Learning objectives / 学习目标
学完本课后,你应能:
- 从 Beer–Lambert attenuation 建立 projection data,并解释 CT 如何由多角度 line integrals 重建 cross-section;
- 由 time-of-flight、acoustic impedance 与 Doppler shift 计算 ultrasound depth、reflection 和 flow velocity;
- 用 Larmor precession、RF excitation、\(T_1/T_2\) relaxation 与 magnetic-field gradients 解释 MRI signal 和 spatial encoding;
- 区分 SPECT 的 single-gamma collimation 与 PET 的 coincidence line of response,并处理 activity decay/counting noise;
- 用 point-spread function、sampling、contrast-to-noise ratio、dose/activity 和 regularization 分析成像 tradeoffs;
- 在每一 modality 中分清 measured data、forward model、reconstruction、display 与 clinical interpretation。
Prerequisites / 先修知识
- exponential law、logarithm、line integral、Fourier/sampling 的基本直觉;
- wave speed、reflection、frequency、Doppler effect 与 acoustic impedance;
- magnetic moment、torque、angular frequency 与 electromagnetic induction;
- radioactive decay、photon detection、Poisson statistics;
- Physical Principles in Living Systems 中的 relaxation、transport、thermal noise 与 model boundary。
Scope and conventions / 范围与约定
- patient/object 固定在 laboratory frame;CT slice 采用 \((x,y)\),ray 用 source angle \(\theta\) 与 detector coordinate \(s\) 表示。
- X-ray linear attenuation coefficient \(\mu\ge0\),单位 \(\mathrm{m^{-1}}\);projection \(p=-\ln(I/I_0)\) 为 dimensionless line integral。
- ultrasound 轴向深度 \(z>0\) 指向体内;pulse-echo round trip 给 \(z=ct/2\)。Doppler angle \(\theta\) 是 beam 与 flow velocity 的夹角。
- MRI 主静磁场写作 \(\mathbf B_0=B_0\hat{\mathbf z}\);本课用 proton gyromagnetic ratio 的 magnitude,\(f_0=(\gamma/2\pi)B_0\)。
- nuclear medicine activity \(A\) 单位 Bq;PET/SPECT detected counts 不等于注射的 decay 总数,efficiency、attenuation、scatter、dead time 与 acquisition window 均影响数据。
- “dose”必须区分 absorbed dose、effective dose、CTDI/DLP 或 administered activity 等不同量;本课不把它们互相等同。
Notation and modeling contract / 符号与建模契约
| Symbol | Meaning | SI / convention |
|---|---|---|
| \(I_0,I\) | incident and transmitted X-ray intensity/count expectation | consistent detector units |
| \(\mu(\mathbf r)\) | linear attenuation coefficient | \(\mathrm{m^{-1}}\) |
| \(p(\theta,s)\) | logarithmic projection / Radon data | 1 |
| \(c,Z=\rho c\) | sound speed and acoustic impedance | \(\mathrm{m\,s^{-1}}\); \(\mathrm{kg\,m^{-2}\,s^{-1}}\) |
| \(f_0,\Delta f\) | transmitted and Doppler-shift frequencies | Hz |
| \(B_0,\gamma\) | main magnetic field and gyromagnetic ratio | T; \(\mathrm{rad\,s^{-1}\,T^{-1}}\) |
| \(M_z,M_{xy}\) | longitudinal and transverse magnetization | \(\mathrm{A\,m^{-1}}\) |
| \(T_1,T_2\) | longitudinal and transverse relaxation times | s |
| \(A,\lambda,T_{1/2}\) | activity, decay constant, half-life | Bq; \(\mathrm{s^{-1}}\); s |
| \(h(\mathbf r)\) | point-spread function (PSF) | normalization stated |
| \(\mathrm{CNR}\) | contrast-to-noise ratio | 1 |
1. Projection imaging and computed tomography / 投影成像与计算机断层
Knowledge ID: P14-U04-L02-K01
monoenergetic, narrow-beam idealization 下,一束 X ray 沿 path \(L\) 穿过 heterogeneous object:
\[I=I_0\exp\left[-\int_L\mu(\mathbf r)\,d\ell\right], \qquad p=-\ln\left(\frac{I}{I_0}\right)=\int_L\mu\,d\ell.\]
plain radiograph 把不同 depth 的 attenuation 叠成一个 projection,overlap 是结构信息损失,不是 detector blur 的同义词。real X-ray source 是 polychromatic spectrum;scatter、beam hardening、detector response、motion 与 finite focal spot 会使 simple exponential model 有偏差。
图 1 把 forward model 画成已知 ray path 上的 attenuation ledger;箭长与颜色不代表 photon energy 或精确强度,not to scale。
CT 采集许多 \(p(\theta,s)\);这些是 attenuation map 的 line integrals(Radon transform)。ideal continuous complete data 可由 inverse Radon transform / filtered backprojection 重建;迭代方法则显式比较 forward projection 与 measured data,并加入 noise model、constraints 或 regularization。少角度、metal、motion 与 truncation 会产生结构化 artifacts,而不只是“图像稍微模糊”。
图 2 说明 tomography 的多角度 constraint。右图不是由四个角度就能高质量重建的承诺;rays 仅为 schematic。
2. Ultrasound imaging and Doppler flow measurement / 超声成像与多普勒血流测量
Knowledge ID: P14-U04-L02-K02
piezoelectric transducer 发出 MHz pulse,也把 return echo 转成 voltage。若 tissue sound speed 近 constant \(c\),echo delay \(t\) 对应 reflector depth
\[z=\frac{ct}{2},\]
因 pulse 往返一次。normal-incidence pressure-amplitude reflection coefficient 为 \((Z_2-Z_1)/(Z_2+Z_1)\);intensity fraction 为
\[R_I=\left(\frac{Z_2-Z_1}{Z_2+Z_1}\right)^2, \qquad Z=\rho c.\]
实际 image brightness 还受 beam pattern、frequency-dependent attenuation、gain、speckle、refraction 与 scattering 影响。higher frequency 通常改善 wavelength-limited resolution,却更快 attenuate,降低 penetration。
图 3 明确 \(1/2\) 因子来自 round trip;real beam 有 finite width/focus,图中 single ray not to scale。
moving red cells scatter ultrasound with Doppler shift。transmit and receive on same probe gives approximately
\[\Delta f=\frac{2f_0v\cos\theta}{c}, \qquad v=\frac{c\Delta f}{2f_0\cos\theta}.\]
factor 2 来自 incident 与 backscattered Doppler shifts。\(\theta\to90^\circ\) 时 \(\cos\theta\to0\),axial Doppler 对 transverse flow insensitive;small angle error 在大 \(\theta\) 会导致大 velocity error。pulsed Doppler 还受 pulse-repetition frequency 和 aliasing 限制。
图 4 定义 Doppler angle;beam/flow arrows 是 vector directions,不代表 vessel wall stress。
3. Magnetic resonance imaging / 磁共振成像
Knowledge ID: P14-U04-L02-K03
在 \(\mathbf B_0\) 中,proton magnetic moments 的 thermal ensemble 产生小净 longitudinal magnetization \(M_0\),并以 Larmor frequency
\[\omega_0=\gamma B_0, \qquad f_0=\frac{\gamma}{2\pi}B_0\]
precess。near-resonant RF field 把 magnetization tip away from \(+z\);receiver coil 检测 transverse magnetization 变化产生的 voltage。MRI 不是“把单个 proton 拍成照片”,而是测许多 spins 的 coherent ensemble signal。
图 5 是 rotating magnetization vector,dashed ellipse 是 vector tip locus,不是 proton orbit;not to scale。
RF preparation 后,若 equilibrium longitudinal magnetization 为 \(M_0\),single-compartment Bloch recovery 的一般式为
\[M_z(t)=M_0+\bigl[M_z(0)-M_0\bigr]e^{-t/T_1}, \qquad M_{xy}(t)=M_{xy}(0)e^{-t/T_2}.\]
\(T_1\) 描述 longitudinal recovery,\(T_2\) 描述 irreversible transverse dephasing。saturation 或 ideal \(90^\circ\) preparation 令 \(M_z(0)=0\),于是 \(M_z=M_0(1-e^{-t/T_1})\);ideal inversion 令 \(M_z(0)=-M_0\),于是 \(M_z=M_0(1-2e^{-t/T_1})\)。field inhomogeneity 另产生更快 \(T_2^*\) decay。sequence 的 repetition time、echo time、flip angle 与 k-space trajectory 决定 tissue weighting,不可把“亮”直接等同某种 material quantity。
spatial encoding 依 gradients:\(B_z(\mathbf r)=B_0+\mathbf G\cdot\mathbf r\) 使 frequency/phase 随 position 变化;measured signal 是 spin density/contrast-weighted distribution 的 Fourier-like samples。undersampling、motion 与 field inhomogeneity 会造成 aliasing/ghosting/distortion。
图 6 对比 \(M_z(0)=0\) 的 normalized saturation/ideal-\(90^\circ\) recovery 与 transverse decay;共用横轴仅展示形状,不表示 \(T_1=T_2\)。真实 tissue 常含多 compartments/exchange。
4. PET and SPECT / PET 与 SPECT
Knowledge ID: P14-U04-L02-K04
SPECT 使用 gamma-emitting radiotracer。mechanical collimator 只允许有限 directions 的 photons 到达 gamma camera,以 rotating projections 重建 activity distribution;collimator improves localization at a severe sensitivity cost。
PET tracer emits positron;positron slows and annihilates with electron,产生近 opposite directions 的 two \(511\ \mathrm{keV}\) photons。detector coincidence 定义 line of response(LOR),不需 SPECT-style absorbing collimator。non-collinearity、positron range、scatter、random coincidences、attenuation、dead time 与 detector timing 限制 resolution/quantification。
图 7 采用 leading-order opposite-photon idealization 展示 PET coincidence geometry;physical non-collinearity 是 resolution limit。LOR 不是已知的 exact annihilation point,tomographic reconstruction 仍需很多 events。
图 8 显示只有 aligned photon 穿过 collimator septa;斜 photons 被吸收的具体路径为 schematic。
activity follows
\[A(t)=A_0e^{-\lambda t}=A_0\,2^{-t/T_{1/2}}.\]
biological clearance 与 physical decay 同时存在时 effective half-life 更短。detected counts 常近 Poisson:\(\sigma_N\approx\sqrt N\),relative noise \(\sigma_N/N\approx1/\sqrt N\);增加 counts 可降 relative noise,却需要 time、activity、sensitivity 或更宽 acceptance,并伴随 motion/dose/resolution tradeoffs。
5. Resolution, contrast, dose, and reconstruction tradeoffs / 分辨率、对比度、剂量与重建权衡
Knowledge ID: P14-U04-L02-K05
spatial resolution 可用 point-spread function(PSF)或 modulation transfer 描述,而不能只用 pixel width。linear shift-invariant idealization 中
\[g(\mathbf r)=\int h(\mathbf r-\mathbf r')f(\mathbf r')\,d\mathbf r'+n(\mathbf r).\]
measured image \(g\) 是 object \(f\) 与 PSF \(h\) 的 blur,再加 noise/artifacts。sampling 太粗会 alias;即使 pixels 很小,finite focal spot、detector aperture、acoustic pulse length、gradient encoding、positron range 或 collimator 仍可限制真实 resolution。
图 9 用 PSF 的 full width at half maximum(FWHM)示意 resolution;真实 PSF 可 anisotropic、space variant 或有 sidelobes。
contrast 是 two tissues/signals 的 difference;detectability 还取决于 noise、background texture、task 与 observer。简单 CNR 可写
\[\mathrm{CNR}=\frac{\lvert S_1-S_2\rvert}{\sigma},\]
但 \(\sigma\) 如何定义必须说明。Poisson-limited acquisition 中 CNR 常随 \(\sqrt N\) 增长:counts 加四倍只使 noise-limited CNR 约加倍。regularization/denoising 可减 variance,却引入 bias、blur 或 hallucinated-looking texture;任何 reconstruction 都应以 task-based validation、phantom、ground truth surrogate 与 uncertainty 检验。
ionizing modalities(X ray/CT/PET/SPECT)需 justification 与 optimization:临床问题能否由 exam 回答?在足以完成 diagnostic task 的前提下,如何减少不必要 exposure?MRI/diagnostic ultrasound 不用 ionizing radiation,但仍有 strong-field/RF/heating、implant、acoustic-output 等 modality-specific safety constraints。
图 10 是 Poisson-limited trend,不是 patient-specific dose–risk or guaranteed diagnostic-performance curve。vertical decision line 需由 validated clinical task/QA 决定。
Worked examples / 完整例题
Worked Example 1 — heterogeneous path 的 X-ray projection
已知: ray 依次穿过 \(L_1=0.100\ \mathrm m\)、\(\mu_1=20.0\ \mathrm{m^{-1}}\) 的 tissue 与 \(L_2=0.0500\ \mathrm m\)、\(\mu_2=10.0\ \mathrm{m^{-1}}\) 的 tissue;忽略 scatter。
研究对象与边界: monoenergetic narrow X-ray beam from source to one detector bin。
参考系/坐标: path coordinate \(\ell\) 从 source 指向 detector。
模型理由: piecewise-constant Beer–Lambert forward model。
\[p=\mu_1L_1+\mu_2L_2=(20.0)(0.100)+(10.0)(0.0500)=2.50,\]
\[\frac{I}{I_0}=e^{-2.50}=0.0821.\]
单位/有效数字: \(\mu L\) dimensionless;transmission 三位有效数字。
sign/reasonableness: \(0<I/I_0<1\),增加任一 path length 会降低 transmission。
模型边界/解释: detector 期望只收到 incident 的 \(8.21\%\);polychromatic beam/scatter 会使测得 ratio 偏离此 ideal value。
Worked Example 2 — ultrasound depth 与 Doppler velocity
已知: echo delay \(t=130\ \mathrm{\mu s}\),sound speed \(c=1540\ \mathrm{m\,s^{-1}}\);同一系统发射 \(f_0=5.00\ \mathrm{MHz}\),测得 \(\Delta f=1.62\ \mathrm{kHz}\),beam–flow angle \(\theta=60.0^\circ\)。
研究对象与边界: straight pulse-echo path and single-scattering Doppler from blood;constant \(c\)。
参考系/坐标: \(+z\) into body;\(v>0\) toward sign convention associated with positive measured shift。
模型理由: round-trip timing 与 backscatter Doppler approximation。
\[z=\frac{ct}{2}=\frac{(1540)(130\times10^{-6})}{2}=0.100\ \mathrm m,\]
\[v=\frac{c\Delta f}{2f_0\cos\theta} =\frac{(1540)(1.62\times10^3)}{2(5.00\times10^6)(0.500)} =0.499\ \mathrm{m\,s^{-1}}.\]
单位/有效数字: depth 三位有效数字;frequency ratio dimensionless,velocity 为 \(\mathrm{m\,s^{-1}}\)。
方向/limit check: \(\cos60^\circ=0.5\);若误把 round trip 或 Doppler factor 2 漏掉,都会差两倍。
解释: \(\theta\) 接近 \(90^\circ\) 时 velocity estimate ill-conditioned;real scanner 需 angle correction 与 aliasing check。
Worked Example 3 — 3 T MRI 的 Larmor frequency 与 relaxation fractions
已知: proton \(\gamma/2\pi=42.58\ \mathrm{MHz\,T^{-1}}\),\(B_0=3.00\ \mathrm T\);ideal recovery \(T_1=1.20\ \mathrm s\)、read time \(t=0.800\ \mathrm s\);transverse \(T_2=0.100\ \mathrm s\)、echo time \(0.0800\ \mathrm s\)。
研究对象与边界: homogeneous single-compartment spin ensemble after ideal excitation;no \(T_2^*\) in echo model。
参考系/坐标: \(\mathbf B_0\) along \(+z\);report frequency magnitude。
模型理由: Larmor and single-exponential Bloch relaxation limits。
\[f_0=(42.58)(3.00)=128\ \mathrm{MHz},\]
\[\frac{M_z}{M_0}=1-e^{-0.800/1.20}=0.487, \qquad \frac{M_{xy}}{M_{xy}(0)}=e^{-0.0800/0.100}=0.449.\]
单位/有效数字: MHz/T times T gives MHz;fractions dimensionless,三位有效数字。
limit check: \(t=0\) gives zero recovered longitudinal and full transverse; long time gives \(M_z/M_0\to1\), \(M_{xy}\to0\)。
解释: tissue brightness 还由 spin density、sequence timing、flip angle、gradients、coil sensitivity 与 reconstruction 决定。
Worked Example 4 — PET tracer activity after decay
已知: \(^{18}\mathrm F\) physical half-life \(T_{1/2}=109.8\ \mathrm{min}\),initial activity \(A_0=370\ \mathrm{MBq}\);等待 \(t=120\ \mathrm{min}\),忽略 biological clearance。
研究对象与边界: sealed tracer population with physical radioactive decay only。
参考系/符号: \(t=0\) at stated calibration time;activity magnitude positive。
模型理由: independent decay exponential。
\[\frac{A}{A_0}=2^{-120/109.8}=0.469, \qquad A=(370)(0.469)=174\ \mathrm{MBq}.\]
单位/有效数字: exponent dimensionless;三位有效数字。
sign/reasonableness: 经过稍多于一 half-life,fraction 应略小于 \(0.5\),符合 \(0.469\)。
模型边界/解释: detected coincidence rate 还乘 geometry/efficiency 并受 attenuation/dead time;人体内 biological clearance 使 effective activity evolution 不同。
Worked Example 5 — Poisson counts 与 CNR tradeoff
已知: lesion ROI 期望 counts \(N_1=40000\),matched background ROI \(N_2=36000\),independent Poisson;定义 difference CNR 为 \((N_1-N_2)/\sqrt{N_1+N_2}\)。低-count protocol 把两者都降为四分之一。
研究对象与边界: equal-size independent ROIs,no reconstruction correlation/systematic bias。
参考系/符号: lesion minus background positive。
模型理由: independent Poisson variances add。
\[\mathrm{CNR}_{high}=\frac{4000}{\sqrt{76000}}=14.5.\]
低-count 时 \(N_1'=10000,N_2'=9000\):
\[\mathrm{CNR}_{low}=\frac{1000}{\sqrt{19000}}=7.25.\]
单位/有效数字: counts treated dimensionless;三位有效数字。
scaling check: counts 降四倍使 CNR 降为一半,符合 \(\sqrt N\) scaling。
模型边界/解释: dose/activity/time reduction 不能仅凭此公式决定;real reconstruction introduces correlated, nonstationary noise and task-specific thresholds。
Lesson synthesis / 全课综合
任何 modality 都可用同一条 measurement-to-decision pipeline:
- 明确 clinical/scientific task 与 system boundary:要定位 anatomy、velocity、relaxation 还是 tracer function?
- 写 forward model:X-ray attenuation、acoustic propagation、spin evolution 或 radioactive emission/detection;
- 列 detector measurement 与 calibration,区分 raw counts/voltages 和 displayed pixel values;
- 检查 sampling、noise、motion、scatter/inhomogeneity 与 incomplete-data mechanisms;
- 选择 reconstruction/regularization,并声明先验与可能 bias;
- 用 phantom、resolution/contrast/noise/task metrics 和 uncertainty 做 validation;
- 对 ionizing exam 做 justification/optimization,对所有 modalities 检查其特有 safety constraints;
- 最终 image 是 evidence,不自动等于 diagnosis;interpretation 需 clinical context 与 qualified professionals。
不同 modality 的 contrast source 不同:CT 主要从 X-ray attenuation,ultrasound 从 acoustic impedance/scattering/motion,MRI 从 proton density 与 sequence-weighted relaxation,PET/SPECT 从 radiotracer distribution。把一种 modality 的“亮”直接映射成另一种 modality 的“dense”会导致类别错误。
Formula-and-decision table / 公式与决策表
| Question | Relation | Use when | Main limitation |
|---|---|---|---|
| X-ray transmission? | \(I/I_0=e^{-\int\mu d\ell}\) | monoenergetic narrow-beam idealization | scatter, spectrum, beam hardening |
| CT projection? | \(p=-\ln(I/I_0)=\int\mu d\ell\) | calibrated positive transmission | zeros, noise and model mismatch |
| Echo depth? | \(z=ct/2\) | straight pulse-echo path, known \(c\) | refraction and sound-speed variation |
| Interface reflection? | \(R_I=[(Z_2-Z_1)/(Z_2+Z_1)]^2\) | normal plane-wave incidence | oblique/diffuse/rough interfaces |
| Doppler velocity? | \(v=c\Delta f/(2f_0\cos\theta)\) | backscatter with known angle | angle error and aliasing |
| MRI frequency? | \(f_0=(\gamma/2\pi)B_0\) | known species and field | chemical shift and field inhomogeneity |
| Longitudinal recovery? | \(M_z=M_0+[M_z(0)-M_0]e^{-t/T_1}\) | known initial \(M_z(0)\), single-exponential recovery toward \(M_0\) | multi-component exchange and sequence details |
| Transverse decay? | \(M_{xy}=M_{xy0}e^{-t/T_2}\) | ideal spin-echo \(T_2\) model | \(T_2^*\) and complex tissue |
| Radioactive activity? | \(A=A_0 2^{-t/T_{1/2}}\) | physical decay only | biological clearance and calibration |
| Counting precision? | \(\sigma_N\approx\sqrt N\) | Poisson-dominated independent counts | corrections and reconstructed correlations |
| Simple detectability? | \(\mathrm{CNR}=\lvert S_1-S_2\rvert/\sigma\) | stated ROI/task/noise definition | observer/task/texture dependence |
Model boundaries / 模型边界
- Beer–Lambert/CT: monoenergetic, narrow beam and accurate calibration are idealizations; polychromatic scatter-corrected systems need detailed forward models.
- Tomographic inversion: complete angle/data assumptions fail with limited view, truncation, metal and motion; regularization cannot create guaranteed truth from missing data.
- Ultrasound ray model: ignores diffraction and finite beam; speckle is coherent interference, not merely electronic noise.
- Doppler equation: returns beam-parallel component; \(\theta\approx90^\circ\) is ill-conditioned and pulsed systems can alias.
- MRI single compartment: \(T_1/T_2\) and signal depend on field, sequence, exchange and tissue microenvironment; MR uses no ionizing radiation but is not risk-free.
- PET/SPECT ideal events: attenuation, scatter, randoms, collimator response, positron range and dead time alter quantification.
- Poisson rule: raw counts may be Poisson, but corrections and iterative reconstruction create correlated spatially varying noise.
- Dose comparison: administered activity, absorbed/effective dose, CTDI and DLP are not interchangeable patient-specific quantities.
- Image quality: resolution, contrast and noise are task-dependent; a sharper-looking image may be biased or artifact-rich。
Misconception clinic / 误区诊所
- “radiograph 是身体内部的一张透视照片。” 它是沿 depth 的 attenuation projection,结构 overlap 不可由肉眼自动分离。
- “CT 只要增加角度就一定完全真实。” calibration、noise、scatter、motion、metal 和 model mismatch 仍会造成 artifacts/bias。
- “ultrasound echo time 乘 \(c\) 就是 depth。” pulse 往返,必须除以 2;且 \(c\) variation 会产生 range error。
- “Doppler 测到血流速度的完整 vector。” 单 beam 主要测 line-of-sight component,依赖 angle correction。
- “MRI 用核辐射,所以有 X-ray dose。” MRI 测 nuclear magnetic resonance,不用 ionizing X ray;但 strong field、RF、gradients 和 implants 有其他风险。
- “\(T_1\) 和 \(T_2\) 是组织永恒不变的标签。” 它们依 field、temperature、microenvironment、sequence/model 与 pathology。
- “PET detector 直接知道 annihilation 点。” coincidence 只给 LOR;positron range 和 non-collinearity 也模糊位置。
- “SPECT 与 PET 只是 tracer 名称不同。” SPECT 用 single-gamma collimation,PET 用 paired coincidence,physics/sensitivity/resolution 不同。
- “pixel 更小就必然分辨率更高。” PSF、SNR、sampling 和 reconstruction 共同限制 resolution。
- “dose 越高图像一定越好,或越低一定越安全合理。” optimization 以 clinical task 为约束;不必要 exposure 应避免,过低导致 nondiagnostic exam 也可能造成 harm。
Practice and retrieval / 练习与检索
- [K01] uniform object \(\mu=15.0\ \mathrm{m^{-1}}\)、thickness \(0.0800\ \mathrm m\),求 projection \(p\) 与 transmission。
Answer 1
\(p=\mu L=1.20\);\(I/I_0=e^{-1.20}=0.301\)。结果只适用于 calibrated monoenergetic narrow-beam model。
- [K01] 为什么把所有 CT projections 直接“向后涂抹”会 blur,而 filtered backprojection 需要 filtering?
Answer 2
simple backprojection 把每个 ray value 沿整条 path 扩散,低 spatial frequencies 被过度加权;filter 在 ideal inverse Radon relation 中校正频率权重。有限/noisy data 时 filtering 会放大高频 noise,故需 window/regularization tradeoff。
- [K02] echo 在 \(65.0\ \mathrm{\mu s}\) 返回,取 \(c=1540\ \mathrm{m\,s^{-1}}\),求 depth。
Answer 3
\(z=ct/2=(1540)(65.0\times10^{-6})/2=0.0501\ \mathrm m=5.01\ \mathrm{cm}\)。
- [K02] Doppler beam 与 flow 成 \(80^\circ\);若 angle 被误设为 \(70^\circ\),速度估计会如何偏?
Answer 4
真值 denominator 用 \(\cos80^\circ=0.174\),误设用 \(\cos70^\circ=0.342\),estimated \(v\) 约为 true 的 \(0.174/0.342=0.508\),低估约 \(49\%\);大角度极其敏感。
- [K03] proton 在 \(1.50\ \mathrm T\) 的 Larmor frequency 是多少?取 \(\gamma/2\pi=42.58\ \mathrm{MHz\,T^{-1}}\)。
Answer 5
\(f_0=(42.58)(1.50)=63.9\ \mathrm{MHz}\)。这是 resonance frequency magnitude;local gradients/chemical shift 会产生小 position/environment-dependent offsets。
- [K03] tissue A、B 的 \(T_2\) 分别 \(60\ \mathrm{ms}\)、\(100\ \mathrm{ms}\),若同初始 transverse signal 且 \(TE=80\ \mathrm{ms}\),哪一个保留更多 signal?
Answer 6
A fraction \(e^{-80/60}=0.264\);B fraction \(e^{-80/100}=0.449\),B 更多。实际 image brightness 还受 \(T_1\)、proton density 和 sequence 影响。
- [K04] tracer 经过三个 half-lives 后 activity fraction 与 Poisson relative noise(若 counts 同比例下降)如何变化?
Answer 7
activity/count expectation 降至 \(1/8\);relative noise \(1/\sqrt N\) 增至 \(\sqrt8=2.83\) 倍,假设 acquisition/sensitivity/background 不变。
- [K04] 分别说出 PET 与 SPECT 定位 photon origin 的主要 geometric mechanism。
Answer 8
PET 以两个 near-opposite \(511\) keV photons 的 time coincidence 定义 LOR;SPECT 用 absorbing collimator 限制 single gamma 的 arrival direction。两者都需 tomography、attenuation/scatter correction。
- [K05] detector pixel width 减半,为何 spatial resolution 不一定加倍?
Answer 9
system PSF 还受 focal spot、detector aperture/collimator、wave pulse/beam、motion、sampling SNR 与 reconstruction 控制;更小 pixel 可能只是 oversample 同一 blur,并使 per-pixel counts 降低。
- [综合] pediatric abdominal question 可由 CT、MRI 或 ultrasound 候选回答。物理层面的 modality decision 至少要比较哪些量?
Answer 10
应比较 clinical task/urgency、target contrast、depth/air/bone access、required resolution、motion/time、ultrasound operator/window、MRI field/implant/sedation constraints、CT ionizing dose and pediatric protocol、contrast-agent risks、availability、expected nondiagnostic/repeat probability。不能只按“谁没有 radiation”或“谁最清楚”单指标决定。
Exit ticket / 迁移检验
一套新 reconstruction 宣称:CT dose 减半后图像“与原图一样锐利”。请设计最小验证,判断这是保留了 diagnostic information,还是只生成锐利外观。
Exit-ticket benchmark
使用 independent raw/phantom and patient-like data,固定 clinical detection/measurement task;比较 dose-calibrated acquisition、PSF/MTF、noise power/correlation、CNR、bias、low-contrast detectability、artifact rate 与 reader performance;用 withheld ground-truth surrogates 检查 small lesions/edges,报告 confidence intervals 和 failure subgroups。sharpness display 或 pixel-level similarity不足;必须确认 dose indices、reconstruction version、preprocessing 与 repeatability,且不能让 test set 泄漏到 algorithm tuning。
References and source notes / 参考资料与使用说明
以下权威直接来源均在 2026-08-31 实际打开并核对。正文重新组织;全部 TikZ 依据 forward models 原创绘制,未复制来源插图或临床图像。
- MIT OpenCourseWare, 22.56J Noninvasive Imaging in Biology and Medicine — X-ray CT、PET/SPECT、MRI 与 molecular imaging 的大学课程范围及 inverse-problem 框架。
- NIH/NIBIB, Computed Tomography (CT) — rotating X-ray source/detector、slice reconstruction、contrast 与 ionizing-radiation context。
- NIH/NIBIB, Ultrasound — piezoelectric transduction、echo imaging、functional flow information 与 safety context。
- NIH/NIBIB, Magnetic Resonance Imaging (MRI) — proton alignment/excitation/relaxation、soft-tissue contrast 与 strong-field constraints。
- NIH/NIBIB, Nuclear Medicine — radiotracer、SPECT camera、PET annihilation coincidence 与 diagnostic exposure context。
- U.S. FDA, Medical X-ray Imaging — X-ray/CT benefit–risk、justification、ALARA optimization、patient-size and QA considerations。
Source-to-content map
| Source | Used for | Knowledge ID / Section | Model boundary or convention |
|---|---|---|---|
| MIT OCW 22.56J | modality comparison and reconstruction/inverse framework | K01, K03, K04, K05 | graduate course source simplified to explicit HS→UG1 mathematics |
| NIBIB CT | CT acquisition, slices, contrast and risk | K01, K05 | public technical overview;正文 adds Beer–Lambert/Radon model and artifact limits |
| NIBIB Ultrasound | transducer, echo, functional flow, nonionizing context | K02 | equations assume straight path, known \(c\), simple backscatter |
| NIBIB MRI | proton excitation/realignment and modality constraints | K03 | single-exponential model is teaching idealization, not tissue certificate |
| NIBIB Nuclear Medicine | tracer, SPECT, PET coincidence | K04 | counts require attenuation/scatter/efficiency corrections |
| FDA Medical X-ray Imaging | justification, optimization, QA, dose context | K01, K05 | CTDI/DLP are scanner/phantom indices, not exact individual organ dose |
复习与拓展
- Radiation Therapy and Health Physics 将从 diagnostic exposure 转向 deliberate dose deposition、treatment planning 与 protection/QA。
- 可扩展:Fourier slice theorem、filtered backprojection derivation、ultrasound beamforming、Nyquist/Doppler aliasing、Bloch equations、k-space trajectories、time-of-flight PET、attenuation/scatter correction 与 task-based image-quality metrics。
- 与 Quantitative Decisions in Applied Physics 的接口:modality choice 是 benefit、hazard、exposure、uncertainty、resource 与 communication 的定量决策。