A cone-beam CT image reconstruction method suitable for portable devices

Abstract

This invention discloses a cone-beam CT image reconstruction method suitable for portable devices, belonging to the field of CT image reconstruction technology. The method includes: weighted preprocessing of cone-beam CT projection data; one-dimensional frequency domain filtering; initialization of a heterogeneous computing environment based on OpenCL and adoption of the CL_MEM_ALLOC_HOST_PTR zero-copy memory allocation strategy; encapsulation of the filtered projection data into an image2d_t texture object to utilize the GPU hardware texture filtering unit; adaptive partitioning of the three-dimensional voxel space into cache matching sub-blocks according to the mobile GPU device parameters and execution of parallel back-projection calculations; and asynchronous pipelined processing of CPU data preloading and GPU back-projection calculations through a command queue supporting out-of-order execution. This invention achieves a speedup of over 44 times on low-frequency mobile GPUs through hardware texture acceleration sampling, zero-copy memory mapping, and heterogeneous pipeline optimization, reducing reconstruction time from 503.45 seconds to 11.3 seconds. It is also compatible with multiple hardware brands and suitable for portable industrial inspection and mobile medical imaging scenarios.

Description

Technical Field

[0001] This application relates to the field of CT image reconstruction technology, and in particular to a cone-beam CT image reconstruction method suitable for portable devices. Background Technology

[0002] Cone-beam CT uses a planar array detector, acquiring multi-slice projection data in a single rotational scan. Compared to parallel-beam or fan-beam CT systems, it offers significant advantages such as faster scanning speed, higher spatial resolution, and higher X-ray utilization, making it a crucial development direction for current CT technology. Throughout the development of cone-beam CT technology, achieving rapid and high-precision image reconstruction has remained a core research topic.

[0003] With the continuous increase in the number of array detector units and the sustained improvement in scanning speed, the amount of projection data generated by cone-beam CT systems per unit time has increased dramatically. Coupled with the computational complexity of 3D image reconstruction algorithms, the reconstruction process faces severe challenges due to massive data transmission and enormous computational demands. Existing technologies, primarily relying on traditional reconstruction schemes that use computer central processing units (CPUs) for serial calculations, can no longer meet the real-time reconstruction efficiency requirements of modern cone-beam CT systems, becoming a key bottleneck restricting the further development and application of this technology. Summary of the Invention

[0004] Therefore, it is necessary to provide a cone-beam CT image reconstruction method suitable for portable devices to address the aforementioned technical problems.

[0005] The following technical solution is adopted in this specification: This manual provides a cone-beam CT image reconstruction method suitable for portable devices, including: The three-dimensional voxel space and projection data of the object to be reconstructed are acquired; wherein the projection data is obtained by cone-beam CT scanning using a detector. The projection data is subjected to weighted sum filtering using the FDK algorithm to obtain filtered projection data, and the filtered projection data is encapsulated into an OpenCL texture object and stored in the hardware texture unit of the portable mobile device. Based on the hardware parameters of the portable mobile device, the three-dimensional voxel space of the object to be reconstructed is divided into sub-blocks that match the cache capacity of the portable mobile device. For each sub-block, based on the hardware texture unit of the OpenCL texture object, the coordinate mapping and bilinear interpolation calculation in the FDK algorithm back projection function are completed, and the calculation results of all projection angles are accumulated to the corresponding voxel; Each sub-block after the back-projection calculation is completed is superimposed, and the resulting three-dimensional voxel data is used as the reconstructed image.

[0006] Furthermore, through a zero-copy memory mapping mechanism, the projection data of the object to be reconstructed is allocated to the physical memory space shared by the CPU and GPU in the portable mobile device.

[0007] Furthermore, the zero-copy memory mapping mechanism is achieved by configuring the CL_MEM_ALLOC_HOST_PTR flag during the initialization of the OpenCL environment.

[0008] Furthermore, the weighting, filtering, and backprojection processes of the FDK algorithm are executed asynchronously and concurrently, specifically including: The FDK algorithm is broken down into a weighted kernel, a filtering kernel, and a backprojection kernel. Create an OpenCL command queue that supports out-of-order execution; Through the OpenCL command queue, while the GPU executes the back-projection kernel for the current projection angle, the CPU executes the loading, weighting, and filtering of projection data for the next projection angle in parallel, thus achieving overlap between computation and data transmission.

[0009] Furthermore, the acquisition of the filtered projection data specifically includes: A one-dimensional fast Fourier transform is performed on the weighted projection data in the horizontal direction to obtain the frequency domain projection data; Multiply the frequency domain projection data by the frequency domain filtering function to obtain the filtered frequency domain projection data; The filtered frequency domain projection data is subjected to an inverse fast Fourier transform to obtain the filtered spatial domain projection data.

[0010] Furthermore, dividing the three-dimensional voxel space of the object to be reconstructed into sub-blocks matching the cache capacity of the portable mobile device specifically includes: Query the CL_DEVICE_MAX_WORK_GROUP_SIZE and cache size parameters of portable mobile devices; Calculate the optimal sub-block size based on the parameters; Based on the optimal sub-block size, the three-dimensional voxel space is divided into multiple sub-blocks along the X, Y, and Z directions.

[0011] Furthermore, the portable mobile device includes laptops equipped with integrated graphics cards from NVIDIA, AMD, or Intel.

[0012] Furthermore, the backprojection calculation process for each voxel includes: For each voxel, calculate the floating-point coordinates of the voxel on the detector plane based on the geometric relationship of the current projection angle; Read the projection value after bilinear interpolation by the hardware texture unit from the OpenCL texture object; the projection value is the contribution of the current projection angle to the voxel; The reconstructed voxel is obtained by summing the contributions of all projection angles of the voxel.

[0013] Furthermore, the voxel size of the reconstructed image is 256×256×100 or higher.

[0014] This specification provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described cone-beam CT image reconstruction method suitable for portable devices.

[0015] The above-mentioned technical solutions adopted in this specification can achieve the following beneficial effects: In a cone-beam CT image reconstruction method for portable devices provided in this specification, the projection data filtered by the FDK algorithm is encapsulated into an OpenCL texture object and stored in the hardware texture unit of the portable mobile device; the three-dimensional voxel space of the object to be reconstructed is divided into sub-blocks that match the cache capacity of the portable mobile device; for each sub-block, FDK back projection calculation is performed based on the hardware texture unit of the OpenCL texture object, and the obtained three-dimensional voxel data is used as the reconstructed image.

[0016] Among them, the OpenCL heterogeneous computing framework is adopted to encapsulate the filtered projection data into texture objects and use texture units to complete the bilinear interpolation calculation in back projection. Through the collaborative innovation of key technologies such as the OpenCL heterogeneous computing framework, hardware texture acceleration, zero-copy memory optimization and asynchronous concurrency, the cone-beam CT reconstruction algorithm has been successfully implemented on portable devices with high efficiency. Significant technical progress has been made in terms of reconstruction speed, energy efficiency ratio, cross-platform compatibility and engineering practicality. Attached Figure Description

[0017] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings:

[0018] Figure 1 This specification provides a schematic flowchart of a cone-beam CT image reconstruction method suitable for portable devices. Figure 2 This is a schematic diagram of an object to be tested provided in this specification; Figure 3 This is a schematic diagram of a z=50 slice representing a reconstruction result provided in this specification. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of this specification clearer, the technical solutions of this application will be clearly and completely described below in conjunction with specific embodiments and corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments in this specification without creative effort are within the scope of protection of this application.

[0020] Currently, CT image reconstruction methods are mainly divided into two categories: algebraic iteration methods and back-projection reconstruction methods. Among them, back-projection reconstruction methods are widely used in practical systems due to their faster computation speed. In the field of commercial cone-beam CT, the most commonly used reconstruction algorithm is the FDK algorithm proposed by Feldkamp, ​​Davis, and Kress, which is based on a simple circular scanning trajectory. However, a circular trajectory does not satisfy the sufficient condition for accurate cone-beam reconstruction, making the FDK algorithm essentially an approximate reconstruction method. When the cone angle is small (e.g., less than 5 degrees), its reconstruction results are acceptable; but as the cone angle increases, increasingly severe artifacts appear in the reconstructed image, causing deviations from the real object structure. Although the FDK algorithm has theoretical inaccuracies, it is still widely used in industry due to its low computational complexity, high execution efficiency, and the ease of engineering implementation of the corresponding circular scanning trajectory.

[0021] The FDK algorithm mainly includes three steps: weighting of projected data, one-dimensional filtering, and back projection. Weighting: The projected data for each division is weighted using a function similar to cosine, appropriately correcting for the distance and angle difference between the volume pixels and the source point. The formula is as follows: Among them, among them, For projection angle Lower detector The original projection data at the coordinates; For projection angle Weighted projection data; The distance from the radiation source to the center of rotation (SID); These are the two-dimensional coordinates of the detector plane; These are the frequency domain filtering coefficients.

[0022] One-dimensional filtering: Perform a one-dimensional FFT transformation in the horizontal direction on the weighted values, as shown in formula (2), and then perform one-dimensional frequency domain filtering in the horizontal direction on the projection data at different projection angles, as shown in formula (3): in, These are the projection data after Fourier transform; Here is the filter function; This is the projected data after filtering and projection.

[0023] Back projection: Three-dimensional back projection is performed along the X-ray direction. The reconstructed volume pixel is the sum of the contributions of rays from all projection angles through that pixel, as shown in the following formula: Among them, formulas (4) and (5) map points on the volume pixel coordinate system to the virtual array detector coordinate system; formula (6) is the back projection formula; and These are the coordinates of the volume data in the virtual array detector coordinate system. Since the back projection calculation is independent at each rotation angle... In computers, cumulative calculations can be used.

[0024] At small cone angles, the image quality reconstructed by the FDK algorithm is comparable to that of 2D fan-beam reconstructed images. However, the FDK algorithm involves large computational and data transfer volumes. When reconstructing high-resolution images, high-performance server clusters are typically used to reduce reconstruction time, which limits the application scenarios of cone-beam CT for motion detection. Since the FDK reconstruction algorithm has parallelism, fully utilizing this parallelism is key to improving reconstruction speed.

[0025] Commonly used techniques for accelerating reconstruction include: It utilizes the concurrency and parallelism technologies of the CPU, where concurrency is achieved through blocking for task scheduling and parallelism is provided by a multi-core CPU. However, due to the limited number of CPU cores, the acceleration in image reconstruction is relatively low. Using dedicated ASIC chips has drawbacks such as being uneditable and having a long development time. The disadvantage of using FPGA is that the programming language is difficult; The drawback of using CUDA is that it can only be used with graphics cards from specific manufacturers.

[0026] The characteristics of various accelerated image reconstruction techniques are shown in Table 1: Table 1 Comparison of Commonly Used Accelerated Image Reconstruction Techniques However, performing image reconstruction on mobile laptops faces the following challenges: Poor mobile hardware compatibility: Existing mainstream acceleration solutions (such as CUDA) are bound to specific manufacturers' hardware, making it impossible for them to run on portable laptops equipped with GPUs from different brands, thus limiting the mobile application of CT reconstruction equipment.

[0027] High power consumption and heat dissipation bottlenecks: Traditional reconstruction acceleration solutions often rely on high-power workstation GPUs, while mobile platforms such as laptops are limited by thermal design power (TDP), which can easily trigger frequency reduction during high-intensity computing (such as back projection), resulting in a sharp drop in reconstruction efficiency.

[0028] High memory access latency: Mobile GPUs share bandwidth with CPUs, and memory bus widths are relatively narrow. Existing technologies, when handling large-scale volume pixel mapping in the FDK algorithm, suffer from severe I / O blocking due to frequent memory addressing, failing to realize the hardware's intended parallel performance.

[0029] In view of the various problems existing in the above-mentioned technologies and their unsuitability for mobile laptops, this technology uses OpenCL to accelerate the image reconstruction process. The technical solutions provided by the various embodiments of this application are described in detail below with reference to the accompanying drawings.

[0030] Figure 1 This is a schematic diagram of a cone-beam CT image reconstruction method suitable for portable devices, as described in this specification, which specifically includes the following steps: 1. Optimization solutions for video memory on mobile devices.

[0031] On mobile devices such as laptops, the GPU and CPU typically share a memory bus, and the video memory bandwidth is the key factor limiting the backprojection stage in the FDK algorithm.

[0032] Based on the projection data with reconstructed image, filtered projection data is obtained after filtering by the FDK algorithm described above.

[0033] To address the frequent non-aligned coordinate addressing during the FDK algorithm's backprojection process, this technique avoids using the traditional cl_mem buffer. Instead, it encapsulates the filtered projection data into an image2d_t texture object. The bilinear interpolation calculation in the backprojection is implemented using the hardware texture filtering unit integrated within the mobile GPU. This not only reduces the computational load on the ALU (Arithmetic Logic Unit) but also leverages the spatial locality of the texture cache to significantly reduce memory bandwidth pressure caused by frequent access to neighboring pixels.

[0034] To address the characteristic of mobile integrated graphics cards sharing physical memory with the CPU, this technology employs the CL_MEM_ALLOC_HOST_PTR flag to implement a zero-copy mapping mechanism during OpenCL environment initialization. This mechanism allows the GPU to directly access the physical space mapped to the host, eliminating the clEnqueueWriteBuffer data migration process required in discrete graphics card architectures, thereby circumventing the bandwidth limitations of the lower PCIe interface on mobile platforms.

[0035] 2. Design of heterogeneous computing pipeline.

[0036] Mobile platforms often struggle to maintain peak GPU frequencies for extended periods due to thermal limitations. This technology addresses this by refactoring the computational pipeline to achieve a more uniform, computationally intensive distribution.

[0037] Asynchronous Out-of-order Execution: This technique breaks down the FDK algorithm into three core kernels: weighting, filtering, and backprojection. By creating an OpenCL command queue that supports out-of-order execution, the GPU asynchronously executes the backprojection calculation for the current frame while the CPU processes the preloading of the next frame's projection data. This overlapping design ensures that the mobile GPU is always in a high-performance operating state, avoiding the energy loss caused by frequent frequency boosting and deboosting due to waiting for I / O.

[0038] Adaptive Workgroup Load Balancing: For hardware architectures with a limited number of compute units (CUs) in mobile GPUs, this technology employs a dynamic workgroup partitioning algorithm. During backprojection, instead of a simple fixed-size partitioning, the 3D voxel space is divided into sub-blocks matching the L1 / L2 cache capacity based on the mobile GPU's CL_DEVICE_MAX_WORK_GROUP_SIZE parameter. This ensures that the data required by each processing thread bundle (Warp / Wavefront) during coordinate mapping and accumulation resides entirely in the cache, significantly improving processing throughput on low-frequency mobile GPUs.

[0039] ③ Experiment and Results Analysis To more intuitively describe the reconstruction process, the entire CT image reconstruction process was conducted using computer simulation. First, a simulation model was constructed... Figure 2 The object to be detected (Phantom) is shown. The Phantom consists of a cube with an attenuation coefficient of 0.02, containing two spheres with attenuation coefficients of 0.1 and 1.0.

[0040] The scanning device parameters are as follows: C-arm turntable; Short Scan Trajectory; Flat panel detector; SID (Source-to-ISOcenter Distance) is 750. Reconstructed volume size: 256×256×100.

[0041] Table 2 shows a comparison of the running efficiency of the same image reconstruction algorithm FDK on CPU and the proposed method on GPU, with the speedup ratio based on CPU: Table 2 Comparison of Operating Efficiency According to experimental tests, in a mobile hardware environment (CPU: Core i7-12700H, GPU: RTX A1000Laptop), this technology exhibits excellent energy efficiency characteristics: 1. Calculation of energy efficiency ratio analysis Non-linear gain of speedup and clock speed: Experimental data shows that although the GPU’s operating frequency (1.192 GHz) is much lower than that of the CPU (4.70 GHz), its reconstruction speed (11.3s) is 44.55 times faster than that of the CPU (503.45s).

[0042] Technical Principle Analysis: This "low clock speed, high throughput" performance is attributed to the deep optimization of the back-projection stage in the FDK algorithm. Through OpenCL's texture mapping mechanism, dedicated hardware interpolation units within the GPU replace traditional software arithmetic logic operations. Given the bandwidth constraints of mobile platforms, the high locality of the texture cache significantly reduces frequent access to system memory, thus overcoming computational bottlenecks in a low-power environment.

[0043] 2. Mobile platform compatibility verification Hardware resource utilization: This technology uses the OpenCL framework and has successfully achieved rapid reconstruction of large-scale volumetric data (256×256×100) on a laptop-level GPU.

[0044] The benefits of zero-copy technology: In mobile architectures where the CPU and GPU share physical memory, this solution reduces unnecessary data copying overhead between the host and device through OpenCL memory mapping technology. This explains why extremely low data transfer latency can still be maintained on mobile platforms, enabling the entire reconstruction process to be completed within 1 second.

[0045] 3. Consistency analysis of reconstructed image quality Accuracy verification: such as Figure 3 As shown, the reconstruction results of the z=50 slice clearly demonstrate the spherical structure with different attenuation coefficients (0.1 and 1.0) in the simulated Phantom model, with sharp edges and no obvious geometric distortion.

[0046] The results analysis shows that this technology significantly improves the reconstruction speed of mobile platforms while fully preserving the mathematical accuracy of the FDK algorithm. The back-projection process optimized through bilinear interpolation sampling not only accelerates computation but also ensures the smoothness of volume pixel mapping to the detector coordinate system, demonstrating the practical value of this scheme in portable industrial inspection or medical auxiliary diagnosis.

[0047] like Figure 1 The cone-beam CT rapid image reconstruction method shown is suitable for portable devices and has the following advantages: Solving the problem of hardware platform universality: By building a heterogeneous computing architecture through OpenCL, it can break through the limitations of graphics cards from specific manufacturers and achieve unified and efficient reconstruction on mobile GPUs, CPUs and FPGAs of different brands (NVIDIA, AMD, Intel), thereby reducing the hardware cost of system deployment.

[0048] Addressing performance degradation issues in low-power platforms: Targeting the characteristics of low-frequency and heat-prone mobile GPUs, this technology aims to provide a highly energy-efficient pipeline scheduling scheme. By optimizing the kernel operator execution sequence, it achieves a very high speedup ratio (e.g., more than 40 times) while maintaining low power consumption, meeting the needs of on-site reconstruction in portable devices.

[0049] Solving the data transfer bottleneck under shared memory: By using OpenCL's texture mapping mechanism and zero-copy technology, the data access path in the voxel-level back projection process is optimized, and the computational load is reduced by using hardware texture interpolation units, thereby solving the problem of computational pauses caused by insufficient video memory bandwidth on mobile platforms.

[0050] It can achieve the following beneficial effects: (1) Excellent energy efficiency ratio and adaptability to mobile platforms This technology addresses the issue of cone-beam CT reconstruction's heavy reliance on high-performance workstations through deep optimization of heterogeneous computing operators. Experimental data shows that on a mobile GPU (RTX A1000 Laptop) with an operating frequency of only 1.192 GHz, this technology achieves a speedup of up to 44.55 times, significantly reducing reconstruction time from 503.45 seconds to 11.3 seconds.

[0051] Addressing the limitations of mobile GPU memory bandwidth and low thermal design power (TDP), this technology optimizes kernel execution efficiency to achieve extremely high processing throughput even at low clock speeds, meeting the real-time reconstruction needs of mobile terminals such as laptops.

[0052] (2) Memory access optimization based on hardware characteristics To address the issue of high data throughput in the back-projection stage of the FDK algorithm, this technique utilizes OpenCL for refined memory management: Texture mapping accelerates sampling: This technique maps projected data to texture objects and utilizes the sampler built into the mobile GPU hardware to perform bilinear interpolation calculations. This not only reduces the burden on the general-purpose computing unit (ALU) but also significantly reduces memory access latency through the spatial locality characteristics of the texture cache.

[0053] Zero-copy data exchange: By leveraging the physical architecture of shared memory between the CPU and GPU of mobile platforms, unnecessary data copying overhead is reduced through a memory mapping mechanism, thus solving the time loss in data transmission in traditional solutions.

[0054] (3) Broad-spectrum cross-platform compatibility and low-cost deployment This technology uses the OpenCL standardized interface, completely overcoming the hardware limitations of existing technologies. Eliminate vendor lock-in: Unlike CUDA, which can only run on graphics cards from specific manufacturers, this technology is widely compatible with GPUs from mainstream brands such as NVIDIA, AMD, and Intel, as well as CPU and FPGA hardware architectures.

[0055] Lowering the system threshold: Users do not need to purchase expensive application-specific integrated circuits (ASICs) or specific high-performance servers. They can use existing ordinary laptops to complete the reconstruction of high-resolution volume data of 256 or more, which greatly reduces the hardware selection cost of CT system on-site inspection.

[0056] (4) Engineering practicality brought by heterogeneous parallel pipelines This solution improves the stability of engineering implementation by refactoring the execution logic of the FDK algorithm. Deep overlap between computation and I / O: Through the OpenCL asynchronous command queue, parallel pipelined operations of projection data loading, pre-weighting, filtering and back-projection are implemented to ensure that hardware resources are always in a state of high-efficiency saturation.

[0057] Balancing speed and accuracy: While achieving a speedup of over 44 times, this technology fully preserves the physical accuracy of the FDK algorithm. Simulation results show that the reconstructed slice images have clear structures and high contrast, fully verifying the practical value of this technology in scenarios with stringent requirements for accuracy and timeliness, such as industrial flaw detection and portable medical imaging.

[0058] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

Claims

1. A cone-beam CT image reconstruction method suitable for portable devices, characterized in that, include: The three-dimensional voxel space and projection data of the object to be reconstructed are acquired; wherein the projection data is obtained by cone-beam CT scanning using a detector. The projection data is allocated to the physical memory space shared by the CPU and GPU of the portable mobile device through a zero-copy memory mapping mechanism; the projection data is then subjected to weighted sum filtering processing of the FDK algorithm to obtain filtered projection data, and the filtered projection data is encapsulated into an OpenCL texture object and stored in the hardware texture unit of the portable mobile device. Based on the hardware parameters of the portable mobile device, the three-dimensional voxel space of the object to be reconstructed is divided into sub-blocks that match the cache capacity of the portable mobile device. For each sub-block, based on the hardware texture unit of the OpenCL texture object, the coordinate mapping and bilinear interpolation calculation in the FDK algorithm back projection function are completed, and the calculation results of all projection angles are accumulated to the corresponding voxel; Each sub-block after the back-projection calculation is completed is superimposed, and the resulting three-dimensional voxel data is used as the reconstructed image.

2. The cone-beam CT image reconstruction method for portable devices as described in claim 1, characterized in that, The weighting, filtering, and backprojection processes of the FDK algorithm are executed asynchronously and concurrently, specifically including: The FDK algorithm is broken down into a weighted kernel, a filtering kernel, and a backprojection kernel. Create an OpenCL command queue that supports out-of-order execution; Through the OpenCL command queue, while the GPU executes the back-projection kernel for the current projection angle, the CPU executes the loading, weighting, and filtering of projection data for the next projection angle in parallel, thus achieving overlap between computation and data transmission.

3. The cone-beam CT image reconstruction method for portable devices as described in claim 1, characterized in that, The zero-copy memory mapping mechanism is achieved by setting the CL_MEM_ALLOC_HOST_PTR flag during OpenCL environment initialization.

4. The cone-beam CT image reconstruction method for portable devices as described in claim 1, characterized in that, The acquisition of the filtered projection data specifically includes: A one-dimensional fast Fourier transform is performed on the weighted projection data in the horizontal direction to obtain the frequency domain projection data; Multiply the frequency domain projection data by the frequency domain filtering function to obtain the filtered frequency domain projection data; The filtered frequency domain projection data is subjected to an inverse fast Fourier transform to obtain the filtered spatial domain projection data.

5. The cone-beam CT image reconstruction method for portable devices as described in claim 1, characterized in that, The step of dividing the three-dimensional voxel space of the object to be reconstructed into sub-blocks matching the cache capacity of the portable mobile device specifically includes: Query the CL_DEVICE_MAX_WORK_GROUP_SIZE and cache size parameters of portable mobile devices; Calculate the optimal sub-block size based on the parameters; Based on the optimal sub-block size, the three-dimensional voxel space is divided into multiple sub-blocks along the X, Y, and Z directions.

6. The cone-beam CT image reconstruction method for portable devices as described in claim 1, characterized in that, The portable mobile device includes laptops equipped with integrated graphics cards from NVIDIA, AMD, or Intel.

7. The cone-beam CT image reconstruction method for portable devices as described in claim 1, characterized in that, The backprojection calculation process for each voxel includes: For each voxel, calculate the floating-point coordinates of the voxel on the detector plane based on the geometric relationship of the current projection angle; Read the projection value after bilinear interpolation by the hardware texture unit from the OpenCL texture object; the projection value is the contribution of the current projection angle to the voxel; The reconstructed voxel is obtained by summing the contributions of all projection angles of the voxel.

8. The cone-beam CT image reconstruction method for portable devices as described in claim 1, characterized in that, The reconstructed image has a voxel size of 256×256×100 or larger.

9. A computer-readable storage medium, characterized in that, The storage medium stores a computer program, which, when executed by a processor, implements the method described in any one of claims 1 to 8.

Images