📚 Computerised Axial Tomography | 计算机轴向断层扫描
Computerised Axial Tomography (CAT), widely known as CT scanning, represents one of the most significant applications of computing in modern medicine. By combining X-ray technology with sophisticated computer algorithms, CAT scanners produce detailed cross-sectional images of the human body, enabling non-invasive diagnosis of a vast range of conditions. This article explores the principles, data processing, image reconstruction techniques, and the computational demands that lie at the heart of this life-saving technology.
计算机轴向断层扫描（CAT），通常称为 CT 扫描，代表了计算技术在现代医学中最重大的应用之一。通过将 X 射线技术与复杂的计算机算法相结合，CAT 扫描仪生成人体的详细横截面图像，从而能够对多种病症进行非侵入性诊断。本文探讨了这项拯救生命的技术背后的原理、数据处理、图像重建技术以及计算需求。
1. Introduction to Computerised Axial Tomography | 计算机轴向断层扫描简介
Computerised Axial Tomography is an imaging procedure that uses a rotating X-ray source and a set of detectors to capture multiple projections through a patient's body. A computer then processes these projections to construct cross-sectional slices, or tomograms, which reveal internal structures with exceptional clarity. The term "axial" originally referred to the plane of the acquired slices, although modern scanners can reconstruct images in any plane.
计算机轴向断层扫描是一种成像程序，它使用旋转的 X 射线源和一组探测器来捕获穿过患者身体的多个投影。然后计算机处理这些投影以构建横截面切片（即断层图像），从而异常清晰地显示内部结构。“轴向”一词最初是指所采集切片的平面，但现代扫描仪可以重建任意平面的图像。
Invented by Sir Godfrey Hounsfield in the early 1970s, the CAT scan revolutionised diagnostic medicine. For his work, Hounsfield shared the 1979 Nobel Prize in Physiology or Medicine. The core innovation was not the X-ray source itself, but the computational method that translated thousands of attenuation measurements into a meaningful image.
CAT 扫描由 Godfrey Hounsfield 爵士于 20 世纪 70 年代初发明，彻底改变了诊断医学。凭借这项工作，Hounsfield 分享了 1979 年诺贝尔生理学或医学奖。核心创新不在于 X 射线源本身，而在于将数千次衰减测量结果转化为有意义图像的计算方法。
2. The Basic Principle of CAT Scanning | CAT 扫描的基本原理
The fundamental principle behind CAT is the measurement of X-ray attenuation as the beam passes through tissue. Different tissues absorb X-rays to varying degrees: bone attenuates heavily, soft tissue moderately, and air very little. An X-ray tube emits a narrow, fan-shaped beam that traverses the body slice from many angles. Detectors positioned opposite the source record the intensity of the emergent radiation.
CAT 背后的基本原理是当 X 射线束穿过组织时对 X 射线衰减的测量。不同组织对 X 射线的吸收程度不同：骨骼衰减很大，软组织中等，空气很小。X 射线管发射一个狭窄的扇形束，从多个角度穿过身体切片。位于射线源对面的探测器记录出射辐射的强度。
sity and μ represents the linear attenuation coefficient along the path. Taking the logarithm yields the line integral of μ, which forms the raw projection data.
测得的强度 I 遵循比尔-朗伯定律，表示为 I = I₀ e⁻∫ μ ds，其中 I₀ 为初始强度，μ 为沿路径的线性衰减系数。取对数得到 μ 的线积分，即原始投影数据。
3. Data Acquisition: X-ray Projections | 数据采集：X 射线投影
During a scan, the gantry rotates around the patient, typically completing a full 360° rotation in under a second in modern systems. At hundreds of angular positions, the detector array captures a one-dimensional profile of attenuation. Each profile is called a projection, and the complete set of projections for one slice is termed a sinogram due to the sinusoidal traces left by individual points in the object.
扫描过程中，机架围绕患者旋转，在现代系统中通常不到一秒钟即可完成完整 360° 旋转。在数百个角度位置，探测器阵列捕获衰减的一维轮廓。每个轮廓称为一个投影，而一个切片的一组完整投影称为正弦图，这是因为物体中各个点留下的正弦轨迹而得名。
Data acquisition is governed by sampling considerations. To avoid aliasing, the number of projections must satisfy the Nyquist criterion relative to the spatial frequency content of the object. A typical scan collects around 1000 projections, each containing 700–1000 detector readings, generating approximately one million data points per slice.
数据采集受采样考虑因素控制。为避免混叠，投影数量必须满足相对于物体空间频率内容的奈奎斯特准则。一次典型扫描收集约 1000 个投影，每个投影包含 700–1000 个探测器读数，每个切片产生约一百万个数据点。
4. From Projections to Images: Image Reconstruction | 从投影到图像：图像重建
Reconstructing a cross-sectional image from its projections is a classic inverse problem. The goal is to determine the distribution of attenuation coefficients μ(x, y) within the slice. Mathematically, the relationship between the object function and its projections is described by the Radon transform. Inversion of the Radon transform lies at the heart of CAT image reconstruction.
从投影重建横截面图像是一个经典的反问题。目标是确定切片内衰减系数 μ(x, y) 的分布。从数学上讲，物体函数与其投影之间的关系由拉东变换描述。拉东变换的逆变换是 CAT 图像重建的核心。
The simplest reconstruction method is back projection, where each projection value is smeared back along the ray path and summed. However, simple back projection produces a blurred image with a characteristic 1/r blurring. To correct this, filtered back projection (FBP) is employed, applying a high-pass filter to the projections before back projection.
最简单的重建方法是反投影，即每个投影值沿射线路径涂抹回去并求和。然而，简单反投影会产生具有特征性 1/r 模糊的模糊图像。为了纠正这一点，采用了滤波反投影（FBP），在反投影之前对投影应用高通滤波器。
5. Filtered Back Projection Algorithm | 滤波反投影算法
The filtered back projection algorithm is the workhorse of clinical CT reconstruction. It consists of two main stages. First, each projection is convolved with a ramp filter in the frequency domain, which boosts high spatial frequencies and suppresses the 1/r blur. Second, the filtered projections are back-projected across the image matrix and summed. The result is a sharp, accurate representation of the original attenuation map.
滤波反投影算法是临床 CT 重建的主力。它包含两个主要阶段。首先，每个投影在频域中与斜坡滤波器进行卷积，增强高空间频率并抑制 1/r 模糊。其次，将滤波后的投影反投影到图像矩阵上并求和。结果是原始衰减图的清晰、准确表示。
μ(x, y) = ∫₀ᴾⁱ [ p(s, θ) * h(s) ] dθ
The mathematical expression above shows that the reconstruction μ(x, y) is obtained by integrating the convolution of the projection p with a filter kernel h over all angles θ. Modern implementations use the Fast Fourier Transform (FFT) to accelerate the convolution step, drastically reducing computation time.
上述数学表达式表明，重建 μ(x, y) 是通过对投影 p 与滤波器核 h 在所有角度 θ 上的卷积进行积分而得到的。现代实现使用快速傅里叶变换（FFT）来加速卷积步骤，从而大幅缩短计算时间。
6. Digitisation and the Role of Pixels and Voxels | 数字化及像素和体素的作用
ach pixel corresponds to the average linear attenuation coefficient within that region. Typically, a CT image has a matrix size of 512 × 512 pixels, although 1024 × 1024 matrices are increasingly common.
重建后的图像是一个二维像素阵列，每个像素代表切片中的一个小正方形区域。每个像素中存储的数值对应于该区域内的平均线性衰减系数。CT 图像的矩阵大小通常为 512 × 512 像素，但 1024 × 1024 矩阵也日益普遍。
When multiple contiguous slices are acquired, the two-dimensional pixels can be stacked to form a three-dimensional volume element, or voxel. The voxel's depth is determined by the slice thickness, which ranges from 0.5 mm to 10 mm. This voxel representation is essential for 3D rendering and multi-planar reformatting.
当采集多个连续切片时，二维像素可以堆叠形成三维体积元素，即体素。体素的深度由切片厚度决定，范围从 0.5 毫米到 10 毫米。这种体素表示对于三维渲染和多平面重组至关重要。
7. Hounsfield Units and Contrast Resolution | 亨氏单位与对比度分辨率
is defined such that water is 0 HU and air is -1000 HU. The formula is: HU = 1000 × (μ - μ_water) / μ_water. Typical values range from -1000 (air) to +3000 (dense bone), covering a dynamic range far beyond the human eye's capability.
CT 图像使用亨氏单位（HU）显示，这是一种将衰减系数相对于水进行归一化的定量标度。该标度的定义使水为 0 HU，空气为 -1000 HU。公式为：HU = 1000 × (μ - μ_water) / μ_water。典型值范围从 -1000（空气）到 +3000（致密骨骼），涵盖的动态范围远超人眼能力。
Because the raw data can distinguish differences as small as a few HU, CT offers excellent contrast resolution for soft tissues. Windowing is used to map a selected HU range to the grey levels of a display monitor, allowing clinicians to focus on specific tissue types.
由于原始数据能够区分小至几个 HU 的差异，CT 为软组织提供了出色的对比度分辨率。窗技术用于将选定的 HU 范围映射到显示器的灰度级，使临床医生能够专注于特定的组织类型。
8. Computer Processing Requirements | 计算机处理要求
ice from 1000 projections with 700 samples each, the FBP algorithm must perform roughly 10¹⁰ floating-point operations. Real-time reconstruction, where images appear within seconds of scanning, demands specialised hardware such as GPU arrays, field-programmable gate arrays (FPGAs), or dedicated image reconstruction processors.
CAT 扫描的计算负荷巨大。对于来自 1000 个投影、每个投影 700 个采样的单个 512 × 512 切片，FBP 算法必须执行大约 10¹⁰ 次浮点运算。实时重建要求图像在扫描后数秒内出现，这需要专用硬件，如 GPU 阵列、现场可编程门阵列（FPGA）或专用图像重建处理器。
Storage is equally challenging. A single CT study can generate 2000–5000 slices, each 0.5 MB in size, resulting in gigabytes of data per patient. Picture Archiving and Communication Systems (PACS) manage this data, employing lossless compression and rapid retrieval algorithms.
存储同样具有挑战性。一次 CT 检查可生成 2000–5000 个切片，每个切片大小为 0.5 MB，每位患者的数据量达到数 GB。影像归档与通信系统（PACS）通过采用无损压缩和快速检索算法来管理这些数据。
9. 3D Reconstruction and Visualisation | 三维重建与可视化
Beyond axial slices, CT data allows powerful three-dimensional visualisations. Volume rendering assigns colour and opacity to each voxel based on its HU value, enabling the creation of life-like 3D models of organs and bones. Surface rendering extracts iso-surfaces to show the boundaries of structures, a technique used in virtual colonoscopy and surgical planning.
除轴向切片外，CT 数据还能实现强大的三维可视化。体绘制根据每个体素的 HU 值为其分配颜色和透明度，从而生成逼真的器官和骨骼三维模型。面绘制提取等值面以显示结构边界，该技术用于虚拟结肠镜检查和手术规划。
These 3D techniques rely heavily on interpolation and segmentation algorithms. Interpolation fills gaps between slices to create an isotropic voxel grid, while segmentation classifies voxels into tissue types using thresholds, region growing, or machine learning. The resulting models can be rotated, sliced, and measured interactively.
这些三维技术严重依赖插值和分割算法。插值填充切片之间的间隙以创建各向同性体素网格，而分割则使用阈值、区域生长或机器学习将体素分类为组织类型。生成的模型可以交互式地旋转、剖切和测量。
10. Clinical Applications and Impact | 临床应用与影响
CAT scanning is indispensable in emergency medicine, neurology, oncology, and cardiology. It rapidly diagnoses strokes, haemorrhages, and traumatic injuries, guiding life-saving interventions. In oncology, CT precisely delineates tumour boundaries, monitors treatment response, and assists in radiotherapy planning. Cardiac CT can non-invasively image coronary arteries, reducing the need for catheter angiography.
CAT 扫描在急诊医学、神经病学、肿瘤学和心脏病学中不可或缺。它能快速诊断中风、出血和创伤，指导挽救生命的干预措施。在肿瘤学中，CT 精确描绘肿瘤边界、监测治疗反应并协助放射治疗计划。心脏 CT 能够无创地对冠状动脉成像，减少了对导管血管造影的需求。
The widespread adoption of CT has dramatically improved patient outcomes. However, it has also contributed to increased population radiation dose, prompting research into low-dose protocols and iterative reconstruction algorithms that maintain image quality while reducing exposure.
CT 的广泛应用显著改善了患者预后。但它也导致了人群辐射剂量的增加，促使研究人员开发低剂量方案和迭代重建算法，以在减少照射的同时保持图像质量。
11. Advantages and Limitations | 优势与局限性
CAT offers several key advantages: short scan times, excellent spatial and contrast resolution, and the ability to image bone, soft tissue, and blood vessels simultaneously. It is less sensitive to patient movement than MRI and can accommodate patients with metallic implants, although artefacts may occur.
CAT 具有几个关键优势：扫描时间短、出色的空间和对比度分辨率，以及能够同时对骨骼、软组织和血管成像。与 MRI 相比，它对患者运动的敏感性较低，并且可以接受带有金属植入物的患者，尽管可能会出现伪影。
Limitations include the use of ionising radiation, which carries a small but cumulative risk of cancer. Contrast resolution, while good, remains inferior to MRI for certain neurological conditions. Beam hardening and metal artefacts can degrade image quality, and obese patients may exceed the gantry's weight or diameter limits.
局限性包括使用电离辐射，这会带来较小但累积的癌症风险。对比度分辨率虽然良好，但在某些神经系统疾病方面仍不及 MRI。射束硬化和金属伪影会降低图像质量，而肥胖患者可能超出机架的重量或直径限制。
12. Future Developments and Computational Advances | 未来发展与计算进展
The future of CAT lies in faster, lower-dose, and more intelligent imaging. Iterative reconstruction algorithms, which model the scanning process and incorporate noise statistics, are replacing FBP. These methods reduce noise and artefacts, enabling dose reductions of up to 80%. Deep learning models now perform image denoising, artefact removal, and even direct image reconstruction from raw sinogram data.
CAT 的未来在于更快、更低剂量和更智能的成像。迭代重建算法对扫描过程建模并结合噪声统计，正在取代 FBP。这些方法减少噪声和伪影，使剂量降低高达 80%。深度学习模型现在可以执行图像去噪、伪影消除，甚至从原始正弦图数据直接重建图像。
Photon-counting detectors represent a hardware breakthrough, providing energy-resolved measurements that improve tissue differentiation and reduce noise. Combined with AI-driven analysis, these advances will transform CAT from an anatomical imaging tool into a quantitative, functional modality, opening new frontiers in personalised medicine.
光子计数探测器代表了一项硬件突破，提供能量分辨测量，改善组织区分度并降低噪声。结合人工智能驱动的分析，这些进步将把 CAT 从解剖成像工具转变为定量功能模态，为个性化医疗开辟新的前沿。
Published by TutorHao | Computer Science Revision Series | aleveler.com
更多咨询请联系16621398022（同微信）
或截图后用「扫一扫 → 相册」选择本图