnetic, optical, solid-state, and/or organic storage), and/or I/O interfaces 1025, any of which may be coupled to a communication mechanism or bus 1030 to communicate information, or embedded in computing device 1005.

According to this example embodiment, the processing associated with neural activity may occur on processor 1010, which is a Central Processing Unit (CPU). Alternatively, it may be replaced with another processor without departing from the inventive concept. For example, and not by way of limitation, a Graphics Processing Unit (GPU) and/or a Neural Processing Unit (NPU) may be used in place of or in conjunction with the CPU to perform the processing for the foregoing example implementations.

Computing device 1005 may be communicatively coupled to input/interface 1035 and output device/interface 1040. One or both of input/interface 1035 and output device/interface 1040 may be a wired or wireless interface and may be removable. Input/interface 1035 may include any device, component, sensor, or physical or virtual interface (e.g., buttons, touch screen interfaces, keyboards, pointing/cursor controls, microphones, cameras, braille, motion sensors, optical readers, etc.) that can be used to provide input.

Output device/interface 1040 may include a display, television, monitor, printer, speakers, braille, etc. In some example implementations, input/interface 1035 (e.g., a user interface) and output device/interface 1040 may be embedded in or physically coupled to computing device 1005. In other example implementations, other computing devices may serve or provide the functions of input/interface 1035 and output device/interface 1040 for computing device 1005.

Examples of computing devices 1005 may include, but are not limited to, highly mobile devices (e.g., smart phones, devices in vehicles and other machines, devices carried by humans and animals, etc.), mobile devices (e.g., tablets, notebooks, laptop computers, personal computers, portable televisions, radios, etc.), and devices of non-mobile design (e.g., desktop computers, server devices, other computers, kiosks, televisions, and/or radios, etc., having one or more processors embedded therein).

Computing device 1005 may be communicatively coupled (e.g., via I/O interface 1025) to external storage 1045 and network 1050 for communication with any number of networked components, devices, and systems, including the same computing device or devices or other configurations. Computing device 1005, or any connected computing device, may function as, or be referred to as, a server, a client, a thin server, a general machine, a special-purpose machine, or another name, providing a service of, or being a server, a client, a thin server, a general machine, a special-purpose machine, or another name. For example, but not by way of limitation, network 1050 may include a blockchain network and/or a cloud.

I/O interface 1025 may include, but is not limited to, using any communication or I/O protocol or standard (e.g., ethernet, 802.11x, general system bus, WiMAX, modem, cellular network protocol, etc.) for communicating information to and from at least all of the connected components, devices, and networks in computing environment 1000. The network 1050 may be any network or combination of networks (e.g., the internet, a local area network, a wide area network, a telephone network, a cellular network, a satellite network, etc.).

Computing device 1005 may use and/or communicate using computer-usable or computer-readable media, including transitory media and non-transitory media. Transitory media include transmission media (e.g., metallic cables, optical fibers), signals, carrier waves, and the like. Non-transitory media include magnetic media (e.g., disks and tapes), optical media (e.g., CD ROM, digital video disks, blu-ray disks), solid state media (e.g., RAM, ROM, flash memory, solid state storage), and other non-volatile storage devices or memories.

Computing device 1005 may be used to implement techniques, methods, applications, processes, or computer-executable instructions in some example computing environments. The computer-executable instructions may be retrieved from a transitory medium and stored in and retrieved from a non-transitory medium. The executable instructions may be derived from one or more of any of a variety of programming, scripting, and machine languages (e.g., C, C + +, C #, Java, Visual Basic, Python, Perl, JavaScript, etc.).

Processor 1010 may execute under any Operating System (OS) (not shown) in a native or virtual environment. One or more applications may be deployed, including logic unit 1055, Application Programming Interface (API) unit 1060, input unit 1065, output unit 1070, training unit 1075, feature extraction unit 1080, bundle adjustment unit 1085, and inter-unit communication mechanism 1095, to enable the different units to communicate with each other, with an operating system, and with other applications (not shown).

For example, training unit 1075, feature extraction unit 1080, and beamlaw smoothing unit 1085 may implement one or more of the processes described above with respect to the above-described structures. The described units and elements may vary in design, function, configuration, or implementation and are not limited to the descriptions provided.

In some example implementations, when information or execution instructions are received by API unit 1060, it may be passed to one or more other units (e.g., logic unit 1055, input unit 1065, training unit 1075, feature extraction unit 1080, and beamlaw adjustment unit 1085).

For example, as described above, training unit 1075 may receive information from simulation data, historical data, or one or more sensors and process the information. The output of training unit 1075 is provided to feature extraction unit 1080, feature extraction unit 1080 performing the necessary operations based on the application of the neural network as described above and shown, for example, in fig. 1-5. Additionally, the beamf iotaeld adjustment unit 1085 may perform operations based on the outputs of the training unit 1075 and the feature extraction unit 1080 and minimize the reprojection error to provide an output signal.

In some cases, in some example implementations described above, logic unit 1055 may be configured to control the flow of information between units and direct the services provided by API unit 1060, input unit 1065, training unit 1075, feature extraction unit 1080, and beamline adjustment unit 1085. For example, the flow of one or more processes or implementations may be controlled by the logic unit 1055 alone or in combination with the API unit 1060.

FIG. 11 illustrates an example environment suitable for certain example implementations. Environment 1100 includes devices 1105-1145, and each device is communicatively connected to at least one other device via, for example, network 1150 (e.g., by a wired and/or wireless connection). Some devices may be communicatively connected to one or more storage devices 1130 and 1145.

Examples of the one or more devices 1105-1145 may each be the computing device 1005 depicted in fig. 10. The device 1105 + 1145 may include, but is not limited to, a computer 1105 (e.g., a laptop device), a mobile device 1110 (e.g., a smart phone or tablet), a television 1115, a device associated with the vehicle 1120, a server computer 1125, a computing device 1135 + 1140, storage devices 1130 and 1145, with a monitor and associated webcam as described above.

In some implementations, the device 1105-1120 may be considered a user device associated with a user who may remotely obtain sensory input that is used as input for the example implementations described above. In this example implementation, one or more of these user devices 1105 and 1120 may be associated with one or more sensors capable of sensing information required by the present example implementation as explained above, such as a camera embedded temporarily or permanently within the user's body, remote from the patient care facility.

While the foregoing example implementations are provided to indicate the scope of the invention, they are not intended to be limiting and other methods or implementations may be substituted or added without departing from the scope of the invention. For example, but not by way of limitation, other imaging techniques may be employed in addition to those disclosed herein.

According to one example implementation, an algorithm such as SuperPoint may be used to train image point detection and determination. Moreover, example implementations may employ alternative image classification algorithms, and/or use other neural network structures (e.g., Siamese networks). Additional approaches incorporate expert knowledge into the region-like actions, apply enhancements to both images by using techniques such as forming, lighting, and/or use a single image-to-depth approach.

Example implementations may have various advantages and benefits, although this is not required. For example, but not by way of limitation, example implementations may operate on small data sets. Further, example implementations provide for constraining a location inside a target tissue, such as a colon or lung. Thus, the surgeon may be able to more accurately locate the position of any person's viewing device by using the video. Furthermore, the example embodiments provide much higher accuracy than the related art methods.

While some example implementations have been shown and described, these example implementations are provided to convey the subject matter described herein to those skilled in the art. It should be understood that the subject matter described herein may be implemented in various forms and is not limited to the example implementations described. The subject matter described herein may be practiced without those specifically defined or described, or without other or different elements or subject matter described. It will be appreciated by those skilled in the art that changes may be made to these example implementations without departing from the subject matter described herein as defined by the appended claims and their equivalents.

Aspects of certain non-limiting embodiments of the present disclosure address the features discussed above and/or other features not described above. However, aspects of the non-limiting embodiments need not address the features described above, and aspects of the non-limiting embodiments of the present disclosure may not address the features described above.