Artificial intelligence driven method for generating white matter fiber tracts based on ct imagery simulation

Abstract

This disclosure relates to an artificial intelligence-driven method, apparatus, device, and medium for simulating and generating white matter fiber tracts based on CT images. The method includes: acquiring a training dataset, which includes CT images of patients with cerebral hemorrhage and diffusion tensor imaging data paired with the CT images; establishing a fiber tract model and generating CST anatomical labels in the diffusion tensor imaging space; mapping the CST anatomical labels in the diffusion tensor imaging space to the corresponding CT space using a linear registration algorithm; annotating the CT images with CST fiber tracking masks based on the CST anatomical labels in the CT space; training a CST prediction model based on the training dataset; and inputting the CT image to be predicted into the CST prediction model to generate a CST prediction mask for the CT image to be predicted. According to the technical solution of this disclosure, damage assessment of key white matter tracts in patients with cerebral hemorrhage can be achieved under resource-constrained conditions.

Description

Technical Field

[0001] This disclosure relates to the field of neuroimaging technology, and in particular to an artificial intelligence-driven method, apparatus, device, and medium for generating white matter fiber bundles based on CT image simulation. Background Technology

[0002] Intracerebral hemorrhage is a major subtype of stroke. In the neuroimaging context, the clinical management of intracerebral hemorrhage is undergoing a major technological transformation from "imaging diagnosis" to "precise prediction of functional prognosis." In the diagnosis and treatment of intracerebral hemorrhage, the corticospinal tract (CST) plays an important role in assessing the severity of the condition, guiding precise treatment, and predicting rehabilitation outcomes.

[0003] Currently, while magnetic resonance diffusion tensor tractography (MRI) can depict specific white matter fiber tracts such as CST, this method is time-consuming and costly, requiring high-field MRI (Magnetic Resonance Imaging) scanners and long scan times. These resources are limited in some scenarios. Furthermore, patients with cerebral hemorrhage are usually in extremely unstable conditions within 24-72 hours of onset and often require various electronic monitoring devices, making MRI scans in a high-risk and difficult clinical setting. Summary of the Invention

[0004] To address the aforementioned technical issues, this disclosure provides an artificial intelligence-driven method, apparatus, device, and medium for generating white matter fiber bundles based on CT image simulation.

[0005] In a first aspect, embodiments of this disclosure provide an artificial intelligence-driven method for generating white matter fiber bundles based on CT image simulation, comprising: Obtain the training dataset; wherein the training data includes CT images of patients with cerebral hemorrhage and diffusion tensor imaging data paired with the CT images; Establish a fiber bundle model and generate CST anatomical labels in the diffusion tensor imaging space; The CST anatomical labels in the diffusion tensor imaging space are mapped to the corresponding CT space using a linear registration algorithm. The CST fiber tracking mask is marked in the CT image according to the CST anatomical label in the CT space; Train the CST prediction model based on the training dataset; The CT image to be predicted is input into the CST prediction model to generate a CST prediction mask for the CT image to be predicted.

[0006] Secondly, embodiments of this disclosure provide an artificial intelligence-driven device for generating white matter fiber bundles based on CT image simulation, comprising: The acquisition module is used to acquire the training dataset; wherein, the training data includes CT images of patients with cerebral hemorrhage and diffusion tensor imaging data paired with the CT images; The fiber bundle tracing module is used to build fiber bundle models and generate CST anatomical labels in the diffusion tensor imaging space; The spatial registration module is used to map the CST anatomical labels in the diffusion tensor imaging space to the corresponding CT space using a linear registration algorithm. The annotation module is used to annotate the CST fiber tracking mask in the CT image based on the CST anatomical label in the CT space; The training module is used to train the CST prediction model based on the training dataset; The prediction module is used to input the CT image to be predicted into the CST prediction model and generate a CST prediction mask for the CT image to be predicted.

[0007] Thirdly, embodiments of this disclosure provide an electronic device, including: a processor; a memory for storing executable instructions of the processor; the processor being configured to read the executable instructions from the memory and execute the instructions to implement the artificial intelligence-driven method for generating white matter fiber bundles based on CT image simulation described in the first aspect above.

[0008] Fourthly, embodiments of this disclosure provide a computer-readable storage medium storing a computer program that, when executed by a processor, implements the artificial intelligence-driven method for generating white matter fiber bundles based on CT image simulation described in the first aspect.

[0009] The technical solution provided in this disclosure has the following advantages compared with the prior art: According to the technical solution of this disclosure, a deep learning model capable of automatically segmenting the corticospinal tract directly on CT is trained using paired CT and DTI data. This enables the segmentation of the corticospinal tract using only conventional CT, achieving high-precision segmentation on low-contrast CT images, improving the clinical accessibility of advanced neuroimaging assessment. The "software algorithm" replaces the expensive and time-consuming DTI-MRI examination, enabling the assessment of critical white matter tract damage in patients with cerebral hemorrhage under resource-constrained conditions such as a lack of advanced imaging equipment. Furthermore, it can generate fiber tract integrity indicators with significant clinical predictive value, achieving an innovation in the paradigm of emergency imaging diagnosis of cerebral hemorrhage. Attached Figure Description

[0010] The accompanying drawings, which are incorporated in and form a part of this specification, illustrate embodiments consistent with this disclosure and, together with the description, serve to explain the principles of this disclosure.

[0011] To more clearly illustrate the technical solutions in the embodiments of this disclosure or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0012] Figure 1 This is a flowchart illustrating an artificial intelligence-driven method for generating white matter fiber bundles based on CT image simulation, as provided in an embodiment of this disclosure. Detailed Implementation

[0013] To better understand the above-mentioned objectives, features, and advantages of this disclosure, the solutions disclosed herein will be further described below. It should be noted that, unless otherwise specified, the embodiments and features described herein can be combined with each other.

[0014] Numerous specific details are set forth in the following description in order to provide a full understanding of this disclosure, but this disclosure may also be implemented in other ways different from those described herein; obviously, the embodiments in the specification are only some, and not all, of the embodiments of this disclosure.

[0015] Figure 1 This is a flowchart illustrating an artificial intelligence-driven method for generating white matter fiber bundles based on CT image simulation, as provided in an embodiment of this disclosure. The method provided in this embodiment can be executed by an artificial intelligence-driven device for generating white matter fiber bundles based on CT image simulation. This device can be implemented in software and / or hardware and can be integrated into any electronic device with computing capabilities.

[0016] like Figure 1 As shown in the embodiments of this disclosure, the artificial intelligence-driven method for generating white matter fiber bundles based on CT image simulation may include: Step 101: Obtain the training dataset.

[0017] The training data includes CT (Computed Tomography) images of patients with cerebral hemorrhage and diffusion tensor imaging data paired with the CT images.

[0018] In one embodiment of this disclosure, obtaining the training dataset includes: processing diffusion tensor imaging data using a deep learning workflow to synthesize distortion-free b0 images, and performing artifact and motion correction on the diffusion tensor imaging data. The training dataset uses data from patients with cerebral hemorrhage, including paired NCCT (non-contrast Computed Tomography), T1-weighted MRI, and high-directional DTI (Diffusion Tensor Imaging). After obtaining the training data, the DTI preprocessing includes: for data with only a single phase encoding direction, synthesizing distortion-free b0 images using the PreQual deep learning workflow, and performing artifact and motion correction using the FSL toolkit.

[0019] Step 102: Generate the gold standard label (Ground Truth) for the training data.

[0020] In this embodiment, a fiber bundle model is established, and CST anatomical labels are generated in the diffusion tensor imaging space to achieve fiber bundle tracing. Then, the CST anatomical labels in the diffusion tensor imaging space are mapped to the corresponding CT space through a linear registration algorithm. Based on the CST anatomical labels in the CT space, CST fiber tracking masks are marked in the CT images.

[0021] As an example, fiber tract tracing includes: establishing a fiber tract model using the Slicer DMRI module of the 3D Slicer software; and generating anatomical labels for the CST within the patient's native space based on the UKFT algorithm. Spatial registration includes: mapping the CST labels in DTI space to the corresponding CT space using the 3DSlicer Elastix linear registration algorithm to ensure spatial consistency of the training data.

[0022] Step 103: Train the CST prediction model based on the training dataset.

[0023] In this embodiment, the CST prediction model adopts an adaptive nnU-Net framework. As an example, to address the problems of blurred white matter boundaries and low signal-to-noise ratio in CT images, the model adopts specific parameter configurations, specifically a 3D full-resolution U-Net structure containing 6 encoder blocks and 5 decoder blocks, with a convolution kernel size of 3×3×3.

[0024] The CST prediction model was trained using the training dataset, including 5-fold cross-validation with 80 pairs of paired data. The training employed a joint loss function of Dice and cross-entropy.

[0025] This breaks the traditional technical limitation that "CT can only see hematomas, while MRI DTI can see nerve fiber bundles." By constructing a cross-modal inference model based on nnU-Net, this model does not rely on the DTI images themselves, but learns from a large number of paired "CT-DTI" data to extract weak spatial context features in CT images. Thus, it can achieve automated and high-precision segmentation of the anatomically "invisible" corticospinal tract (CST) on extremely low-contrast NCCT images.

[0026] Step 104, Model Inference Application.

[0027] In this embodiment, the CT image to be predicted is input into the CST prediction model to generate a CST prediction mask for the CT image to be predicted, so as to segment the corticospinal tract using conventional CT.

[0028] According to the technical solution of this disclosure, a deep learning model capable of automatically segmenting the corticospinal tract directly on CT is trained using paired CT and DTI data. This enables the segmentation of the corticospinal tract using only conventional CT, achieving high-precision segmentation on low-contrast CT images and improving the clinical accessibility of advanced neuroimaging assessment. The "software algorithm" replaces the expensive and time-consuming DTI-MRI examination, enabling damage assessment of key white matter tracts in patients with cerebral hemorrhage under resource-constrained conditions such as a lack of advanced imaging equipment. Furthermore, addressing the problem that traditional DTI fiber tractography often fails (typically with a failure rate of around 10%) around hematomas and edema (low anisotropy regions), this solution's model can still locate the CST even in cases of severe anatomical deformities such as midline shift. Compared to physical modeling relying on water molecule diffusion signals, the deep learning model, by learning spatial context information, exhibits stronger robustness at the complex pathophysiological environment of hematoma edges, providing anatomical inferences that traditional methods cannot obtain, thus solving the technical problem of traditional DTI methods failing in lesion areas. Compared to the traditional atlas overlay method, which is problematic because cerebral hemorrhage can cause significant midline shift and anatomical distortion, this approach uses deep learning to learn the spatial distribution patterns of CSTs under conditions of tissue compression, displacement, and low anisotropy at the lesion margin. Even when significant changes occur in brain structure, it can still accurately locate CSTs, thus solving the problem of large registration bias in the fixed atlas method.

[0029] Based on the above embodiments, the method further includes: annotating hematoma masks in CT images, training a hematoma prediction model based on a training dataset, and training the hematoma prediction model in parallel for automatic segmentation of hematoma lesions on CT. Optionally, the hematoma prediction model uses the same training framework, and the training is supervised by manually segmenting CT images. Inference application: The trained model is applied to diagnostic CT scans of patients in the MISTIE III trial to generate predictive masks for CST and hematoma.

[0030] In this embodiment, the model inference application further includes: inputting the CT image to be predicted into the hematoma prediction model to generate a hematoma prediction mask for the CT image to be predicted.

[0031] In one embodiment of this disclosure, a hematoma fiber bundle overlap index is determined on a spatial voxel based on a CST prediction mask and a hematoma prediction mask; a fiber bundle transection index is determined based on the detection of spatial continuity interruptions of the CST on an axial section. A multiple linear regression model is used to analyze the predictive power of the CST integrity index for 180-day and 365-day functional prognosis.

[0032] The CST integrity index comprises the hematoma-fiber tract overlap index and the fiber tract transection index. Key features are defined as follows: Hematoma-fiber tract overlap is defined as the overlap between the CST segmentation map and the hematoma segmentation map on a spatial voxel; tract tract transection is defined as the detection of a spatial discontinuity (i.e., a "break") in the CST on an axial plane. The CST integrity index reflects the degree of nerve damage, including signs such as hematoma overlap and fiber tract transection.

[0033] The regression analysis included: using a multiple linear regression model, after controlling for confounding factors such as age, gender, and hematoma volume, to analyze the predictive power of CST integrity indices for functional prognosis at 180 days and 365 days, in order to validate clinical prognosis prediction. For example, in a large clinical trial dataset (MISTIE III), the correlation between the above indices and the patients' acute and long-term neurological function recovery (such as NIHSS motor score and mRS score) was validated.

[0034] According to embodiments of this disclosure, long-term functional prognostic biomarkers with significant predictive power are provided, and a fiber tract integrity index with significant clinical predictive value is generated, achieving innovation in the emergency imaging diagnostic paradigm for intracerebral hemorrhage. Through multiple linear regression analysis of CT-simulated fiber tract image datasets, the "fiber tract transection" index extracted by the model predicts the 180-day NIHSS motor score and the 365-day mRS score, thereby predicting the patient's clinical prognosis and providing clinicians with an objective quantitative tool for accurately assessing the patient's rehabilitation potential after hemorrhage. Furthermore, this solution can be integrated into surgical navigation or screening systems. By assessing the compression / rupture status of the CST, it automatically screens out the patient subgroups most likely to benefit from minimally invasive surgery (e.g., patients with no CST rupture but significant compression). Based on the automatic segmentation results, intracerebral hemorrhage patients are stratified, and an automated recommendation system is provided regarding whether to recommend hematoma evacuation (such as the MISTIE procedure). Additionally, it integrates two parallel models—hematoma segmentation and CST prediction—and generates clinical decision support signals by calculating their three-dimensional spatial interaction, improving surgical success rate and resource utilization.

[0035] In one embodiment of this disclosure, the prediction model employs a Transformer architecture (such as Swin-UNETR, nnFormer) or a hybrid architecture (such as TransUNet) to more accurately infer the path of the CST based on distal anatomical landmarks (such as skull shape, midbrain location) by capturing long-range dependencies in medical images.

[0036] In one embodiment of this disclosure, the prediction model employs variants such as Auto-nnU-Net, which, through hyperparameter optimization (HPO) and neural architecture search (NAS), can automatically find the optimal network topology for a specific brain hemorrhage dataset (such as CT scans with different scanning protocols), thereby further improving the Dice coefficient.

[0037] In one embodiment of this disclosure, the image processing logic employs cross-modal synthesis. Optionally, generative synthesis from CT to MRI: Generative adversarial networks or diffusion models are used to convert non-enhanced CT into synthetic MRI. Synthetic MRI possesses higher tissue contrast and can directly utilize established MRI white matter segmentation tools, exhibiting high realism in guided image registration and anatomical recognition. Optionally, direct prediction of diffusion indices: In addition to predicting the spatial location of the CST, scalar maps, such as fractional anisotropy (FA) maps, can be directly synthesized from CT images. Deep learning models are used to directly synthesize missing image modalities or quantization parameters to provide richer microstructural integrity information than binary segmentation maps.

[0038] In one embodiment of this disclosure, damage assessment features and biomarkers are expanded. Optionally, weighted damage burden (wCST-LL): a weighted CST damage burden is introduced, assigning different weights to the contribution of the CST to functional recovery in different segments (e.g., internal capsule vs. corona radiata) to improve predictive accuracy. Optionally, integrated clinical and radiomics features: AI-automated imaging features are combined with the patient's baseline clinical data (e.g., age, NIHSS, level of consciousness) or radiomics features of the tissue surrounding the hematoma to construct a multimodal nomogram for more comprehensive prognostic assessment.

[0039] In one embodiment of this disclosure, assisted surgical navigation is provided. Real-time surgical planning and augmented reality: Prognostic prediction is integrated with the neurosurgical navigation system. AI-segmented 3D models of the CST assist in selecting the approach for hematoma evacuation. An augmented reality system displays the CST location in real-time during the operation to maximize the protection of functional areas. Automated decision support system: A decision support framework specifically designed for primary care hospitals is developed. AI automatically analyzes CT scans and outputs risk level scores to help non-expert physicians quickly determine whether patients need to be referred to advanced stroke centers for interventional treatment.

[0040] This disclosure also provides a schematic diagram of the structure of an AI-driven device for generating white matter fiber bundles based on CT image simulation. The AI-driven device for generating white matter fiber bundles based on CT image simulation includes: The acquisition module is used to acquire the training dataset; wherein, the training data includes CT images of patients with cerebral hemorrhage and diffusion tensor imaging data paired with the CT images; The fiber bundle tracing module is used to build fiber bundle models and generate CST anatomical labels in the diffusion tensor imaging space; The spatial registration module is used to map the CST anatomical labels in the diffusion tensor imaging space to the corresponding CT space using a linear registration algorithm. The annotation module is used to annotate the CST fiber tracking mask in the CT image based on the CST anatomical label in the CT space; The training module is used to train the CST prediction model based on the training dataset; The prediction module is used to input the CT image to be predicted into the CST prediction model and generate a CST prediction mask for the CT image to be predicted.

[0041] The AI-driven apparatus for generating white matter fiber bundles based on CT image simulation provided in this disclosure can execute any AI-driven method for generating white matter fiber bundles based on CT image simulation provided in this disclosure, and has the corresponding functional modules and beneficial effects for executing the method. Content not described in detail in the apparatus embodiments of this disclosure can be referred to the descriptions in any method embodiments of this disclosure.

[0042] This disclosure also provides an electronic device including one or more processors and a memory. The processor may be a central processing unit (CPU) or other processing unit with data processing capabilities and / or instruction execution capabilities, and may control other components in the electronic device to perform desired functions. The memory may include one or more computer program products, which may include various forms of computer-readable storage media, such as volatile memory and / or non-volatile memory. Volatile memory may include, for example, random access memory (RAM) and / or cache memory. Non-volatile memory may include, for example, read-only memory (ROM), hard disk, flash memory, etc. One or more computer program instructions may be stored on the computer-readable storage medium, and the processor may execute the program instructions to implement the methods of the embodiments of this disclosure above and / or other desired functions. Various contents such as input signals, signal components, and noise components may also be stored in the computer-readable storage medium.

[0043] In one example, the electronic device may also include input and output devices, which are interconnected via a bus system and / or other forms of connection. Furthermore, the input device may include, for example, a keyboard, a mouse, etc. The output device can output various information to the outside, including determined distance information, direction information, etc. The output device may include, for example, a display, a speaker, a printer, and a communication network and its connected remote output devices, etc. In addition, depending on the specific application, the electronic device may include any other suitable components such as a bus, input / output interfaces, etc.

[0044] In addition to the methods and apparatus described above, embodiments of this disclosure may also be computer program products, which include computer program instructions that, when executed by a processor, cause the processor to perform any of the methods provided in the embodiments of this disclosure.

[0045] Computer program products can be written in any combination of one or more programming languages ​​to perform the operations of embodiments of this disclosure. The programming languages ​​include object-oriented programming languages ​​such as Java and C++, as well as conventional procedural programming languages ​​such as C or similar languages. The program code can be executed entirely on a user's computing device, partially on a user's computing device, as a standalone software package, partially on a user's computing device and partially on a remote computing device, or entirely on a remote computing device or server.

[0046] Furthermore, embodiments of this disclosure may also be computer-readable storage media storing computer program instructions that, when executed by a processor, cause the processor to perform any of the methods provided in the embodiments of this disclosure.

[0047] Computer-readable storage media may take the form of any combination of one or more readable media. A readable medium may be a readable signal medium or a readable storage medium. A readable storage medium may, for example, include, but is not limited to, electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatuses, or devices, or any combination thereof. More specific examples of readable storage media (a non-exhaustive list) include: electrical connections having one or more wires, portable disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof.

[0048] It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0049] The above description is merely a specific embodiment of this disclosure, enabling those skilled in the art to understand or implement it. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this disclosure. Therefore, this disclosure is not to be limited to the embodiments described herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. An artificial intelligence-driven method for generating white matter fiber bundles based on CT image simulation, characterized in that, The method includes: Obtain the training dataset; wherein the training data includes CT images of patients with cerebral hemorrhage and diffusion tensor imaging data paired with the CT images; Establish a fiber bundle model and generate CST anatomical labels in the diffusion tensor imaging space; The CST anatomical labels in the diffusion tensor imaging space are mapped to the corresponding CT space using a linear registration algorithm. The CST fiber tracking mask is marked in the CT image according to the CST anatomical label in the CT space; Train the CST prediction model based on the training dataset; The CT image to be predicted is input into the CST prediction model to generate a CST prediction mask for the CT image to be predicted.

2. The method as described in claim 1, characterized in that, The acquisition of the training dataset includes: The diffusion tensor imaging data is processed using a deep learning process to synthesize a distortion-free b0 image; Artifact and motion corrections are applied to the diffusion tensor imaging data.

3. The method as described in claim 1, characterized in that, The CST prediction model adopts an adaptive nnU-Net framework, configured as a 3D full-resolution U-Net structure, containing 6 encoder blocks and 5 decoder blocks, with a convolutional kernel size of 3×3×3; The training of the CST prediction model based on the training dataset includes: Five-fold cross-validation was performed using 80 pairs of paired data; training employed a joint loss function of Dice and cross-entropy.

4. The method as described in claim 1, characterized in that, The method further includes: Hematoma masks are annotated in the CT images, and a hematoma prediction model is trained based on the training dataset; The CT image to be predicted is input into the hematoma prediction model to generate a hematoma prediction mask for the CT image to be predicted.

5. The method as described in claim 4, characterized in that, The method further includes: Based on the CST prediction mask and the hematoma prediction mask, the hematoma fiber bundle overlap index on the spatial voxel is determined; Based on the spatial continuity interruption of CST detected on the axial section, the fiber bundle transection index is determined; wherein, the hematoma fiber bundle overlap index and the fiber bundle transection index constitute the CST integrity index. A multiple linear regression model was used to analyze the predictive power of the CST integrity index on 180-day and 365-day functional prognosis.

6. An artificial intelligence-driven device for generating white matter fiber bundles based on CT image simulation, characterized in that, include: The acquisition module is used to acquire the training dataset; wherein, the training data includes CT images of patients with cerebral hemorrhage and diffusion tensor imaging data paired with the CT images; The fiber bundle tracing module is used to build fiber bundle models and generate CST anatomical labels in the diffusion tensor imaging space; The spatial registration module is used to map the CST anatomical labels in the diffusion tensor imaging space to the corresponding CT space using a linear registration algorithm. The annotation module is used to annotate the CST fiber tracking mask in the CT image based on the CST anatomical label in the CT space; The training module is used to train the CST prediction model based on the training dataset; The prediction module is used to input the CT image to be predicted into the CST prediction model and generate a CST prediction mask for the CT image to be predicted.

7. An electronic device, characterized in that, include: processor; Memory used to store the processor's executable instructions; The processor is configured to read the executable instructions from the memory and execute the instructions to implement the artificial intelligence-driven method for generating white matter fiber bundles based on CT image simulation as described in any one of claims 1-5.

8. A computer-readable storage medium, characterized in that, The storage medium stores a computer program, which, when executed by a processor, implements the artificial intelligence-driven method for generating white matter fiber bundles based on CT image simulation as described in any one of claims 1-5.

Images