Radiomic fingerprint prediction of pulmonary embolism clot types

A radiomic fingerprinting method using machine learning models to predict PE clot types enhances treatment planning for intermediate-risk patients by providing quantitative insights into clot characteristics, addressing the lack of consensus in clinical guidelines.

WO2026236746A1PCT designated stage Publication Date: 2026-11-12MEDICAL COLLEGE OF WISCONSIN INC
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
PCT/US2026/027603
Authority / Receiving Office
WO · WO
Patent Type
Applications
Current Assignee / Owner
Priority Date
2025-05-09
Filing Date
2026-05-11
Publication Date
2026-11-12

AI Technical Summary

Technical Problem

Current clinical guidelines lack consensus on optimal management strategies for intermediate-risk pulmonary embolism (PE) patients, complicating treatment decisions due to varying clinical factors and patient-specific conditions.

Method used

A radiomic fingerprinting method using machine learning models to predict PE clot types by analyzing medical imaging data, extracting radiomic features, and segmenting clots to generate classified feature data for improved treatment planning.

Benefits of technology

Provides quantitative insights into clot characteristics, aiding clinicians in determining appropriate interventions for intermediate-risk PE patients, potentially reducing treatment uncertainties and improving patient outcomes.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure US2026027603_12112026_PF_FP_ABST
    Figure US2026027603_12112026_PF_FP_ABST
Patent Text Reader

Abstract

A method for predicting pulmonary embolism (PE) clot types is provided herein. The method includes receiving medical imaging data that depict one or more pulmonary arteries of a patient, segmenting one or more PE clots from the medical imaging data, extracting radiomic features from the segmented PE clots, and inputting the radiomic features to a trained machine learning model. The trained machine learning model generates classified feature data as an output, wherein the classified feature data indicate a prediction of a clot type for each of the segmented PE clots.
Need to check novelty before this filing date? Find Prior Art

Description

MOW C2408650053.01305RADIOMIC FINGERPRINT PREDICTION OF PULMONARY EMBOLISM CLOT TYPES CROSS-REFERENCE TO RELATED APPLICATIONS

[0001] This application claims the benefit of U.S. Provisional Patent Application Serial No. 63 / 803,410, filed on May 9, 2025, and entitled “RADIOMIC FINGERPRINT PREDICTION OF PULMONARY EMBOLISM CLOT TYPES which is herein incorporated by reference in its entirety.BACKGROUND

[0002] Venous thromboembolism, which includes pulmonary embolism (PE), is a significant cardiovascular condition. Current medical practice categorizes acute PEs into different risk levels based on symptoms and hemodynamic status. Management of intermediate-risk PEs. defined as symptomatic cases without cardiogenic shock, presents challenges due to varying clinical guidelines and patient-specific factors.

[0003] Radiomics, an emerging field in medical imaging analysis, has shown potential in developing quantitative signatures associated with tissue characteristics and clinical outcomes. This approach may offer new insights into PE clot t pes and burden, potentially aiding in treatment decisions for patients with intermediate-risk PEs.SUMMARY OF THE DISCLOSURE

[0004] According to an aspect of the present disclosure, a method for predicting pulmonary embolism (PE) clot types is provided. The method includes receiving medical imaging data that depict one or more pulmonary arteries of a patient. The method includes segmenting one or more PE clots from the medical imaging data. The method includes extracting radiomic features from the segmented PE clots. The method includes inputting the radiomic features to a trained machine learning model, generating classified feature data as an output. The classified feature data indicate a prediction of a clot t pe for each of the segmented PE clots.

[0005] According to another aspect of the present disclosure, a system for predicting pulmonary embolism (PE) clot types is provided. The system includes a computing device comprising a processor and a memory. The memory stores a trained machine learning model. The processor is configured to receive medical imaging data that depict one or more pulmonary 1QB\650053.01305\102412286.1MCW C2408650053.01305arteries of a patient. The processor is configured to segment one or more PE clots from the medical imaging data. The processor is configured to extract radiomic features from the segmented PE clots. The processor is configured to input the radiomic features to the trained machine learning model, generating classified feature data as an output. The classified feature data indicate a prediction of a clot type for each of the segmented PE clots.

[0006] According to another aspect of the present disclosure, a non-transitory computer-readable medium is provided. The non-transitory computer-readable medium stores instructions that, when executed by a processor, cause the processor to receive medical imaging data that depict one or more pulmonary' arteries of a patient. The instructions cause the processor to segment one or more PE clots from the medical imaging data. The instructions cause the processor to extract radiomic features from the segmented PE clots. The instructions cause the processor to input the radiomic features to a trained machine learning model, generating classified feature data as an output. The classified feature data indicate a prediction of a clot type for each of the segmented PE clots.BRIEF DESCRIPTION OF THE DRAWINGS

[0007] FIG. 1 is a flowchart of a method for generating classified feature data using a trained neural network.

[0008] FIG. 2 is a flowchart of a method for training a neural network on training data.

[0009] FIG. 3 is a block diagram of an example system for predicting PE clot ty pes using a radiomic fingerprint analysis.

[0010] FIG. 4 is a block diagram of example components that can implement the system of FIG. 3.DETAILED DESCRIPTION

[0011] Described here are systems and methods for predicting pulmonary' embolism clot types using radiomics. Radiomics refers to the extraction and analysis of large amounts of quantitative features from medical images. This approach may allow for the identification of imaging biomarkers and patterns that are not apparent to the human eye. By applying radiomic techniques to imaging data from pulmonary embolism (PE) patients, it may be possible to develop a “fingerprint” that characterizes different clot types and distributions.

[0012] PE is a serious cardiovascular condition that occurs when a blood clot becomes lodged in the pulmonary arteries. Current clinical practice categorizes PE cases into low-risk,2QB\650053.01305\ 102412286.1MCW C2408650053.01305intermediate-risk (defined as symptomatic PE without cardiogenic shock), and high-risk (defined as symptomatic with evidence of cardiogenic shock) groups based on factors such as hemodynamic stability and evidence of right ventricular dysfunction. While treatment guidelines exist for low-risk and high-risk PE cases, there is less consensus on optimal management strategies for intermediate-risk patients.

[0013] The systems and methods described herein aim to leverage radiomics to predict PE clot types and burden, particularly for intermediate-risk patients. This information may assist clinicians in determining appropriate interventions and treatment plans. The radiomic approach may provide additional insights beyond conventional imaging assessment to support clinical decision-making for this challenging patient population.

[0014] As described above, venous thromboembolism, which includes PE is the third leading cause of acute cardiovascular syndrome behind myocardial infarction and stroke. The Pulmonary Embolism Response Team (PERT) is a multidisciplinary group of medical and surgical sub-specialists who collaborate to discuss what acute treatment is best for intermediate- and high-risk patients with a pulmonary embolism. PERT encompasses a method of decision-making that can effectively decrease the delay in patient care in potentially lifethreatening presentations. Implementation of a PERT has been shown to reduce 30-day mortality rates, lower rates of major bleeding, and shorter times to intervention. Sub-specialists on PERTs include hospitalists, pulmonology, critical care, interventional radiologists, radiology, cardiolog}', cardiothoracic surgery, vascular surgery, and hematology among others.

[0015] Current research categorizes acute PEs into low-risk, intermediate-risk (defined as symptomatic PE without cardiogenic shock), and high-risk (defined as symptomatic with evidence of cardiogenic shock) events. The challenge with treating and managing pulmonary emboli is multifactorial. There is no consensus among the many major clinical guidelines, including American Heart Association (AHA), European Society of Cardiology (ESC), American Society of Hematology (ASH), and American College of Chest Physicians CHEST guidelines regarding interventions. As such, even the ESC has recommended PERT as a Class Ila Level C recommendation for intermediate- and high-risk PEs, for which management can be difficult and very patient-specific.

[0016] Challenges affecting PE treatment include the time frame of decision making, comorbidities, time to first anticoagulation, other initial treatments and / or interventions, available resources for intervention, available resources for diagnosis, disposition, and hemodynamic instability . PERT is activated at MCW for intermediate- and high-risk PE3QB\650053.01305\ 102412286.1MCW C2408650053.01305patients, typically as defined by Pulmonary Embolism Severity Index (PESI or sPESI) 30-day mortality prognostic criteria. Radiomics has been associated with different clinical end points over the years, one of which developing a fingerprint associated with tissue microenvironment using medical imaging. The purpose of this study is to establish a PERT Database that will allow for quality improvement and retrospective studies, with the intention of ongoing data retrieval to allow developing a radiomic fingerprint of PE clot types and burden.

[0017] Referring now to FIG. 1, a flowchart is illustrated as setting forth the steps of an example method for generating classified feature data using a suitably trained neural network or other machine learning algorithm. As will be described, the neural network or other machine learning algorithm takes radiomic features extracted from medical imaging data of PE clots as input data and generates classified feature data as output data. As an example, the classified feature data can be indicative of PE clot types, such as saddle, main, segmental, or subsegmental clots based on their anatomical location within the pulmonary vasculature.

[0018] The method includes accessing medical imaging data with a computer system, as indicated at step 102. The medical imaging data may include computed tomography (CT) angiography images and / or magnetic resonance angiography (MRA) images that depict one or more pulmonary' arteries of a patient. Accessing the medical imaging data may include retrieving such data from a memory' or other suitable data storage device or medium. Additionally or alternatively, accessing the medical imaging data may include acquiring such data with a CT scanner, MR1 system, or other medical imaging device and transferring or otherwise communicating the data to the computer system, which may be a part of the medical imaging system or a separate workstation. In some implementations, the collected medical images may undergo preprocessing to enhance image quality and facilitate accurate clot segmentation. This preprocessing may involve image normalization, noise reduction, and contrast enhancement techniques.

[0019] The method may include segmenting one or more PE clots from the medical imaging data, as indicated at step 104. Clot segmentation may be performed using specialized medical imaging software. In some implementations, MIM Software™ may be utilized for segmenting PE clots from the medical images. The segmentation process may involve delineating the boundaries of the clots within the pulmonary' arteries. The segmented clot regions may be classified according to their anatomical location within the pulmonary vasculature. In some implementations, the clots may be categorized into four groups such as saddle, main, segmental, or subsegmental, based on their origin and extent within the4QB\650053.01305\ 102412286.1MCW C2408650053.01305pulmonary arteries. In other implementations, the clots may be categorized into three groups: saddle (SD). main (MN), and segmental or subsegmental (SSEG), where the segmental and subsegmental categories are combined into a single category.

[0020] The method may include extracting radiomic features from the segmented PE clots, as indicated at step 106. The feature extraction process may involve analyzing the medical image data to quantify various characteristics of the clots. Volumetric measurements may be obtained to assess the size and extent of each clot. In some implementations, the maximum intensify value within the segmented clot region may be determined. Additionally, kurtosis, which describes the shape of the intensify distribution, may be calculated for each clot. Skewness, which measures the asymmetry of the intensity distribution within the clot region, may also be extracted as a radiomic feature. In still other cases, other texture features or statistical measures may be computed. Feature extraction may be performed using specialized medical image analysis software. In some cases, a large number of radiomic features may be initially extracted from the image data.

[0021] The method may include selecting significant radiomic features for input to the neural network, as indicated at step 108. Statistical analysis techniques such as analysis of variance (ANOVA) and regression analysis may be applied to identify the most significant features for distinguishing between different clot types. In some implementations, the volume, maximum intensify, skewness, and kurtosis measurements may be found to be particularly informative features. These features may be incorporated as inputs to a neural network model for clot type classification. The use of these specific radiomic features may allow the model to capture important characteristics related to clot size, density, and heterogeneity'.

[0022] A trained neural network (or other suitable machine learning algorithm) is then accessed with the computer system, as indicated at step 110. In general, the neural network is trained, or has been trained, on training data in order to differentiate between PE clot types based on anatomical location within the pulmonary' vasculature. Accessing the trained neural network may include accessing network parameters (e.g., weights, biases, or both) that have been optimized or otherwise estimated by training the neural network on training data. In some instances, retrieving the neural network can also include retrieving, constructing, or otherwise accessing the particular neural network architecture to be implemented. For instance, data pertaining to the layers in the neural network architecture (e.g., number of layers, type of layers, ordering of layers, connections between layers, hyperparameters for layers) may be retrieved, selected, constructed, or otherwise accessed.5QB\650053.01305\ 102412286.1MCW C2408650053.01305

[0023] An artificial neural network generally includes an input layer, one or more hidden layers (or nodes), and an output layer. Typically, the input layer includes as many nodes as inputs provided to the artificial neural network. The number (and the type) of inputs provided to the artificial neural network may vary based on the particular task for the artificial neural network. In some implementations, the input layer may receive radiomic features such as volume, maximum intensity, skewness, and kurtosis.

[0024] The input layer connects to one or more hidden layers. The number of hidden layers varies and may depend on the particular task for the artificial neural network. Additionally, each hidden layer may have a different number of nodes and may be connected to the next layer differently. For example, each node of the input layer may be connected to each node of the first hidden layer. The connection between each node of the input layer and each node of the first hidden layer may be assigned a weight parameter. Additionally, each node of the neural network may also be assigned a bias value. In some configurations, each node of the first hidden layer may not be connected to each node of the second hidden layer. That is, there may be some nodes of the first hidden layer that are not connected to all of the nodes of the second hidden layer. The connections between the nodes of the first hidden layers and the second hidden layers are each assigned different weight parameters. Each node of the hidden layer is generally associated with an activation function. The activation function defines how the hidden layer is to process the input received from the input layer or from a previous input or hidden layer. These activation functions may vary and be based on the type of task associated with the artificial neural network and also on the specific type of hidden layer implemented.

[0025] Each hidden layer may perform a different function. For example, some hidden layers can be convolutional hidden layers which can. in some instances, reduce the dimensionality of the inputs. Other hidden layers can perform statistical functions such as max pooling, which may reduce a group of inputs to the maximum value; an averaging layer; batch normalization; and other such functions. In some of the hidden layers each node is connected to each node of the next hidden layer, which may be referred to then as dense layers. Some neural networks including more than, for example, three hidden layers may be considered deep neural networks.

[0026] The last hidden layer in the artificial neural network is connected to the output layer. Similar to the input layer, the output layer typically has the same number of nodes as the possible outputs. In an example in which the artificial neural network classifies PE clot types.6QB\650053.01305\ 102412286.1MCW C2408650053.01305the output layer may include, for example, a number of different nodes, where each different node corresponds to a different clot type classification. A first node may indicate a saddle (SD) clot, a second node may indicate a main (MN) clot, a third node may indicate a segmental (SEG) clot, and a fourth node may indicate a subsegmental (SSEG) clot. In some implementations, the output may be in the form of probabilities for each clot type, enabling the model to provide confidence levels for its predictions.

[0027] The radiomic features (e.g., the selected subset of significant features, the full set of radiomic features, etc.) are then input to the one or more trained neural networks, generating output as classified feature data, as indicated at step 112. For example, the classified feature data may include a clot type prediction. The clot type prediction can provide physicians or other clinicians with information regarding the anatomical location and characteristics of PE clots, which may aid in determining appropriate interventions for intermediate-risk patients.

[0028] As another example, the classified feature data may indicate the probability for a particular clot type classification (i.e., the probability that the radiomic features include patterns, features, or characteristics indicative of detecting, differentiating, and / or determining the type of one or more PE clots).

[0029] Additionally or alternatively, the classified feature data may classify the radiomic features as indicating a particular PE clot type. In these instances, the classified feature data can differentiate between different clot types, such as saddle, main, segmental, and subsegmental clots. In still other embodiments, the classified feature data may indicate a severity or burden of pulmonary embolism. For example, the classified feature data may include a burden score that quantifies the extent of pulmonary embolism based on clot characteristics.

[0030] The classified feature data generated by inputting the radiomic features to the trained neural network(s) can then be displayed to a user, stored for later use or further processing, or both, as indicated at step 114. In some implementations, the classified feature data may be presented to members of a Pulmonary Embolism Response Team (PERT), which is a multidisciplinary group of medical and surgical sub-specialists who collaborate to discuss acute treatment options for intermediate- and high-risk patients with pulmonary embolism. The classified feature data may facilitate treatment discussions among PERT members by providing quantitative information regarding clot types and characteristics. The radiomic features extracted from medical images may offer insights into clot characteristics that are not readily apparent through visual inspection alone. In some implementations, the prediction results may7QB\650053.01305\ 102412286.1MCW C2408650053.01305be incorporated into existing clinical decision support systems used by PERTs. The model outputs, such as predicted clot type and associated probabilities, may be presented alongside other clinical and imaging data to aid in treatment planning discussions.

[0031] In some cases, the classified feature data may help guide selection between different treatment options. For example, given the lack of consensus among major clinical guidelines, including American Heart Association (AHA), European Society of Cardiology (ESC), American Society of Hematology (ASH), and American College of Chest Physicians CHEST guidelines regarding interventions for intermediate-risk PE, the classified feature data may provide additional information to aid in determining whether anticoagulation alone, catheter-directed thrombolysis, or surgical embolectomy may be appropriate for a particular patient. The radiomic fingerprint may potentially help identify subgroups of intermediate-risk patients who may benefit from more aggressive interventions. In some cases, patients with certain predicted clot characteristics may be considered for catheter-directed thrombolysis or surgical embolectomy, while others may be managed with anticoagulation alone.

[0032] In some implementations, the classified feature data may be used for risk stratification and prognostication. For example, the classified feature data may supplement existing risk indices, such as the Pulmonary Embolism Severity Index (PESI) or simplified PESI (sPESI), which are used for 30-day mortality prognostication. Additionally or alternatively, the classified feature data may be used to predict pulmonary embolism burden associated with right ventricular cardiac function, which may aid in determining the type of intervention appropriate for intermediate-risk clots. In some implementations, the radiomic fingerprint prediction may be combined with other clinical variables in a multivariate risk model. This integrated approach may provide a more comprehensive assessment of patient risk and guide selection of optimal management strategies.

[0033] In some cases, the classified feature data may be integrated into radiology reporting workflows. For example, automated analysis and inclusion of radiomic findings in standardized reports may facilitate rapid communication of relevant clot characteristics to treating clinicians. In some cases, the prediction model may be implemented as part of a clinical decision algorithm, with specific management recommendations based on the predicted clot characteristics and other clinical factors.

[0034] Referring now to FIG. 2, a flowchart is illustrated as setting forth the steps of an example method for training one or more neural networks (or other suitable machine learning algorithms) on training data, such that the one or more neural networks are trained to8QB\650053.01305\ 102412286.1MCW C2408650053.01305receive radiomic features extracted from medical imaging data of PE clots as input data in order to generate classified feature data as output data, where the classified feature data are indicative of PE clot types based on anatomical location within the pulmonary vasculature.

[0035] In general, the neural network(s) can implement any number of different neural network architectures. In some implementations, the neural network(s) may implement a bilayered neural network architecture with three fully connected layers. The neural network(s) may include hidden layers with non-linear activation functions, such as rectified linear unit (RELU) activation functions, allowing the model to learn complex relationships between the input features and clot types. Alternatively, the neural network(s) could be replaced with other suitable machine learning or artificial intelligence algorithms, such as those based on supervised learning, unsupervised learning, deep learning, ensemble learning, dimensionality reduction, and so on.

[0036] The method includes accessing training data with a computer system, as indicated at step 202. Accessing the training data may include retrieving such data from a memory or other suitable data storage device or medium. Alternatively, accessing the training data may include acquiring such data with a CT scanner, MRI system, or other medical imaging device and transferring or otherwise communicating the data to the computer system.

[0037] In general, the training data can include medical imaging data depicting PE clots in pulmonary arteries. Additionally, the training data may include other data, such as patient clinical information, hemodynamic status, and treatment outcomes. In some embodiments, the training data may include PE clot images that have been labeled (e.g., labeled as containing patterns, features, or characteristics indicative of saddle, main, segmental, or subsegmental clot types; and the like).

[0038] The method can include assembling training data from a PERT database or other clinical imaging repository using a computer system. This step may include assembling the medical imaging data into an appropriate data structure on which the neural network or other machine learning algorithm can be trained. Assembling the training data may include assembling medical imaging data, segmented PE clot regions, and other relevant data. For instance, assembling the training data may include generating labeled data and including the labeled data in the training data. Labeled data may include medical imaging data, segmented PE clot regions, or other relevant data that have been labeled as belonging to, or otherwise being associated with, one or more different classifications or categories. For instance, labeled data may include medical imaging data and / or segmented PE clot regions that have been9QB\650053.01305\ 102412286.1MCW C2408650053.01305labeled as being associated with saddle, main, segmental, or subsegmental clot ty pes based on their anatomical origin within the pulmonary vasculature.

[0039] One or more neural networks (or other suitable machine learning algorithms) are trained on the training data, as indicated at step 204. In general, the neural network can be trained by optimizing network parameters (e.g., weights, biases, or both) based on minimizing a loss function. As one non-limiting example, the loss function may be a mean squared error loss function or a cross-entropy loss function. In some implementations, hyperparameter tuning may be performed using Bayesian optimization with a probability of improvement acquisition function. The hyperparameters may be optimized at regular intervals, such as every 30 iterations, during the training process. In some cases, the training process may have an iteration limit, such as 1000 iterations.

[0040] Training a neural network may include initializing the neural network, such as by computing, estimating, or otherwise selecting initial network parameters (e.g., weights, biases, or both). During training, an artificial neural network receives the inputs for a training example and generates an output using the bias for each node, and the connections between each node and the corresponding weights. For instance, training data can be input to the initialized neural network, generating output as classified feature data indicating predicted clot ty pes. The artificial neural network then compares the generated output with the actual output of the training example in order to evaluate the quality of the clot type predictions. For instance, the predicted clot types can be passed to a loss function to compute an error. The current neural network can then be updated based on the calculated error (e.g., using backpropagation methods based on the calculated error). For instance, the current neural network can be updated by updating the network parameters (e.g., weights, biases, or both) in order to minimize the loss according to the loss function. The training continues until a training condition is met. The training condition may correspond to, for example, a predetermined number of training examples being used, a minimum accuracy threshold being reached during training and validation, a predetermined number of validation iterations being completed, and the like. When the training condition has been met (e.g.. by determining whether an error threshold or other stopping criterion has been satisfied), the current neural network and its associated network parameters represent the trained neural network. Different types of training processes can be used to adjust the bias values and the weights of the node connections based on the training examples. The training processes may include, for example, gradient descent, Newton's method, conjugate gradient, quasi-Newton, Levenberg-Marquardt, among others.10QB\650053.01305\ 102412286.1MCW C2408650053.01305

[0041] The artificial neural network can be constructed or otherwise trained based on training data using one or more different learning techniques, such as supervised learning, unsupervised learning, reinforcement learning, ensemble learning, active learning, transfer learning, or other suitable learning techniques for neural networks. As an example, supervised learning involves presenting a computer system with example inputs and their actual outputs (e.g., clot type categorizations). In these instances, the artificial neural network is configured to learn a general rule or model that maps the inputs to the outputs based on the provided example input-output pairs. To evaluate the performance of the model, cross-validation techniques may be employed. In some implementations, the area under the receiver operating characteristic curve (AUC-ROC) may be used as a metric to assess the model's ability to distinguish between different clot types.

[0042] The one or more trained neural networks are then stored for later use, as indicated at step 206. Storing the neural network(s) may include storing network parameters (e.g., weights, biases, or both), which have been computed or otherwise estimated by training the neural network(s) on the training data. Storing the trained neural network(s) may also include storing the particular neural network architecture to be implemented. For instance, data pertaining to the layers in the neural network architecture (e.g., number of layers, type of layers, ordering of layers, connections between layers, hyperparameters for layers) may be stored.

[0043] An example study was conducted to evaluate the potential of developing a radiomic fingerprint for predicting PE clot types and aiding in the determination of appropriate interventions for intermediate-risk patients. Venous thromboembolism, which includes pulmonary embolism (PE), represents a significant cause of acute cardiovascular syndrome. Current research categorizes acute PEs into low-risk, intermediate-risk (defined as symptomatic PE without cardiogenic shock), and high-risk (defined as symptomatic with evidence of cardiogenic shock) events. However, there is no consensus among the many major clinical guidelines, including American Heart Association (AHA), European Society7of Cardiology (ESC), American Society of Hematology (ASH), and American College of Chest Physicians CHEST guidelines regarding interventions for intermediate-risk PE. Radiomics has been associated with different clinical end points over the years, including the development of fingerprints associated with tissue microenvironment using medical imaging.

[0044] In this example study, a retrospective analysis was performed on 61 individual PE clots identified from 23 patients in a PERT database. The clots were segmented using MIM Software™ by an experienced hematology oncology fellow and cross-validated by senior11QB\650053.01305\ 102412286.1MCW C2408650053.01305radiologists. The clots were then divided into groups according to their anatomical origin, including saddle, main, segmental, and subsegmental categories. Radiomic features were extracted from each clot type, and ANOVA and regression analysis were applied to identify significant features for building a classification model.

[0045] Cross-validated bi-layered neural networks were developed to predict different clot types. The performance of each model was evaluated using the cross-validated area under the receiver operating characteristic curve (CV-AUC). The analysis demonstrated that a bilayered neural network with three fully connected layers incorporating volume, maximum intensify, and kurtosis may differentiate saddle (SD) clots from segmental / subsegmental (SSEG) clots with a CV-AUC of 0.98, main (MN) clots from saddle clots with a CV-AUC of 0.96, and main clots from segmental / subsegmental clots with a CV-AUC of 0.81.

[0046] These results suggest that developing a radiomic fingerprint to predict PE clot type may be feasible. With larger verification studies, this approach may potentially be developed into a useful tool for predicting pulmonary embolism burden associated with right ventricular cardiac function, which may aid in determining the type of intervention appropriate for intermediate-risk clots.

[0047] FIG. 3 shows an example of a system 300 for predicting PE clot types from radiomic fingerprinting in accordance with some embodiments described in the present disclosure. As shown in FIG. 3, a computing device 350 can receive one or more types of data (e.g., medical image data) from data source 302. In some embodiments, computing device 350 can execute at least a portion of a radiomic fingerprint-based PE clot type prediction system 304 to predict PE clot types from data received from the data source 302.

[0048] Additionally or alternatively, in some embodiments, the computing device 350 can communicate information about data received from the data source 302 to a server 352 over a communication network 354, which can execute at least a portion of the radiomic fingerprint-based PE clot type prediction system 304. In such embodiments, the server 352 can return information to the computing device 350 (and / or any other suitable computing device) indicative of an output of the radiomic fingerprint-based PE clot type prediction system 304.

[0049] In some embodiments, computing device 350 and / or server 352 can be any suitable computing device or combination of devices, such as a desktop computer, a laptop computer, a smartphone, a tablet computer, a wearable computer, a server computer, a virtual machine being executed by a physical computing device, and so on. The computing device 350 and / or server 352 can also reconstruct images from the data.12QB\650053.01305\ 102412286.1MCW C2408650053.01305

[0050] In some embodiments, data source 302 can be any suitable source of data (e.g., measurement data, images reconstructed from measurement data, processed image data), such as a medical imaging system, another computing device (e g., a server storing measurement data, images reconstructed from measurement data, processed image data), and so on. In some embodiments, data source 302 can be local to computing device 350. For example, data source 302 can be incorporated with computing device 350 (e.g.. computing device 350 can be configured as part of a device for measuring, recording, estimating, acquiring, or otherwise collecting or storing data). As another example, data source 302 can be connected to computing device 350 by a cable, a direct wireless link, and so on. Additionally or alternatively, in some embodiments, data source 302 can be located locally and / or remotely from computing device 350, and can communicate data to computing device 350 (and / or server 352) via a communication network (e.g., communication network 354).

[0051] In some embodiments, communication network 354 can be any suitable communication network or combination of communication networks. For example, communication network 354 can include a Wi-Fi network (which can include one or more wireless routers, one or more switches, etc.), a peer-to-peer network (e.g., a Bluetooth network), a cellular network (e.g., a 3G network, a 4G network, etc., complying with any suitable standard, such as CDMA, GSM, LTE, LTE Advanced, WiMAX, etc.), other ty pes of wireless network, a wired network, and so on. In some embodiments, communication network 354 can be a local area network, a wide area network, a public network (e.g., the Internet), a private or semi-private network (e.g., a corporate or university7intranet), any other suitable type of netw ork, or any suitable combination of netw orks. Communications links shown in FIG. 3 can each be any suitable communications link or combination of communications links, such as wired links, fiber optic links, Wi-Fi links. Bluetooth links, cellular links, and so on.

[0052] Referring now to FIG. 4, an example of hardware 400 that can be used to implement data source 302, computing device 350, and server 352 in accordance with some embodiments of the systems and methods described in the present disclosure is shown.

[0053] As shown in FIG. 4, in some embodiments, computing device 350 can include a processor 402, a display 404, one or more inputs 406, one or more communication systems 408, and / or memory 410. In some embodiments, processor 402 can be any suitable hardware processor or combination of processors, such as a central processing unit (CPU), a graphics processing unit (GPU), and so on. In some embodiments, display 404 can include any suitable display devices, such as a liquid crystal display (LCD) screen, a light-emitting diode (LED)13QB\650053.01305\ 102412286.1MCW C2408650053.01305display, an organic LED (OLED) display, an electrophoretic display (e.g., an “e-ink’‘ display), a computer monitor, a touchscreen, a television, and so on. In some embodiments, inputs 406 can include any suitable input devices and / or sensors that can be used to receive user input, such as a keyboard, a mouse, a touchscreen, a microphone, and so on.

[0054] In some embodiments, communications systems 408 can include any suitable hardware, firmware, and / or software for communicating information over communication network 354 and / or any other suitable communication networks. For example, communications systems 408 can include one or more transceivers, one or more communication chips and / or chip sets, and so on. In a more particular example, communications systems 408 can include hardware, firmware, and / or software that can be used to establish a Wi-Fi connection, a Bluetooth connection, a cellular connection, an Ethernet connection, and so on.

[0055] In some embodiments, memory 410 can include any suitable storage device or devices that can be used to store instructions, values, data, or the like, that can be used, for example, by processor 402 to present content using display 404, to communicate with server 352 via communications system(s) 408, and so on. Memory 410 can include any suitable volatile memory, non-volatile memory, storage, or any suitable combination thereof. For example, memory 410 can include random-access memory' (RAM), read-only memory (ROM), electrically programmable ROM (EPROM), electrically erasable ROM (EEPROM), other forms of volatile memory, other forms of non-volatile memory, one or more forms of semivolatile memory, one or more flash drives, one or more hard disks, one or more solid state drives, one or more optical drives, and so on. In some embodiments, memory 410 can have encoded thereon, or otherwise stored therein, a computer program for controlling operation of computing device 350. In such embodiments, processor 402 can execute at least a portion of the computer program to present content (e.g., images, user interfaces, graphics, tables), receive content from server 352, transmit information to server 352, and so on. For example, the processor 402 and the memory' 410 can be configured to perform the methods described herein.

[0056] In some embodiments, server 352 can include a processor 412, a display 414, one or more inputs 416. one or more communications systems 418. and / or memory 420. In some embodiments, processor 412 can be any suitable hardware processor or combination of processors, such as a CPU, a GPU, and so on. In some embodiments, display 414 can include any suitable display devices, such as an LCD screen, LED display, OLED display, electrophoretic display, a computer monitor, a touchscreen, a television, and so on. In some14QB\650053.01305\ 102412286.1MCW C2408650053.01305embodiments, inputs 416 can include any suitable input devices and / or sensors that can be used to receive user input, such as a keyboard, a mouse, a touchscreen, a microphone, and so on.

[0057] In some embodiments, communications systems 418 can include any suitable hardware, firmware, and / or software for communicating information over communication network 354 and / or any other suitable communication networks. For example, communications systems 418 can include one or more transceivers, one or more communication chips and / or chip sets, and so on. In a more particular example, communications systems 418 can include hardware, firmware, and / or software that can be used to establish a Wi-Fi connection, a Bluetooth connection, a cellular connection, an Ethernet connection, and so on.

[0058] In some embodiments, memory’ 420 can include any suitable storage device or devices that can be used to store instructions, values, data, or the like, that can be used, for example, by processor 412 to present content using display 414, to communicate with one or more computing devices 350, and so on. Memory’ 420 can include any suitable volatile memory, non-volatile memory', storage, or any suitable combination thereof. For example, memory 420 can include RAM. ROM, EPROM, EEPROM, other ty pes of volatile memory, other ty pes of non-volatile memory, one or more types of semi-volatile memory, one or more flash drives, one or more hard disks, one or more solid state drives, one or more optical drives, and so on. In some embodiments, memory 420 can have encoded thereon a server program for controlling operation of server 352. In such embodiments, processor 412 can execute at least a portion of the server program to transmit information and / or content (e.g., data, images, a user interface) to one or more computing devices 350, receive information and / or content from one or more computing devices 350, receive instructions from one or more devices (e.g., a personal computer, a laptop computer, a tablet computer, a smartphone), and so on.

[0059] In some embodiments, the server 352 is configured to perform the methods described in the present disclosure. For example, the processor 412 and memory 420 can be configured to perform the methods described herein.

[0060] In some embodiments, data source 302 can include a processor 422, one or more data acquisition systems 424, one or more communications systems 426, and / or memory 428. In some embodiments, processor 422 can be any suitable hardware processor or combination of processors, such as a CPU, a GPU, and so on. In some embodiments, the one or more data acquisition systems 424 are generally configured to acquire data, images, or both, and can include a medical imaging system. Additionally or alternatively, in some embodiments, the one or more data acquisition systems 424 can include any suitable hardware, firmware, and / or15QB\650053.01305\ 102412286.1MCW C2408650053.01305software for coupling to and / or controlling operations of a medical imaging system. In some embodiments, one or more portions of the data acquisition system(s) 424 can be removable and / or replaceable.

[0061] Note that, although not shown, data source 302 can include any suitable inputs and / or outputs. For example, data source 302 can include input devices and / or sensors that can be used to receive user input, such as a keyboard, a mouse, a touchscreen, a microphone, a trackpad, a trackball, and so on. As another example, data source 302 can include any suitable display devices, such as an LCD screen, an LED display, an OLED display, an electrophoretic display, a computer monitor, a touchscreen, a television, etc., one or more speakers, and so on.

[0062] In some embodiments, communications systems 426 can include any suitable hardware, firmware, and / or software for communicating information to computing device 350 (and, in some embodiments, over communication network 354 and / or any other suitable communication networks). For example, communications systems 426 can include one or more transceivers, one or more communication chips and / or chip sets, and so on. In a more particular example, communications systems 426 can include hardware, firmware, and / or software that can be used to establish a wired connection using any suitable port and / or communication standard (e.g., VGA, DVI video, USB, RS-232, etc.), Wi-Fi connection, a Bluetooth connection, a cellular connection, an Ethernet connection, and so on.

[0063] In some embodiments, memory 428 can include any suitable storage device or devices that can be used to store instructions, values, data, or the like, that can be used, for example, by processor 422 to control the one or more data acquisition systems 424, and / or receive data from the one or more data acquisition systems 424; to generate images from data; present content (e.g., data, images, a user interface) using a display; communicate with one or more computing devices 350; and so on. Memory 428 can include any suitable volatile memory, non-volatile memory, storage, or any suitable combination thereof. For example, memory 428 can include RAM, ROM, EPROM, EEPROM, other types of volatile memory, other types of non-volatile memory, one or more types of semi-volatile memory, one or more flash drives, one or more hard disks, one or more solid state drives, one or more optical drives, and so on. In some embodiments, memory 428 can have encoded thereon, or otherwise stored therein, a program for controlling operation of data source 302. In such embodiments, processor 422 can execute at least a portion of the program to generate images, transmit information and / or content (e.g., data, images, a user interface) to one or more computing devices 350, receive information and / or content from one or more computing devices 350,16QB\650053.01305\ 102412286.1MCW C2408650053.01305receive instructions from one or more devices (e.g., a personal computer, a laptop computer, a tablet computer, a smartphone, etc.), and so on.

[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 magnetic media (e.g., hard disks, floppy disks), optical media (e.g., compact discs, digital video discs, Blu-ray discs), semiconductor media (e.g., RAM, flash memory, EPROM, EEPROM), 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, or any suitable media that is fleeting and devoid of any semblance of permanence during transmission, and / or any suitable intangible media.

[0065] As used herein in the context of computer implementation, unless otherwise specified or limited, the terms “component / ’ “system,” “module,” “framework,” and the like are intended to encompass part or all of computer-related systems that include hardware, software, a combination of hardware and software, or software in execution. For example, a component may be, but is not limited to being, a processor device, a process being executed (or executable) by a processor device, an object, an executable, a thread of execution, a computer program, or a computer. By way of illustration, both an application running on a computer and the computer can be a component. One or more components (or system, module, and so on) may reside within a process or thread of execution, may be localized on one computer, may be distributed between two or more computers or other processor devices, or may be included within another component (or system, module, and so on).

[0066] In some implementations, devices or systems disclosed herein can be utilized or installed using methods embodying aspects of the disclosure. Correspondingly, description herein of particular features, capabilities, or intended purposes of a device or system is generally intended to inherently include disclosure of a method of using such features for the intended purposes, a method of implementing such capabilities, and a method of installing disclosed (or otherwise known) components to support these purposes or capabilities. Similarly, unless otherwise indicated or limited, discussion herein of any method of manufacturing or using a particular device or system, including installing the device or system,17QB\650053.01305\ 102412286.1MCW C2408650053.01305is intended to inherently include disclosure, as embodiments of the disclosure, of the utilized features and implemented capabilities of such device or system.

[0067] The present disclosure has described one or more preferred embodiments, and it should be appreciated that many equivalents, alternatives, variations, and modifications, aside from those expressly stated, are possible and within the scope of the invention.18QB\650053.01305\ 102412286.1

Claims

MOW C2408650053.01305CLAIMS1. A method for predicting pulmonary embolism (PE) clot types, comprising: receiving medical imaging data that depict one or more pulmonary arteries of a patient;segmenting one or more PE clots from the medical imaging data;extracting radiomic features from the segmented PE clots: andinputting the radiomic features to a trained machine learning model, generating classified feature data as an output, wherein the classified feature data indicate a prediction of a clot type for each of the segmented PE clots.

2. The method of claim 1. wherein the medical imaging data comprise computed tomography (CT) angiography images.

3. The method of claim 1, wherein the radiomic features comprise at least one of volume, maximum intensity, skewness, and kurtosis of the segmented PE clots.

4. The method of claim 1, wherein the machine learning model comprises a bilayered neural network with three fully connected layers and rectified linear unit (RELU) activation functions.

5. The method of claim 4, wherein the bi-layered neural network is trained to differentiate between pulmonary embolism clot types based on anatomical location within the pulmonary vasculature.

6. The method of claim 5, wherein the clot types comprise saddle, main, segmental, and subsegmental pulmonary embolism clots.

7. The method of claim 5. wherein the clot types comprise saddle, main, and combined segmental or subsegmental pulmonary embolism clots.19QB\650053.01305\102412286.1MCW C2408650053.013058. The method of claim 1. further comprising preprocessing the medical imaging data prior to segmenting the one or more PE clots, wherein the preprocessing comprises at least one of image normalization, noise reduction, or contrast enhancement.

9. The method of claim 1, wherein the medical imaging data comprise magnetic resonance angiography (MRA) images.

10. The method of claim 4, wherein the bi-layered neural network is trained using Bayesian optimization with a probability of improvement acquisition function to determine hyperparameters.

11. The method of claim 1 , wherein the classified feature data comprise probability values for each clot type.

12. The method of claim 1. further comprising generating a pulmonary embolism burden score based on the classified feature data.

13. A system for predicting pulmonary embolism (PE) clot types, comprising: a computing device comprising a processor and a memory, the memory storing a trained machine learning model;wherein the processor is configured to:receive medical imaging data that depict one or more pulmonary' arteries of a patient;segment one or more PE clots from the medical imaging data; extract radiomic features from the segmented PE clots; andinput the radiomic features to the trained machine learning model, generating classified feature data as an output, wherein the classified feature data indicate a prediction of a clot type for each of the segmented PE clots.

14. The system of claim 13, wherein the medical imaging data comprise at least one of computed tomography (CT) angiography images or magnetic resonance angiography (MRA) images.20QB\650053.01305\ 102412286.1MCW C2408650053.0130515. The system of claim 13, wherein the radiomic features comprise at least one of volume, maximum intensity, skewness, and kurtosis of the segmented PE clots.

16. The system of claim 13, wherein the trained machine learning model comprises a bi-layered neural network with three fully connected layers.

17. The system of claim 13, wherein the trained machine learning model is configured to differentiate between saddle, main, segmental, and subsegmental pulmonary embolism clots.

18. A non-transitory computer-readable medium stonng instructions that, when executed by a processor, cause the processor to:receive medical imaging data that depict one or more pulmonary arteries of a patient; segment one or more PE clots from the medical imaging data;extract radiomic features from the segmented PE clots; andinput the radiomic features to a trained machine learning model, generating classified feature data as an output, wherein the classified feature data indicate a prediction of a clot ty pe for each of the segmented PE clots.

19. wherein the trained machine learning model compnses a bi-layered neural network with three fully connected layers20. The non-transitory computer-readable medium of claim 18, wherein the trained machine learning model is configured to differentiate between saddle, main, segmental, and subsegmental pulmonary' embolism clots.21QB\650053.01305\ 102412286.1