Evidence level: B
血管几何模型算出的谐波能量谱,与中医师摸出的浮沉弦滑数迟,可能是同一套信息的两种编码——把”弦脉”翻译成”基频能量比偏高、谐波衰减率偏低”,再用盲法对照去检验这个翻译准不准。
为什么这个问题值得做
上一篇文章《无创血管几何重建1-D模型预测脉搏频谱的逐谐波能量分布闭合检验》证明了:用超声/MRI重建个体血管几何,喂进1-D血流动力学模型,可以算出这个人桡动脉处的脉搏频谱,且逐谐波能量分布理论上有希望逼近实测。但”模型算得准”只回答了物理层的问题。真正悬而未决的是临床层的问题:这套物理上可解释的谐波能量谱,和中医师手指底下摸出来的脉象分类,到底是不是同一回事?
中医把脉象分成浮、沉、弦、滑、数、迟等二十多种,靠手指对压力、速度、光滑度、张力的触觉综合。现代脉搏波分析(PWA)证明这些”手感”不是幻觉:一项针对高血压患者的横断面研究显示,浮/沉(深度)、强/弱(力度)、快/慢(速度)三个要素与PWV、心率、脉压等血流动力学指标有显著的多变量关联(PMID 26988431),浮脉对应心率≥58.25次/分,数脉对应心率≥69.6次/分。也就是说,中医脉象要素在现代血流动力学里有可量化的对应物。那频谱呢?傅里叶级数分析已经显示,正常、弦(taut)、滑(slippery)三类3D脉搏图像在第一谐波比值上存在显著差异(PMID 31885653)——弦脉的第一谐波占比系统性高于正常脉和滑脉。这暗示:脉象分类的频域签名是真实存在的。
1-D模型与脉象之间的桥梁
把两件事拼起来看,逻辑就通了:①1-D模型能算出个体化的谐波能量谱(PMID 21622820验证患者特异1-D模型显著优于通用模型;PMID 32142412验证62人个性化0D-1D模型压力波MSE仅7.1 mmHg²、颈动脉血流相关0.911);②中医脉象在频域有可分离的签名(PMID 31885653)。如果①的频谱能预测②的分类,那就说明:中医师摸脉时感受到的张力和形状差异,本质上是血管几何在谐波结构上的投影——”弦脉”是基频能量比偏高、谐波衰减率偏低的频谱,”滑脉”是某一高阶谐波能量相对突出的频谱。这与王唯工”谐波对应脏腑”的框架一致:不同谐波阶次携带不同生理信息,而脉象就是这些谐波比例的触觉摘要。
这不是空想。机器学习已经能仅凭桡动脉脉搏波以约80%准确率分类冠心病/健康/妊娠(PMID 41087424,加临床参数达91%)。但这些模型是”黑箱”——它们用大量时域/时频特征做判别,不解释为什么。1-D模型的独特价值在于:它的谐波特征全部由物理参数(血管几何、弹性、边界条件)决定,可解释、可溯源。如果这套物理可解释的频谱能盲法预测中医脉象,那就不是”机器学习刚好找到一个统计规律”,而是”脉象=血管几何的频域指纹”获得了独立验证。
盲法系统对照怎么设计
核心设计是医生盲法评级 × 模型盲法预测 × 判别AUC裁决:
- 采集端:对同一批受试者(建议覆盖浮沉弦滑数迟等6类主要脉象,每类≥30人),由≥3位高年资中医师独立盲法评级(只摸脉、不知其他信息),kappa一致性达标(≥0.6)者入组;同时做桡动脉/颈动脉超声/MRI几何重建+桡动脉实测脉搏波采集。
- 预测端:用几何重建跑1-D模型,输出模型预测的谐波特征——基频能量比(C1占比)、谐波衰减率(C2-C10能量下降斜率)、共振峰位置(谱峰对应谐波阶次)。
- 对照端:医生评级标签 vs 模型谐波特征做判别建模(判别AUC),检验模型频谱能否盲法区分浮/沉/弦/滑/数/迟六类。三种结局都是知识:
– 显著命中(AUC>0.75且跨中心可复现)→ 脉象的频域基础获物理模型级证据,”血管几何决定脉象频谱”从波形层面推进到分类层面;
– 部分命中(只命中数/迟等心率相关类)→ 脉象分类里只有一部分(心率类)可被几何-频谱解释,形状类(浮沉弦滑)依赖几何之外的因素(管壁弹性、微循环边界条件);
– 完全随机 → 脉象分类编码的不是频谱信息,该假说需重审。
与现有研究的差异化
这项设计不是又一个”机器学习分类脉象”——那些工作已经很多,准确率也高。差异化在于三点:①特征可解释:用的是1-D模型输出的物理谐波特征,不是黑箱时频特征;②盲法双端:医生不知仪器结果、建模者不知医生评级,防”评级-特征循环论证”;③与#266闭合检验衔接:上一篇证明模型频谱≈实测频谱,本篇证明实测频谱≈脉象分类——两环扣起来,”血管几何→谐波频谱→脉象分类”的因果链才有完整证据。
证据与局限
已验证: 中医脉象要素与血流动力学指标显著关联(PMID 26988431);正常/弦/滑三类3D脉像在第一谐波比值上可分离(PMID 31885653);1-D模型可个体化预测脉搏波形(PMID 21622820、32142412);机器学习可凭脉搏波分类疾病(PMID 41087424、39285363)。
部分验证: 1-D模型的”逐谐波能量分布”完整闭合检验尚无先例(#266已提案);脉象六分类的频域判别力只有小样本/单中心证据。
待验证: ①1-D模型谐波特征能否盲法判别六类脉象,AUC阈值与跨中心稳定性;②谐波特征中哪些维度对应哪些脉象要素(深度→低频能量?张力→衰减率?);③控制心率后”数/迟”类是否只剩频率信息、几何无关。
相关阅读
证据等级: B(脉象-血流动力学关联、脉象-谐波分离、1-D模型波形预测各有A级直接支撑;”1-D模型频谱×脉象分类盲法系统对照”本身无先例,属研究设计提案)
参考文献: PMID 26988431, PMID 31885653, PMID 21622820, PMID 32142412, PMID 41087424, PMID 39285363, PMID 22732024
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 the Pulse Spectrum Predicted by the 1-D Model Blindly Match Traditional Chinese Medicine’s Floating, Sunken, Wiry, Slippery, Rapid, and Delayed?
Evidence level: B
The harmonic energy spectrum calculated by vascular geometric models may represent the same information encoded differently as the pulse qualities (floating, sinking, wiry, slippery, rapid, slow) felt by TCM practitioners. Translating a “wiry pulse” into “elevated fundamental frequency energy ratio and reduced harmonic attenuation rate,” followed by blind validation to assess the accuracy of this translation.
Why This Question Deserves Investigation
The previous article, “Closed-Loop Validation of Harmonic Energy Distribution in 1-D Model Predicted Pulse Spectra via Non-Invasive Vascular Geometric Reconstruction,” demonstrated that reconstructing individual vascular geometry using ultrasound/MRI and feeding it into a 1-D hemodynamic model can calculate the pulse spectrum at the radial artery, theoretically approaching measured values in terms of harmonic energy distribution. However, “model accuracy” addresses only the physical layer. The unresolved issue lies at the clinical level: Does this physically interpretable harmonic energy spectrum correspond to the pulse category classifications felt beneath a TCM practitioner’s fingers?
TCM categorizes pulse qualities into over twenty types, including floating, sinking, wiry, slippery, rapid, and slow, relying on tactile integration of pressure, velocity, smoothness, and tension. Modern Pulse Wave Analysis (PWA) confirms these “sensations” are not illusions: a cross-sectional study on hypertensive patients showed significant multivariate associations between the three elements of floating/sinking (depth), strong/weak (force), and fast/slow (velocity) with hemodynamic indicators such as PWV, heart rate, and pulse pressure (PMID 26988431). Specifically, a floating pulse corresponds to a heart rate ≥58.25 bpm, while a rapid pulse corresponds to ≥69.6 bpm. Thus, TCM pulse elements have quantifiable counterparts in modern hemodynamics. What about the spectrum? Fourier analysis has shown significant differences in the first harmonic ratio among normal, wiry (taut), and slippery (slippery) 3D pulse images (PMID 31885653)—the first harmonic proportion is systematically higher in wiry pulses than in normal or slippery pulses. This suggests: Pulse category signatures exist in the frequency domain.
Bridging 1-D Models and Pulse Qualities
Viewing both aspects together clarifies the logic: ① The 1-D model calculates individualized harmonic energy spectra (patient-specific 1-D models significantly outperform generic models per PMID 21622820; personalized 0D-1D models show mean squared error of only 7.1 mmHg² and strong correlation with carotid flow per PMID 32142412); ② TCM pulse qualities possess separable signatures in the frequency domain (PMID 31885653). If the spectrum from ① can predict the classification in ②, it indicates that the tension and shape differences felt by TCM practitioners are essentially projections of vascular geometry onto harmonic structures—a “wiry pulse” represents a spectrum with elevated fundamental energy ratio and low harmonic attenuation, while a “slippery pulse” represents one with relatively prominent higher-order harmonics. This aligns with Wang Weigong’s framework of “harmonic correspondence to organs”: different harmonic orders carry distinct physiological information, and pulse qualities serve as tactile summaries of these harmonic proportions.
This is not mere speculation. Machine learning can already classify coronary heart disease, health, and pregnancy based solely on radial artery pulse waves with ~80% accuracy (PMID 41087424), reaching 91% with clinical parameters. However, these models are “black boxes”—they utilize extensive time-domain/time-frequency features for discrimination without explaining the underlying mechanisms. The unique value of the 1-D model lies in its harmonic features being entirely determined by physical parameters (vascular geometry, elasticity, boundary conditions), making them interpretable and traceable. If such a physically interpretable spectrum can blindly predict TCM pulse qualities, it is not merely “machine learning finding a statistical coincidence,” but rather obtaining independent validation for the hypothesis that “pulse quality equals the frequency-domain fingerprint of vascular geometry.”
Designing a Blind System Comparison
The core design involves Doctor Blind Rating × Model Blind Prediction × Discriminant AUC Decision:
- Acquisition Phase: For a cohort of subjects covering the six major pulse types (floating, sinking, wiry, slippery, rapid, slow) with ≥30 subjects per category, ≥3 senior TCM practitioners independently perform blind ratings (palpating only, unaware of other data). Subjects with kappa consistency ≥0.6 are enrolled. Simultaneously, perform vascular geometric reconstruction via radial/carotid ultrasound/MRI and collect measured pulse waves.
- Prediction Phase: Run the 1-D model using geometric reconstruction to output predicted harmonic features: fundamental energy ratio (C1 proportion), harmonic attenuation rate (energy decline slope from C2 to C10), and resonance peak locations (spectral peaks corresponding to harmonic orders).
- Comparison Phase: Perform discriminant modeling comparing doctor rating labels against model harmonic features to calculate Discriminant AUC, testing whether the model spectrum can blindly distinguish the six categories. All three outcomes yield knowledge:
- Significant Hit (AUC > 0.75 with cross-center reproducibility) → Physical model-level evidence supports the frequency-domain basis of pulse qualities, advancing from waveform to classification levels;
- Partial Hit (only hits rate-related categories like rapid/slow) → Only certain pulse categories (rate-related) can be explained by geometry-spectrum, while shape-related ones (floating/sinking/wiry/slippery) depend on factors beyond geometry (wall elasticity, microcirculatory boundary conditions);
- Complete Randomness → Pulse category encoding does not reflect spectral information; the hypothesis requires re-evaluation.
Differentiation from Existing Research
This design is not another “machine learning classification of pulse qualities,” as such work is abundant and already achieves high accuracy. Differentiation rests on three points: ① Feature Interpretability: Uses physical harmonic features from the 1-D model rather than black-box time-frequency features; ② Dual-Blind Design: Doctors are unaware of instrument results, and modelers are unaware of doctor ratings, preventing “rating-feature circular reasoning”; ③ Integration with Closed-Loop #266: The previous study proved model spectrum ≈ measured spectrum; this study proves measured spectrum ≈ pulse classification—linking the two rings completes the causal chain from “vascular geometry → harmonic spectrum → pulse classification.”
Evidence and Limitations
Verified: Significant associations between TCM pulse elements and hemodynamic indicators (PMID 26988431); separability of normal/wiry/slippery 3D pulse images in first harmonic ratios (PMID 31885653); ability of 1-D models to predict individual pulse waveforms (PMID 21622820, 32142412); machine learning capability to classify diseases based on pulse waves (PMID 41087424, 39285363).
Partially Verified: No prior precedent exists for a complete closed-loop validation of “harmonic energy distribution” by the 1-D model (proposed in #266); frequency-domain discriminative power for six pulse categories relies on small sample/single-center evidence.
To Be Verified: ① Whether 1-D model harmonic features can blindly discriminate six pulse types, including AUC thresholds and cross-center stability; ② Which dimensions of harmonic features correspond to which pulse elements (depth → low-frequency energy? tension → attenuation rate?); ③ Whether “rapid/slow” categories rely solely on frequency information and are geometry-independent after controlling for heart rate.
Related Reading
- Vascular Geometry 1-D Model Closed-Loop Validation
- Harmonic Energy Distribution Closed-Loop Validation
- Circadian and Seasonal Stability of Pulse Spectra
- Research Pool: Environment-Life Frequency Interface
Evidence Level: B (Associations between pulse qualities and hemodynamics, pulse quality-harmonic separation, and 1-D model waveform prediction each have direct A-level support; the “1-D model spectrum × pulse classification blind system comparison” itself is a novel research design proposal)
References: PMID 26988431, PMID 31885653, PMID 21622820, PMID 32142412, PMID 41087424, PMID 39285363, PMID 22732024
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.
Leave a Reply