Systems, methods, and media for estimating pulmonary embolism data based on medical data

Abstract

Mechanisms for indicating estimated pulmonary embolism data are provided, the mechanisms including: receiving medical data; processing the medical data to provide processed data; evaluating the processed data using a first machine learning algorithm using a hardware processor; generating the estimated pulmonary embolism data based on the evaluating the processed data; and causing the expected outcome to be presented to a user. In some of these mechanisms, the medical data is 3D matrix image data from a computed tomography (CT) scan. In some of these embodiments, processing the medical data includes at least one of decompressing, intensity scaling, and resizing the medical data. In some of these mechanisms, the first machine learning algorithm is a deep convolutional neural network.

Description

SYSTEMS, METHODS, AND MEDIA FOR ESTIMATING PULMONARY EMBOLISM DATA BASED ON MEDICAL DATACross-Reference to Related Application

[0001] This application claims the benefit of United States Provisional Patent Application No. 63 / 654,633, filed May 31, 2024 and United States Provisional Patent Application No. 63 / 548,557, filed February 1, 2024, each of which is hereby incorporated by reference herein in its entirety.Background

[0002] Pulmonary embolisms (PEs) are found inside pulmonary arteries or pulmonary artery vessels of the lungs. PEs are mostly composed of thrombi or blood clots made of fibrin, platelets and other blood cells. In most cases, those clots come from the lower limb veins. PEs can be in the central, lobar, segmental, sub-segmental, and / or distal parts of the pulmonary arteries. Acute PE mortality can be between 10% and 30% depending on the extent and severity of a corresponding PE. PEs can manifest themselves by chest pain, by loss of consciousness, and, in severe cases, by cardiac arrest.

[0003] PE extension and its effect on the right heart cavities reflect the severity of pulmonary embolisms. Pulmonary embolisms' compositions reflect their acuteness or age, which have repercussions on the best treatment solutions to apply.

[0004] Computed tomography (CT) is the main way to detect a PE and evaluate its extent. CTs are often performed with injection of contrast medium and called angio-CT or CTPA (computed tomography pulmonary angiogram). On an angio-CT, PEs manifest themselves by a dark region surrounded by brighter contrast filled blood of the pulmonary artery vessel. Bilateral and central PEs are more severe and are associated with higher mortality than a unilateral anddistal PEs. PE detection often requires extensive reading by experienced radiologists who scroll slice by slice through the 3D images of a CT scan. As such, PE detection can be a slow and expensive process.

[0005] Accordingly, new mechanisms, including systems, methods, and media for estimating pulmonary embolism data based on medical data are desirable.Summary

[0006] In accordance with some embodiments, mechanisms, which can include systems, methods, and media for estimating pulmonary embolism data based on medical data are provided.

[0007] In some embodiments, systems for indicating estimated pulmonary embolism data are provided, the systems comprising: memory; and at least one hardware processor coupled to the memory and collectively configured to at least: receive medical data; process the medical data to provide processed data; evaluate the processed data using a first machine learning algorithm; generate the estimated pulmonary embolism data based on the evaluating the processed data; and cause the expected outcome to be presented to a user. In some of these embodiments, the medical data is 3D matrix image data from a computed tomography (CT) scan. In some of these embodiments, processing the medical data includes at least one of decompressing, intensity scaling, and resizing the medical data. In some of these embodiments, the first machine learning algorithm is a deep convolutional neural network. In some of these embodiments, the at least one hardware processor is further configured to: generate a plurality of pixels and / or voxels for an estimated pulmonary embolism; and cause the plurality of pixels and / or voxels to be presented with CT image data. In some of these embodiments, the at least one hardware processor is further configured to indicate a general location of a pulmonary embolism. In someof these embodiments, the at least one hardware processor is further configured to indicate a severity of a pulmonary embolism.

[0008] In some embodiments, methods for indicating estimated pulmonary embolism data are provided, the methods comprising: receiving medical data; processing the medical data to provide processed data; evaluating the processed data using a first machine learning algorithm using a hardware processor; generating the estimated pulmonary embolism data based on the evaluating the processed data; and causing the expected outcome to be presented to a user. In some of these embodiments, the medical data is 3D matrix image data from a computed tomography (CT) scan. In some of these embodiments, processing the medical data includes at least one of decompressing, intensity scaling, and resizing the medical data. In some of these embodiments, the first machine learning algorithm is a deep convolutional neural network. In some of these embodiments, the method further comprises: generating a plurality of pixels and / or voxels for an estimated pulmonary embolism; and causing the plurality of pixels and / or voxels to be presented with CT image data. In some of these embodiments, the method further comprises indicating a general location of a pulmonary embolism. In some of these embodiments, the method further comprises indicating a severity of a pulmonary embolism.

[0009] In some embodiments, non-transitory computer-readable media containing computer executable instructions that, when executed by a processor, cause the processor to perform a method for indicating estimated pulmonary embolism data are provided, the method comprising: receiving medical data; processing the medical data to provide processed data; evaluating the processed data using a first machine learning algorithm; generating the estimated pulmonary embolism data based on the evaluating the processed data; and causing the expected outcome to be presented to a user. In some of these embodiments, the medical data is 3D matrix image datafrom a computed tomography (CT) scan. In some of these embodiments, processing the medical data includes at least one of decompressing, intensity scaling, and resizing the medical data. In some of these embodiments, the first machine learning algorithm is a deep convolutional neural network. In some of these embodiments, the method further comprises: generating a plurality of pixels and / or voxels for an estimated pulmonary embolism; and causing the plurality of pixels and / or voxels to be presented with CT image data. In some of these embodiments, the method further comprises indicating a general location of a pulmonary embolism. In some of these embodiments, the method further comprises indicating a severity of a pulmonary embolism.Brief Description of the Drawings

[0010] FIG. 1 is an example of a block diagram of a system architecture in accordance with some embodiments.

[0011] FIG. 2 is an example of a block diagram of hardware that can be used to implement certain components of FIG. 1 in accordance with some embodiments.

[0012] FIG. 3 is an example of a high-level flow diagram in accordance with some embodiments.

[0013] FIG. 4 is an example of a flow diagram of a process for training a machine learning algorithm in accordance with some embodiments.

[0014] FIG. 5 is an example of a flow diagram of a process for determining and presenting possible outcome data in accordance with some embodiments.

[0015] FIG. 6 illustrates an example of a user interface showing 2D representations of pulmonary embolism(s) in accordance with some embodiments.

[0016] FIGS. 7 and 8 illustrate another example of a user interface showing 3D representations of pulmonary embolism(s) in accordance with some embodiments.

[0017] FIGS. 9-11 illustrate yet another example of a user interface showing 2D and 3D representations of pulmonary embolism(s) in accordance with some embodiments.Detailed Description

[0018] Mechanisms, which can include systems, methods, and media for estimating pulmonary embolism data based on medical data are provided. In some embodiments, the mechanism described herein use a trained machine learning algorithm to estimate pulmonary embolism data based on medical data. In some of these embodiments, the medial data is computed tomography (CT) data, such as 3D images of CT data. In some of these embodiments, the medial data is scanography data, such as scanography image data. In some of these embodiments, a deep-learning, artificial-neural-network-based artificial intelligence (Al) algorithm is used to estimate a segmentation mask of a pulmonary embolism automatically. In some embodiments, the segmentation mask identifies and segments a PE inside a subject’s pulmonary arteries. In some embodiments, the segmentation mask is used to generate a volumetric representation of the pulmonary embolism which can be overlayed on the imaging study. In some embodiments, such a volumetric representation can be used to generate pixels and / or voxels representing a PE that can be presented with (e.g., on top of 2D and / or 3D images from CT data. In some embodiments, a deep-learning, artificial-neural-network-based artificial intelligence (Al) algorithm is used to estimate the presence, the location, the extent, and / or the composition of a PE.

[0019] As described above, in some embodiments, the pulmonary embolism data can be determined using one or more machine learning algorithms. Any suitable number and type of machine learning algorithms can be used, in some embodiments. For example, in some embodiments, a convolutional neural network (CNN), which can be a deep CNN in some suchembodiments, can be used as a machine learning algorithm that is used in the mechanisms described herein. As another example, in some embodiments, a U-Net convolutional neural network (CNN) with an encoded-decoder structure can be used as a machine learning algorithm that is used in the mechanisms described herein. As still another example, in some embodiments, Swin U-Net Transformers, which have a switching window transformer encoder and a 3D U-Net decoder with convolutional blocks, can be used as a machine learning algorithm that is used in the mechanisms described herein.

[0020] Turning to FIG. 1, an example block diagram 100 of a system architecture in accordance with some embodiments is shown. As illustrated, the system includes medical data source(s) 102 (which can include medical image data source(s) (which can provide medical image data (e.g., CT image data)) 104, and PE data source(s) (which can indicate any suitable PE data) 106. Although two medical data sources are shown in FIG. 1, any suitable number of medical data sources can be provided, and they can be provided in any suitable one or more devices, in some embodiments. Each medical data source can provide any suitable data in any suitable format, in some embodiments. For example, in some embodiments, each medical data source can be implemented using specialty equipment, such as a CT imager, or general-purpose equipment, such as a computer, a database, a server, etc.

[0021] The system can also include one or more machine learning servers 110 that can each implement any suitable one or more machine learning algorithms. The machine learning server(s) can be administered by an administrator using a machine learning admin device 112.

[0022] The system can further include one or more web interface servers 120 that can each provide one or more web interfaces that can be accessed by a user for interacting with thesystem. The web interface server(s) can be administered by an administrator using a web interface admin device 122.

[0023] The system can still further include one or more user devices 116 and 118 that can be used by a user to access interfaces provided by the web interface server(s).

[0024] The system can further include a communication network 122 for connecting the above-described components. The communication network can be any suitable combination of one or more wired and / or wireless networks in some embodiments. For example, in some embodiments, the communication network can include any one or more of the Internet, a mobile data network, a satellite network, a local area network, a wide area network, a telephone network, a cable television network, a WiFi network, a WiMax network, and / or any other suitable communication network.

[0025] In some embodiments, the medical data source(s), the machine learning server(s), the machine learning admin device, the web interface server(s), the web interface admin device, and / or the user device(s) can be implemented in or utilize any suitable computing device(s). For example, in some embodiments, these components can be implemented using any suitable general-purpose computer or special-purpose computer(s). Any such general-purpose computer or special-purpose computer can include any suitable hardware. For example, as illustrated in example hardware 200 of FIG. 2, such hardware can include a hardware processor 202, memory and / or storage 204, an input device controller 206, an input device 208, display / audio drivers 210, display and audio output circuitry 212, communication interface(s) 214, an antenna 216, and a bus 218.

[0026] Hardware processor 202 can include any suitable hardware processor, such as a graphical processing unit (GPU), tensor processing unit (TPU) and quantum processing unit(QPU), a microprocessor, a micro-controller, digital signal processor(s), dedicated logic, and / or any other suitable circuitry for controlling the functioning of a general-purpose computer or a special purpose computer in some embodiments. For example, when used to implement a machine learning server, the hardware processor can be implemented using eight NVIDIA A 10G Tensor Core GPUs and 192 vCPUs.

[0027] Memory and / or storage 204 can be any suitable memory and / or storage for storing programs, data, and / or any other suitable information in some embodiments. For example, memory and / or storage 204 can include: hard disk storage; optical media; Random Access Memory (RAM) (such as dynamic RAM, static RAM, etc.); NAND flash memory; NOR flash memory; any other suitable flash technology; phase change memory technology; and / or other any other suitable volatile and / or non-volatile memory storage technology. For example, when used to implement a machine learning server, the memory can include 192 GiB GPU memory and 768 GiB Random Access Memory (RAM).

[0028] Input device controller 206 can be any suitable circuitry for controlling and receiving input from input device(s) 208, in some embodiments. For example, input device controller 206 can be circuitry for receiving input from an input device 208 such as a touch screen, from one or more buttons (such as on a keyboard), from a voice recognition circuit, from a microphone, from a camera, from an optical sensor, from an accelerometer, from a temperature sensor, from a near field sensor, and / or any other type of input device.

[0029] Display / audio drivers 210 can be any suitable circuitry for controlling and driving output to one or more display / audio output circuitries 212 in some embodiments. For example, display / audio drivers 210 can be circuitry for driving one or more display / audio output circuitries 212, such as an LCD display, a speaker, an LED, or any other type of output device.

[0030] Communication interface(s) 214 can be any suitable circuitry for interfacing with one or more communication networks (such as the communication network shown in FIG. 1). For example, interface(s) 214 can include network interface card circuitry, wireless communication circuitry, and / or any other suitable type of communication network circuitry.

[0031] Antenna 216 can be any suitable one or more antennas for wirelessly communicating with a communication network in some embodiments. In some embodiments, antenna 216 can be omitted when not needed.

[0032] Bus 218 can be any suitable mechanism for communicating between two or more components 202, 204, 206, 210, and 214 in some embodiments.

[0033] Any other suitable components can additionally or alternatively be included in hardware 200 in accordance with some embodiments.

[0034] In some embodiments, two or more of devices 104, 106, 108, 110, 112, 116, 118, 120, and 122 can be combined. For example, devices 104, 110, 116, and 120 can be implemented in a single device, such as a CT imager, in some embodiments.

[0035] As noted above and as illustrated in FIG. 3, in some embodiments, the mechanisms described herein can estimate PE data 306 based on medical data 302, such as computed tomography (CT) scan data, using one or more machine learning algorithms 304. In some embodiments, the PE data can include the presence, the location, the extent, and / or the composition of one or more PEs. In some embodiments, the PE data can be used to generate pixels and / or voxels that can be presented with (e g., overlayed on) 2D and / or 3D CT images.

[0036] Training a machine learning algorithm to be used in the described mechanisms can be performed in any suitable manner, in some embodiments.

[0037] Turning to FIG. 4, an example 400 of a process for training a machine learning algorithm in accordance with some embodiments is shown. In some embodiments, this process can begin by receiving CT scan data (for example) and corresponding PE data (for example) at 402. Any suitable type of CT scan data and corresponding PE data, any suitable format of same, and any suitable amount of same, can be received, in some embodiments.

[0038] For example, in some embodiments, the CT scan data can represent CT scans of chests with contrast injection or non-injected chest scans. As another example, in some embodiments, CT scans can be received in DICOM or NIFTY format. In some embodiments, the CT scan data can be received in a compressed format.

[0039] As another example, in some embodiments, the corresponding PE data can include any suitable data related to a PE corresponding to the CT scan data, such as that a PE occurred, one or more general locations of a PE (e.g., right, left, bilateral, proximal, distal, multiple, saddle, lobar, etc.), one or more specific locations of a PE (e.g., pixel and / or voxel locations of a PE in one or more corresponding 2D and / or 3D CT images), extent / severity data, PE-related- condition data (e.g., dilated right ventricle, pulmonary artery dilation), and / or any other suitable data relating to an indication that a PE occurred.

[0040] In some embodiments, the PE data can indicate that no PE occurred in a given piece of training (e.g., CT image) data. As such, the PE data can indicate that a PE did not occur, that there were no general locations of a PE (e.g., a PE did not occur in right, left, bilateral, proximal, distal, multiple, saddle, lobar, etc.), that there were no specific locations of a PE (e.g., identify that there were no pixel and / or voxel locations of a PE), that there was no extent / severity data, that there was no PE-related-condition data (e.g., no dilated right ventricle, no pulmonary artery dilation), and / or any other suitable data relating to an indication that a PE did not occur.

[0041] In some embodiments, the CT scan data and the corresponding PE data can be received from any suitable source in any suitable manner. For example, in some embodiments, the CT scan data can be received as 3D image data in a 3D matrix format that is fed into an image dataset loader, and the PE data can be received from a database of same the correlates the PE data to the CT scan data.

[0042] Next, as shown in FIG. 4, in some embodiments, the CT scan data can be processed at 404 prior to being used for training. Any suitable processing can be performed in some embodiments. For example, in some embodiments, the CT scan data can be processed to perform intensity scaling, to perform image resizing (e.g., to a 96*96*96-pixel format), to perform random rotation (e.g., 90°), to perform spacing (e.g., by adding pixels with zero intensity), to add random gaussian noise, to perform random affine transformations, to perform random cropping, and / or to perform any other suitable functions. While these functions may be random, these functions may similarly be pseudorandom, or non-random, in some embodiments.

[0043] Then, as shown in FIG. 4, the machine learning algorithm can be trained at 406. In some embodiments, any suitable portion of the training dataset can be used to train the machine learning algorithm (e.g., such as 80%) and another portion of the training dataset can be used to test or validate the training of the machine learning algorithm (e.g., such as 20%).

[0044] Training of the machine learning algorithm can be performed in any suitable manner, in some embodiments. In some embodiments, an Adaptive Moment Estimation (Adam) optimizer can be used while training the machine learning algorithm. In some embodiments, a cross-entropy loss function can be used in the machine learning algorithm. In some embodiments, during training, the machine learning algorithm can implement input data shuffling between epochs.

[0045] Any suitable parameters can be used in the machine learning algorithm, in some embodiments. For example, in some embodiments, the machine learning algorithm can use any suitable number of epochs, such as 300 epochs. As another example, in some embodiments, the machine learning algorithm can use any suitable size of batches, such as batches of four. As yet another example, in some embodiments, the machine learning algorithm can implement any suitable number of parallel workers, such as 20 parallel workers. As still another example, in some embodiments, the machine learning algorithm can implement any suitable dropout rate, such as a 10% dropout rate. As yet another example, in some embodiments, the machine learning algorithm can implement any suitable number of channels, such as one channel. As still another example, in some embodiments, the machine learning algorithm can implement any suitable number of output classes, such as two output classes (e.g., whether the patient is likely to have had a PE or not). As yet another example, in some embodiments, the machine learning algorithm can implement any suitable learning rate, such as a learning rate of 0.0001.

[0046] Next, at 408, the process can validate the training of the machine learning algorithm. More particularly, in some embodiments, CT image data that was reserved for testing / validating the trained machine learning algorithm (e.g., 20% of the full training data) can be submitted to the machine learning algorithm and estimated PE data generated by the machine learning algorithm can be compared to PE data already established as corresponding to the CT image data to determine if the trained machine learning algorithm is properly estimating PE data. In some embodiments, any suitable parameters can be used to determine whether the trained machine learning algorithm is properly estimating PE data, such as DICE coefficient (a pixel-wise agreement between a estimate pixels and / or voxels of a PE its corresponding ground truth, accuracy, precision, recall, area under the curve (AUC), and / or any other suitable parameter.

[0047] In some embodiments, as shown in FIG. 4, after training, model weights of the machine learning algorithm can be checked for compatibility with a GPU and / or a CPU to be used in production at 410. Model weights of the machine learning algorithm can be checked for compatibility with a GPU and / or a CPU to be used in production in any suitable manner, in some embodiments.

[0048] In some embodiment, as also shown in FIG. 4, after training, the performance of the machine learning algorithm can be checked with fixed weights on the entire dataset to confirm that the algorithm has adequate performance at 412. The performance of the machine learning algorithm can be checked with fixed weights on the entire dataset to confirm that the algorithm has adequate performance in any suitable manner, in some embodiments.

[0049] Turning to FIG. 5, an example 500 of a production (i.e., not training or validation)) process using the machine learning algorithm to determine PE data in accordance with some embodiments is shown.

[0050] As illustrated, at 502, this process can receive CT scan data (e.g., a CT scan's 3D matrix image data) having the same or similar characteristics as the CT scan data received for training, as described above, in some embodiments. This data can be received from any suitable source and in any suitable manner, in some embodiments.

[0051] Next, as shown, at 504, the received data can be processed in any suitable manner. For example, in some embodiments, in production, the received CT scan data can be processed in the same or similar manner to that in which the training data was processed. More particularly, for example, the data can be decompressed, intensity scaled, and resized to have any suitable image such (e.g., such as 96*96*96 pixels).

[0052] Then, the received data can be evaluated at 506 to determine PE data using the trained machine learning algorithm. Before doing so, in some embodiments, the machine learning algorithm can be loaded using its initial architecture (e.g., U-Net or Swin U-Net Transformers) and the trained parameters, weights, biases, and / or any other suitable parameters applied, in some embodiments. In some embodiments, the model can be loaded on a GPU or a CPU.

[0053] Any suitable PE data can be generated by the trained machine learning algorithm, in some embodiments. For example, in some embodiments, the PE data can include any suitable data related to a PE corresponding to the CT scan data, such as that a PE occurred, one or more general locations of a PE (e.g., right, left, bilateral, proximal, distal, multiple, saddle, lobar, etc.), one or more specific locations of a PE (e.g., pixel and / or voxel locations of a PE in one or more corresponding 2D and / or 3D CT images), extent / severity data, PE-related-condition data (e.g., dilated right ventricle, pulmonary artery dilation), and / or any other suitable data relating to an indication that a PE occurred.

[0054] In some embodiments, the PE data can indicate that no PE occurred in a given piece of CT image data. As such, the PE data can indicate that a PE did not occur, that there were no general locations of a PE (e.g., a PE did not occur in right, left, bilateral, proximal, distal, multiple, saddle, lobar, etc.), that there were no specific locations of a PE (e.g., identify that there were no pixel and / or voxel locations of a PE), that there was no extent / severity data, that there was no PE-related-condition data (e.g., no dilated right ventricle, no pulmonary artery dilation), and / or any other suitable data relating to an indication that a PE did not occur.

[0055] In some embodiments, indications of whether a PE is estimated to have occurred or not can be based on whether a suspected PE corresponds to at least a threshold number of pixels and / or voxels in the CT image data. In some embodiments, this threshold can be established inany suitable manner. For example, in some embodiments, this threshold can be established using a logistic regression with the threshold corresponding to a maximum combined sensitivity and specificity. In some embodiments, PE data output to a clinician can include how the PE pixels and / or voxels compared to the threshold, the area under the curve (AUC), and accuracy.

[0056] In some embodiments, PE location can be deduced by splitting CT image data into quadrants. In some embodiments, using these quadrants, PE location can be reflected as being right, left, bilateral, proximal, distal, multiple, saddle, lobar, etc.

[0057] In some embodiments, PE severity can be interpreted from the PE's extension (e.g., a bilateral PE indicates a severe PE) and indirect signs like right ventricle dilation and pulmonary artery dilation.

[0058] Finally, as shown in FIG. 5, outcome data can be presented at 508. The outcome data can be presented in any suitable manner, in some embodiments. For example, in some embodiments, a web application user interface as shown in one or more of FIGS. 6-11 can be provided to present one or more of the output value and an interpretation to the clinician user for decision support. In some embodiments, slices of DICOM images used for prediction can be extracted and shown to a user in the web application with pixels and / or voxels overlayed thereon to indicate estimated locations of one or more PEs. In some embodiments, any suitable number of ordered and evenly distributed slices can be selected regularly from the pool of slices to provide a sample representation from the pool. In some embodiment, the slices can be selected based on their relevance / importance to the algorithm by selecting the slices with the highest density of relevant pixels as provided by occlusion and heat maps as described below.

[0059] In some embodiments, anatomy areas that the machine learning algorithm has deemed important for prediction can be presented to the user. In some embodiments, heat mapsand / or occlusions maps based on importance of anatomy areas in estimating the PE data can be presented to the user. In some embodiments, the most-relevant areas can be identified as an aggregation of pixels on a given slice or an aggregation of voxels on a given 3D volume, wherein important pixels and voxels can be identified through the relationship input-weights.

[0060] Turning to FIG. 6, an example 600 of a user interface that can be presented to a user in accordance with some embodiments is shown. As illustrated, user interface 600 includes region 602 in which a user can drop a CT scan image file so that the file is received by process 500 at 502. After processing and evaluating the image(s) at 504 and 506, at 508, interface 600 can indicate at 604 that a pulmonary embolism is suspected and can generate pixels 606 and 608 that overlay the CT images to reflect that this is a location of a suspected embolism. The overlayed pixels can be in any suitable color and / or have any other suitable characteristics (e.g., be blinking on and off). Although not shown in the example of FIG. 6, in some embodiments, user interfaces can additionally or alternatively indicate one or more general locations of a PE (e g., right, left, bilateral, proximal, distal, multiple, saddle, lobar, etc.), extent / severity data, PE- related-condition data (e.g., dilated right ventricle, pulmonary artery dilation), and / or any other suitable data relating to an indication that a PE occurred.

[0061] Turning to FIGS. 7 and 8, examples 700 and 800 of user interfaces that can be presented to a user in accordance with some embodiments are shown. As illustrated in FIG. 7, user interface 700 shows a 3D view of a pulmonary embolism mask (i.e., voxels suspect to represent one or more pulmonary embolism locations). These voxels can be shown in any suitable color in some embodiments. In some embodiments, a user can use a cursor to identify a voxel in the user interface and view coordinates corresponding to the location of the voxel. As illustrated in FIG. 8, user interface 800 shows a 3D view of a connected components ofpulmonary embolism(s) (i.e., voxels suspected of belonging to common group of voxels). In some embodiments, each group can be given a color so that the voxels common to a group all have the same color. In some embodiments, a user can use a cursor to identify a voxel in the user interface and view coordinates corresponding to the location of the voxel as well as a group identifier (e.g., "Region 11" as shown in the figure).

[0062] Turning to FIGS. 9-11, examples 900, 1000, and 1100 of user interfaces that can be presented to a user in accordance with some embodiments are shown. As illustrated in FIG. 9, user interface 900 shows still 2D images of CT image data that can be overlay ed with pixels identifying PE locations. As illustrated in FIG. 10, user interface 1000 shows moving 2D images (i.e., video) of CT image data that can be overlayed with pixels identifying PE locations. As illustrated in FIG.11, user interface 1100 shows 3D representations identifying PE locations as described above in connection with FIGS. 7 and 8. As shown on the left side of interfaces 900, 1000, and 1100, these interfaces can show further information about suspected PE(s), such as that a PE is suspected, how many PE pixels / voxels were detected, how many pixels / voxels were confidently identified as being PE pixels (e.g., based on a threshold), how many PE regions are suspected, how many PE regions are significant, the size of a largest PE region, and / or any other suitable data.

[0063] It should be understood that at least some of the above-described blocks of the processes of FIGS. 4 and / or 5 can be executed or performed in any order or sequence not limited to the order and sequence shown in and described in the figures. Also, some of the above blocks of the processes of FIGS. 4 and / or 5 can be executed or performed substantially simultaneously where appropriate or in parallel to reduce latency and processing times. Additionally oralternatively, some of the above described blocks of the processes of FIGS. 4 and / or 5 can be omitted.

[0064] In some embodiments, any suitable computer readable media can be used for storing instructions for performing the functions and / or processes described herein. For example, in some embodiments, computer readable media can be transitory or non-transitory. For example, non-transitory computer readable media can include media such as non-transitory magnetic media (such as hard disks, floppy disks, and / or any other suitable magnetic media), non- transitory optical media (such as compact discs, digital video discs, Blu-ray discs, and / or any other suitable optical media), non-transitory semiconductor media (such as flash memory, electrically programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), and / or any other suitable semiconductor media), any suitable media that is not fleeting or devoid of any semblance of permanence during transmission, and / or any suitable tangible media. As another example, transitory computer readable media can include signals on networks, in wires, conductors, optical fibers, circuits, any suitable media that is fleeting and devoid of any semblance of permanence during transmission, and / or any suitable intangible media.

[0065] Although the invention has been described and illustrated in the foregoing illustrative embodiments, it is understood that the present disclosure has been made only by way of example, and that numerous changes in the details of implementation of the invention can be made without departing from the spirit and scope of the invention, which is limited only by the claims that follow. Features of the disclosed embodiments can be combined and rearranged in various ways.

Claims

What is claimed is:

1. A system for indicating estimated pulmonary embolism data, comprising: memory; and at least one hardware processor coupled to the memory and collectively configured to at least: receive medical data; process the medical data to provide processed data; evaluate the processed data using a first machine learning algorithm; generate the estimated pulmonary embolism data based on the evaluating the processed data; and cause the expected outcome to be presented to a user.

2. The system of claim 1, wherein the medical data is 3D matrix image data from a computed tomography (CT) scan.

3. The system of claim 1, wherein processing the medical data includes at least one of decompressing, intensity scaling, and resizing the medical data.

4. The system of claim 1, wherein the first machine learning algorithm is a deep convolutional neural network.

5. The system of claim 1, wherein the at least one hardware processor is further configured to:generate a plurality of pixels and / or voxels for an estimated pulmonary embolism; and cause the plurality of pixels and / or voxels to be presented with CT image data.

6. The system of claim 5, wherein the at least one hardware processor is further configured to indicate a general location of a pulmonary embolism.

7. The system of claim 1, wherein the at least one hardware processor is further configured to indicate a severity of a pulmonary embolism.

8. A method for indicating estimated pulmonary embolism data, comprising: receiving medical data; processing the medical data to provide processed data; evaluating the processed data using a first machine learning algorithm using a hardware processor; generating the estimated pulmonary embolism data based on the evaluating the processed data; and causing the expected outcome to be presented to a user.

9. The method of claim 8, wherein the medical data is 3D matrix image data from a computed tomography (CT) scan.

10. The method of claim 8, wherein processing the medical data includes at least one of decompressing, intensity scaling, and resizing the medical data.

11. The method of claim 8, wherein the first machine learning algorithm is a deep convolutional neural network.

12. The method of claim 8, further comprising: generating a plurality of pixels and / or voxels for an estimated pulmonary embolism; and causing the plurality of pixels and / or voxels to be presented with CT image data.

13. The method of claim 8, further comprising indicating a general location of a pulmonary embolism.

14. The method of claim 8, further comprising indicating a severity of a pulmonary embolism.

15. A non-transitory computer-readable medium containing computer executable instructions that, when executed by a processor, cause the processor to perform a method for indicating estimated pulmonary embolism data, the method comprising: receiving medical data; processing the medical data to provide processed data; evaluating the processed data using a first machine learning algorithm; generating the estimated pulmonary embolism data based on the evaluating the processed data; and causing the expected outcome to be presented to a user.

16. The non-transitory computer-readable medium of claim 15, wherein the medical data is 3D matrix image data from a computed tomography (CT) scan.

17. The non-transitory computer-readable medium of claim 15, wherein processing the medical data includes at least one of decompressing, intensity scaling, and resizing the medical data.

18. The non-transitory computer-readable medium of claim 15, wherein the first machine learning algorithm is a deep convolutional neural network.

19. The non-transitory computer-readable medium of claim 15, wherein the method further comprises: generating a plurality of pixels and / or voxels for an estimated pulmonary embolism; and causing the plurality of pixels and / or voxels to be presented with CT image data.

20. The non-transitory computer-readable medium of claim 15, wherein the method further comprises indicating a general location of a pulmonary embolism.21 . The non-transitory computer-readable medium of claim 15, wherein the method further comprises indicating a severity of a pulmonary embolism.

Images