「肝动脉没变」这句话,在一根桡动脉上到底能不能读出来:从单点脉搏谐波剖面里分离各供血支路贡献的可分离性判据

Written by

in

Language

Evidence level: C

桡动脉只有一处观测点,它拿到的是五条支路经波反射叠加后的合成波形——要在这里面证明「肝动脉没变」而「肠系膜变了」,先得证明这个欠定方程组能解。

正文

一、一个外科级的前提

上一篇把「一脏一谐波」的表改成了逐器官裁决,配了一条锋利的设计:把肝动脉当阴性对照——因为餐后肝动脉的缓冲反应「在各种条件下都被保住」(conserved under all conditions, PMID 15086358),它是一支已知不该变的支路。若同一模型在肠系膜上动脉报出大改写、在肝动脉上也报出大改写,改写量的来源就是共模(波反射、僵硬度、心率),不是分配。

这个逻辑没错,但它踩在一个从未被检验的前提上:从一根桡动脉上,到底能不能把「肠系膜支路变了」和「肝动脉没变」分开?

桡动脉只有一处观测点。它拿到的不是五条支路的各自信号,而是五条支路经波反射叠加后的合成波形——没有任何一条支路的标签。要从合成波形里还原各支路的贡献,等价于解一个欠定方程组:未知数(各支路阻力、流量份额)比观测约束(一个波形的前若干阶谐波)多。

这一跳在血流动力学里早有明确的病态性证据。

二、多床反解本来就是病态的

最直接的参照是 67 段 16 床的手臂模型:要靠解非线性方程组反解末端阻力,还必须专门做敏感性分析才敢用——因为解对初值和噪声极其敏感(PMID 22771032)。这说明「从合成波形反推多床参数」这件事本身不是默认能成的,是需要单独论证的条件命题。

第二层麻烦是共线。谐波失真在 24 小时内随收缩压线性下降,又与血管切向模量、脉搏波速度线性相关(PMID 35433650)——波形的前若干阶谐波变化里至少混着血压、僵硬度、出口阻抗三个共线来源。同一段波形变化可以被归因给不同的输入,而数据本身分不出谁对谁错。

对 Windkessel 类模型,四元件里的惯性参数 L 早已被指认是估计噪声的主源——「离群值几乎全部由于 L 的估计误差」(PMID 20172779)。若各支路都带 L,反解的不确定性会系统性集中在惯性参数上,而不是均匀分布在各支路阻力上。

三、这个问题不是「加大样本」能解决的

可辨识性分两层,必须分开判:结构层(拓扑决定,加样本无用)与实践层(噪声与采样决定,可救)。对心血管模型的先验全局可辨识性分析(PMID 26726299)已经证明:同一类拓扑里,有的模型全局可辨识,有的(如 T-Tube「一入口两出口」)被判非全局可辨识——这是拓扑层面的病态,不是样本量层面的。

五床(五条供血支路)比 T-Tube 更复杂。若每个床带阻力 R 与顺应性 C 两个参数、而观测只定住总阻力与总顺应性,十个参数只对两个观测约束——即便每支路另有波形形状信息可用,自由度是否够到十个也必须算清,不能凭「有相位关系」一句带过。这是代数问题,不是测量问题。

四、可分离性判据:三个数字三档结论

要检验「肝动脉没变」能不能从单点波形里读出来,不必新增受试者——复用 1-D 模型仿真加真实数据双轨即可:

  • 仿真轨道:在模型里只改肠系膜支路出口阻力(复现 70% 血流升幅, PMID 40445944),记录桡动脉前 10 阶谐波幅值/相位的响应向量 v₁;再只改肝动脉(幅度按 20% 上限, PMID 15086358),记录 v₂。算 v₁ 与 v₂ 的余弦相似度与条件数。
  • 真实轨道:用 4D flow MRI 数据(同批受试者餐前餐后同测八条血管, PMID 33547918)作已知标签,反推「肝动脉份额不变」这一约束在单点拟合里能否被满足。
  • 替代方案:若单点不可分离,改用双点观测(桡动脉 + 股动脉)或直接引入超声独立测肝动脉作旁路标签。

判据分三档:

第一档:v₁ 与 v₂ 余弦相似度 > 0.9 或条件数 > 10 → 单点不可分离。#1176 的第二、三格判据(阴性对照裁决)作废,改走双点观测或超声旁路。这不是样本量问题——加大样本也分不开两个在观测点上同向的机制。

第二档:相似度 < 0.5 且条件数 < 3 → 单点可分离,原判据照用。

第三档:相似度落在 0.5–0.9 → 须为每个器官限定「可判时相窗」——只在灵敏度最高的时段采。比如肠系膜上动脉的 70% 升幅在餐后 20–40 分钟最显著(PMID 40445944),而肝动脉全程守恒,那么在峰值窗口采样能最大化两机制在观测点上的分离度。

五、这件事为什么绕不过去

「一脏一谐波」整条研究线的地基,是把谐波剖面读成器官身份的窗口。窗口之所以是窗口,前提是剖面的变化能定位到具体的支路。如果这个可分离性前提不成立,「肝动脉没变」就不是一个可检验的命题——它只是模型的一个假设,被写进了结论里。

而反过来,如果仿真显示单点确实可分(相似度低、条件数小),肝动脉阴性对照的设计就站得住,逐器官裁决就有了一个真正的锚。这一测,决定的是后面所有逐器官裁决文章的地基稳不稳——它不是可选项。

证据与局限

已验证:餐后内脏血流确实大幅重分配(肠系膜上动脉 +70%, PMID 40445944;腹腔干可被病理单独抹掉, PMID 32857602;肝动脉守恒, PMID 15086358);多床反解是病态的(67 段 16 床须敏感性分析, PMID 22771032;Windkessel 的 L 是噪声主源, PMID 20172779);谐波剖面与血压/僵硬度/阻抗共线(PMID 35433650);同一类心血管拓扑的可辨识性可分层判定(PMID 26726299)。

部分验证:4D flow MRI 已能同批同测八条血管的餐后反应(PMID 33547918),证明「已知标签的真实数据」在技术上可获得——可分离性检验的真实轨道是可执行的。

待验证:① 肠系膜与肝动脉两机制在桡动脉观测点上的响应向量相似度与条件数,从未被算过——这是本文要填的第一个数;② 单点拟合能否满足「肝动脉份额不变」约束,未有实证;③ 若不可分离,双点(桡+股)或超声旁路能否恢复可分离性,未有实证。

L1 网络状态:该主题尚未进入 L1 对应网络的「单点可分离性」层。CORRESPONDENCE_NETWORK.md 生理侧节点表仅八项横断面观测项(HRV/皮温皮电/呼吸率/肌电/EEG/胃肠蠕动/气感报告/体质分型),无「器官血流分配」「波反射叠加」「欠定方程组」任何节点;物理侧仅「相位/节律 — 周期性耦合的关键」一行,不带份额/叠加/可解性语义;边表止于 E-050(纳音↔频率,现代证据栏原文「尚无任何可测来源」)。EDGE_TABLE_science-yixue.md 概念表中「脉象」73 篇最高 A 级但 A 级数仅 5、「五脏」26 篇最高 B 级、A 级数 0——该表统计「概念被写过几篇」,不统计「同一观测量能否定位到具体支路」,无任何以可分离性、条件数或响应向量为源或目标的边。本主题对应「器官血流分配 ↔ 脉搏谐波阶次」簇(母文 #1176 已登记簇首条)的深化边:「单点观测的可分离性(响应向量余弦相似度 / 条件数 / 欠定自由度)」尚未入网。

相关阅读

证据等级: C
参考文献: PMID 40445944, PMID 32857602, PMID 15086358, PMID 22771032, PMID 20172779, PMID 35433650, PMID 26726299, PMID 33547918


免责说明:本文属传统知识与科学文献的整理、对照与研究设计讨论,仅供理解与参考,不构成医疗建议,不能替代专业诊断与治疗。文中涉及的效应量、频率区间与暴露结论多为研究报告层面的推断,个体情况差异很大;如有身体不适或健康疑虑,请就医并由专业医师判断。

Can contributions of individual supply branches be separated from single-point pulse wave harmonics: a separability criterion

Evidence level: C

The radial artery has only one observation point. It receives a composite waveform formed by the superposition of wave reflections from five branches. To prove that “the hepatic artery remained unchanged” while “the superior mesenteric artery changed,” one must first demonstrate that this underdetermined system of equations is solvable.

Body

I. A Surgical-Level Premise

The previous study converted the “one organ, one harmonic” table into an organ-by-organ adjudication, introducing a sharp design: using the hepatic artery as the negative control—because postprandial hepatic arterial buffering responses are “conserved under all conditions” (PMID 15086358), it is a branch known not to change. If the same model reports major changes in the superior mesenteric artery and also major changes in the hepatic artery, the source of these changes must be common-mode (wave reflection, stiffness, heart rate), not allocation.

This logic is sound, but it rests on a never-tested premise: Can one actually separate “the superior mesenteric branch changed” from “the hepatic artery did not change” using only a single radial artery?

The radial artery has only one observation point. It does not receive the individual signals of the five branches, but rather a composite waveform formed by their superposition—without any branch labels. Reconstructing the contribution of each branch from this composite waveform is equivalent to solving an underdetermined system of equations: the number of unknowns (resistance and flow share of each branch) exceeds the number of observational constraints (the first few harmonics of a single waveform).

This jump in hemodynamics already has clear evidence of ill-posedness.

II. Multi-Bed Inversion Is Inherently Ill-Posed

The most direct reference is the 67-segment, 16-bed arm model: to invert terminal resistance from a nonlinear system of equations, one must perform sensitivity analysis before using the results—because the solution is extremely sensitive to initial values and noise (PMID 22771032). This indicates that “inverting multi-bed parameters from a composite waveform” is not automatically possible; it requires specific argumentation as a conditional proposition.

A second complication is collinearity. Harmonic distortion decreases linearly with systolic pressure over 24 hours and is also linearly correlated with vascular tangential modulus and pulse wave velocity (PMID 35433650)—at least three collinear sources (blood pressure, stiffness, outlet impedance) are mixed within the changes of the first few harmonics of the waveform. Changes in the same waveform segment can be attributed to different inputs, while the data itself cannot distinguish which is correct.

For Windkessel-type models, the inertial parameter L in the four-element model has long been identified as the main source of estimation noise—”outliers are almost entirely due to L estimation error” (PMID 20172779). If each branch carries an L, the uncertainty in inversion will systematically concentrate on inertial parameters rather than being evenly distributed across branch resistances.

III. This Problem Cannot Be Solved by “Increasing Sample Size”

Identifiability has two layers that must be judged separately: the structural layer (determined by topology, adding samples is useless) and the practical layer (determined by noise and sampling, potentially salvageable). A prior global identifiability analysis of cardiovascular models (PMID 26726299) has proven that within the same class of topology, some models are globally identifiable while others (such as the T-Tube “one inlet, two outlets”) are judged non-globally identifiable—this is a topological ill-posedness, not an issue of sample size.

The five-bed (five supplying branches) system is more complex than the T-Tube. If each bed carries two parameters—resistance R and compliance C—and observations only fix total resistance and total compliance, then ten parameters correspond to only two observational constraints. Even if waveform shape information is available for each branch, whether the degrees of freedom suffice for ten parameters must be calculated precisely; one cannot pass this over with “there is a phase relationship.” This is an algebraic problem, not a measurement problem.

IV. Separability Criteria: Three Numbers, Three Conclusions

To test whether “hepatic artery unchanged” can be read from a single-point waveform, no new subjects need to be added—simulating the 1-D model alongside real data on dual tracks suffices:

  • Simulation track: In the model, only change the outlet resistance of the superior mesenteric branch (reproducing a 70% blood flow increase, PMID 40445944), record the response vector v₁ of the first 10 harmonic amplitudes/phases at the radial artery; then only change the hepatic artery (amplitude capped at 20%, PMID 15086358), record v₂. Calculate the cosine similarity and condition number between v₁ and v₂.
  • Real track: Use 4D flow MRI data (same subjects measured eight vessels pre- and postprandially, PMID 33547918) as known labels to reverse-engineer whether the constraint “hepatic artery share unchanged” can be satisfied in single-point fitting.
  • Alternative: If single-point separation fails, use dual-point observation (radial + femoral arteries) or introduce ultrasound-independent hepatic artery measurement as a bypass label.

The criteria fall into three tiers:

Tier 1: Cosine similarity between v₁ and v₂ > 0.9 or condition number > 10 → single-point inseparable. #1176’s second and third criteria (negative control adjudication) are void; switch to dual-point observation or ultrasound bypass. This is not a sample size issue—increasing samples cannot separate two mechanisms that move in the same direction at the observation point.

Tier 2: Similarity < 0.5 and condition number < 3 → single-point separable, original criteria hold.

Tier 3: Similarity falls between 0.5–0.9 → must define a “sensitive time window” for each organ—sample only during the period of highest sensitivity. For example, the 70% increase in the superior mesenteric artery is most pronounced 20–40 minutes postprandially (PMID 40445944), while the hepatic artery remains conserved throughout; sampling at the peak window maximizes separation of the two mechanisms at the observation point.

V. Why This Cannot Be Circumvented

The foundation of the entire “one organ, one harmonic” research line is reading the harmonic profile as a window into organ identity. The reason this is a “window” is precisely that profile changes can be localized to specific branches. If this separability premise does not hold, “hepatic artery unchanged” is not a testable proposition—it is merely an assumption embedded in the conclusion.

Conversely, if simulations show single-point separation is indeed possible (low similarity, low condition number), the design of the hepatic artery negative control stands firm, and organ-by-organ adjudication gains a true anchor. This test determines the stability of the foundation for all subsequent organ-by-organ adjudication papers—it is not optional.

Evidence and Limitations

Verified: Postprandial visceral blood flow indeed undergoes major redistribution (superior mesenteric artery +70%, PMID 40445944; celiac trunk can be pathologically isolated, PMID 32857602; hepatic artery conserved, PMID 15086358); multi-bed inversion is ill-posed (67-segment 16-bed requires sensitivity analysis, PMID 22771032; Windkessel L is the main noise source, PMID 20172779); harmonic profiles are collinear with blood pressure/stiffness/impedance (PMID 35433650); identifiability of cardiovascular topologies can be stratified (PMID 26726299).

Partially verified: 4D flow MRI can measure postprandial responses of eight vessels simultaneously in the same batch (PMID 33547918), proving that “real data with known labels” is technically obtainable—the real track for separability testing is executable.

Pending verification: ① The cosine similarity and condition number of the response vectors for superior mesenteric vs. hepatic artery mechanisms at the radial artery observation point have never been calculated—this is the first number this paper fills; ② Whether single-point fitting can satisfy the “hepatic artery share unchanged” constraint lacks empirical evidence; ③ If inseparable, whether dual-point (radial+femoral) or ultrasound bypass can restore separability lacks empirical evidence.

L1 Network Status: This topic has not yet entered the “single-point separability” layer of the L1-corresponding network. The CORRESPONDENCE_NETWORK.md physiological node table contains only eight cross-sectional observation items (HRV, skin temperature/skin conductance, respiration rate, EMG, EEG, gastrointestinal motility, air sensation report, constitution typing), with no nodes for “organ blood flow allocation,” “wave reflection superposition,” or “underdetermined equation system”; the physical side contains only the row “phase/rhythm — key periodic coupling,” without share/superposition/solvability semantics; the edge table stops at E-050 (Na Yin ↔ frequency, modern evidence column states “no measurable source yet”). The EDGE_TABLE_science-yixue.md concept table lists “pulse diagnosis” with 73 papers at highest evidence level A (only 5 A-level papers) and “five organs” with 26 papers at highest evidence level B (zero A-level papers)—this table counts “how many times a concept has been written,” not “whether the same observable can be localized to specific branches.” No edges originate from or target separability, condition number, or response vectors. This topic corresponds to the deepening edge of “organ blood flow allocation ↔ pulse harmonic order” cluster (cluster head already registered in parent paper #1176): “single-point observability separability (response vector cosine similarity / condition number / underdetermined degrees of freedom)” has not yet entered the network.

Related Reading

Evidence Level: C
References: PMID 40445944, PMID 32857602, PMID 15086358, PMID 22771032, PMID 20172779, PMID 35433650, PMID 26726299, PMID 33547918


Disclaimer: This article compiles traditional knowledge and scientific literature for comparison and research-design discussion only. It is not medical advice and cannot replace professional diagnosis or treatment. Effect sizes, frequency ranges and exposure conclusions cited here are largely inferences at the research-report level, and individual circumstances vary widely. If you have any discomfort or health concerns, please consult a physician.

Comments

Leave a Reply

Your email address will not be published. Required fields are marked *