Evidence level: C
把「这套脉位映射到底有没有点位—脏器特异对应」从一场文献争论,变成一个可以算出显著性与效应量的数:对角线强度指数。
正文
汉医寸关尺分部候脏与藏医多点候脏,是两个独立发展出来的「手腕分区—内脏状态」查表系统。两套系统各有一张映射表,也各有自己的辩护者。历史上裁决这类争论的办法是引经据典,但经可以各引各的。本文提出的裁决方式不同:不投票,做判别——把两套映射各自变成一个 AUC 矩阵,再用一个统计量直接对比它们。
第一步,建矩阵。 设某套映射有 5 个脉位(寸、关、尺 × 双手,藏医侧取其对应五点位),每点位对应一个理论脏器。在每个点位单独采集脉学特征(时域 + 频域特征),训练「脏器状态判别器」,然后让每个点位的判别器去判所有脏器标签,得到一个 5×5 的判别 AUC 矩阵。矩阵对角线是「点位判它理论对应的脏器」,非对角线是「点位判别的脏器与映射表不符」。若映射真有特异对应,对角线应系统性地高于非对角线;若只是脉波整体携带全身信息、点位无关,则矩阵应接近均匀。
第二步,定义指数。 对角线强度指数 D = 对角线均值 − 非对角线均值。D=0 表示点位与脏器配对完全无信息;D 越大,映射的位置特异对应越强。这个指数的好处是把 25 个格子(5×5)压缩成一个数,避免逐格检验带来的多重比较爆炸——25 次检验在 5% 水平下期望产生约 1.25 个假阳性,逐格看星号基本是噪声放大器。
第三步,三层推断。 ① 排列检验:打乱「点位—脏器」配对关系(保持矩阵结构、只重排映射表本身)若干千次,每次重算 D,得到零分布,观察值落在分布外的比例即经验 p 值——这是对「配对本身是否比随机配对好」的直接检验,AUC 的排列推断已有统一理论框架(PMID 27502845,学生化排列区间在小样本下覆盖率良好)。② 自助法置信区间:以受试者而非记录为重采样单元,重算 D 的 95% 置信区间——脉图数据是同体多点位记录,天然聚簇,逐条记录重采样会低估方差(PMID 10845400 在聚簇诊断数据上证明受试者层面自助的覆盖率更优)。③ 随机点位基准:用随机指定的伪映射算一遍 D,作为「无映射时应看到什么」的具象参照。
第四步,双矩阵头对头。 两套映射在同一批受试者、同一次采集中各算一个 D,记 D1(汉)与 D2(藏)。ΔD = D1 − D2 做配对自助检验:同一受试者的两次重采样必须配对,否则方差被重复计入。这一步回答的不是「有没有对应」,而是「哪套映射的配对更特异」——一个纯数据层的问题,文献投票答不了。
数据层的前提已经具备。三点位同步采集的硬件已有多个独立实现:三柔性传感器阵列可同时读寸关尺(PMID 36296052),三部九候同步测力传感器已通过国际标准校验、与临床专家判读一致率 92.3%(PMID 33825371);脉波特征的判别能力也已被反复证实,机器学习模型对高血压判别 AUC 达 0.83–0.86(PMID 30534557),深度学习脉象分类准确率 91.68%(PMID 40925696)。也就是说,缺的不是信号、不是判别器,而是这套「映射表本身」的统计裁决设计——这正是对角线强度指数要补的位置。
需要注意的陷阱有三个。其一,两点位解剖距离只有几厘米,脉波信号空间相关,矩阵格子之间不独立,所以检验对象必须是聚合后的 D 而非单格;其二,类别不平衡时 AUC 虽然稳健,排列时也必须保持每类样本数固定,否则零分布被污染;其三,D 显著不等于「脏器对应」成立,它只证明「这张映射表的配对优于随机配对」,机制解释仍须另行检验。
L1 网络状态:该主题尚未进入 L1 对应网络(CORRESPONDENCE_NETWORK 与 EDGE_TABLE 中均无「脉位映射判别 AUC 矩阵 / 对角线强度指数」对应边;EDGE_TABLE 中脉象相关仅有一条 PMID 21622820 的盲分类边,属不同主题)。
证据与局限
已验证:脉波特征判别疾病/生理状态可行(PMID 30534557、PMID 40925696);三点位同步采集硬件成熟且标准化(PMID 36296052、PMID 33825371);AUC 排列推断与聚簇数据自助区间的方法学基础扎实(PMID 27502845、PMID 10845400)。
部分验证:判别 AUC 矩阵结构在其他诊断领域(如影像组学)常见,用于脉位映射裁决尚未有已发表实例。
待验证:D 指数对两套真实映射的区分力;对角线-非对角线差异能否在跨中心数据中复现;ΔD 的配对自助检验功效。
相关阅读
证据等级: C
参考文献: PMID 27502845, PMID 10845400, PMID 36296052, PMID 33825371, PMID 30534557, PMID 40925696
证据等级:A=有PubMed直接支撑,B=科学类比/间接证据,C=传统文化共识/初步探索
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。
Statistical Test Design for the \”Diagonal Strength Index\” of AUC Matrices in Two Pulse Position Mapping Discrimination
Evidence level: C
The “Diagonal Strength Index” transforms the scholarly debate over whether this pulse-location mapping possesses specific point–organ correspondences into a quantitative measure of significance and effect size.
Statistical Framework for Diagonal Strength
| Component | Description |
|---|---|
| Matrix Structure | AUC (Area Under the Curve) matrix mapping pulse locations to organ functions. |
| Diagonal Elements | Represent the strength of direct, specific correspondences (e.g., specific pulse point → specific organ). |
| Off-Diagonal Elements | Represent non-specific or diffuse associations. |
| Diagonal Strength Index (DSI) | A composite statistic quantifying the dominance of diagonal elements over off-diagonal ones. |
Calculation Methodology
The Diagonal Strength Index is derived from the ratio of diagonal to total matrix values, adjusted for expected background noise:
$$
DSI = \frac{\sum_{i} AUC_{ii}}{\sum_{i,j} AUC_{ij}} \times \left( 1 – \frac{\sigma_{noise}}{\mu_{diagonal}} \right)
$$
Where:
* $AUC_{ii}$ = Area Under the Curve for diagonal element $i$.
* $\sigma_{noise}$ = Standard deviation of off-diagonal elements (background noise).
* $\mu_{diagonal}$ = Mean of diagonal elements.
Evidence and Limitations
- Evidence Level: A (Direct calculation from matrix data).
- Traditional Consensus: The concept of specific point–organ mapping is a traditional consensus in Yijing Studies, though the degree of specificity varies.
- Structural Analogy: This approach uses a structural analogy to medical imaging, where the “diagonal” represents high-resolution focus and the “off-diagonal” represents background radiation.
Limitations
- Sensitivity to Noise: High off-diagonal values (diffuse associations) can artificially inflate the denominator, reducing the DSI.
- Assumption of Linearity: The current model assumes linear relationships between pulse and organ, which may not hold for complex physiological interactions.
- Sample Size: Requires a sufficiently large dataset of pulse readings to accurately estimate the AUC values.
Body
The Han medical寸关尺分部候脏 (wrist partitioning for organ assessment) and the Tibetan medical多点候脏 (multi-point organ assessment) are two independent mapping systems developed separately for assessing internal organ states via wrist pulse locations. Each system has its own lookup table and its own proponents. Historically, such disputes were resolved by citing classical texts, but each side could cite different authorities. This paper proposes a different resolution method: instead of voting, we perform discrimination by converting each mapping into an AUC matrix and directly comparing them using a single statistic.
Step 1, Build the matrix. Suppose a mapping has 5 pulse locations (寸关尺 × both hands; for Tibetan medicine, take the corresponding five points). Each location corresponds to a theoretical organ. At each location, collect pulse features (time-domain + frequency-domain features), train an「organ state discriminator」, then let each location’s discriminator classify all organ labels, yielding a 5×5 discrimination AUC matrix. The diagonal represents「the location judging its theoretically corresponding organ」; off-diagonal elements represent「the organ judged by the location does not match the mapping table」. If the mapping truly has specific correspondence, the diagonal should systematically exceed off-diagonal values; if pulse waves carry whole-body information regardless of location, the matrix should be nearly uniform.
Step 2, Define the index. Diagonal strength index D = mean diagonal − mean off-diagonal. D=0 indicates no information in the location-organ pairing; larger D implies stronger positional specificity of the mapping. This index compresses the 25 cells (5×5) into one number, avoiding multiple comparison explosion from cell-by-cell testing—25 tests at the 5% level would expect about 1.25 false positives; inspecting stars cell-by-cell essentially amplifies noise.
Step 3, Three-tier inference. ① Permutation test: randomly shuffle the「location-organ」pairings (keeping matrix structure, only reordering the mapping table itself) thousands of times, recalculate D each time to obtain the null distribution; the proportion of observed values falling outside this distribution gives the empirical p-value—this directly tests「whether the pairing itself is better than random pairing」. AUC permutation inference already has a unified theoretical framework (PMID 27502845; studentized permutation intervals have good coverage in small samples). ② Bootstrap confidence interval: resample by subject rather than record to compute the 95% CI for D—pulse chart data are inherently clustered as multiple locations per subject; resampling individual records would underestimate variance (PMID 10845400 demonstrates subject-level bootstrap has better coverage for clustered diagnostic data). ③ Random location baseline: compute D using pseudo-mappings with randomly assigned locations as a concrete reference for「what should be seen if there were no mapping」.
Step 4, Head-to-head comparison of two matrices. For both mappings applied to the same cohort of subjects and single pulse acquisition, compute D1 (Han) and D2 (Tibetan). ΔD = D1 − D2 undergoes paired bootstrap testing: resampled data from the same subject must be paired; otherwise variance is double-counted. This step answers not「whether correspondence exists」, but「which mapping shows stronger specific pairing」—a pure data-layer question that literature voting cannot resolve.
The data-layer prerequisites are already in place. Hardware for synchronized three-location pulse acquisition exists with multiple independent implementations: a tri-flexible sensor array can read寸关尺 simultaneously (PMID 36296052); the three-section nine-position synchronous force sensor has passed international standard calibration and matches clinical expert judgment with 92.3% agreement (PMID 33825371). Pulse feature discriminative power has been repeatedly validated: machine learning models achieve hypertension discrimination AUC of 0.83–0.86 (PMID 30534557), and deep learning pulse classification accuracy reaches 91.68% (PMID 40925696). In short, what is missing is not the signal, nor the discriminator, but the statistical decision design for「the mapping table itself」—this is precisely where the diagonal strength index fills the gap.
Three pitfalls require attention. First, since only a few centimeters separate two pulse locations, pulse wave signals are spatially correlated; matrix cells are not independent, so the test object must be the aggregated D rather than individual cells. Second, under class imbalance, although AUC remains robust, permutation testing must keep sample counts fixed per class; otherwise the null distribution becomes contaminated. Third, significant D does not prove「organ correspondence」holds; it only demonstrates「this mapping table’s pairings outperform random pairings」, and mechanistic explanation requires separate validation.
L1 Network Status: This topic has not yet entered the L1 correspondence network (neither the CORRESPONDENCE_NETWORK nor the EDGE_TABLE contains edges corresponding to「pulse location mapping discrimination AUC matrix / diagonal strength index」; the EDGE_TABLE contains only one blind classification edge related to pulse patterns, PMID 21622820, which belongs to a different topic).
Evidence and Limitations
Verified: Distinguishing disease/physiological states via pulse wave characteristics is feasible (PMID 30534557, PMID 40925696); three-point synchronized acquisition hardware is mature and standardized (PMID 36296052, PMID 33825371); the methodological foundation for AUC-based ranking inference and clustering data bootstrap intervals is solid (PMID 27502845, PMID 10845400).
Partially Verified: AUC matrix structure discrimination is common in other diagnostic fields (e.g., radiomics); however, no published instances exist yet for applying this to pulse-position mapping decisions.
Pending Verification: The discriminative power of the D index between two true mappings; whether diagonal-non-diagonal differences can be replicated across-center data; and the power of ΔD paired bootstrap testing.
Related Reading
- Single-point discriminative power of the six-position specificity model across organ heterogeneity
- Cross-center/cross-instrument generalization stability of the six-position specificity model
- HRV response convergence in Han individuals with lung-type constitution under acute high-altitude exposure
- Research pool: cross-cultural frequency cognition
Evidence level: C
References: PMID 27502845, PMID 10845400, PMID 36296052, PMID 33825371, PMID 30534557, PMID 40925696
Evidence level: A = direct PubMed support; B = scientific analogy/indirect evidence; C = traditional consensus/preliminary exploration.
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