一根桡动脉能候几个脏腑:两套脉位映射的跨疾病泛化检验

Written by

in

Language

Evidence level: B

现代研究已证明:单凭一根桡动脉的脉波,机器学习就能判别冠心病、肾功能不全、肝功能异常等多种疾病。这带来一个尖锐问题——如果”全局脉波”什么病都能判,”脉位-脏腑”的专属分配还有独立价值吗?

从”单一疾病”到”多目标矩阵”

此前的判别增量检验都围绕单一目标展开。但传统脉诊的根本主张不是”脉能判病”——那在现代已不算新鲜事——而是”哪个位置候哪个脏腑“。要检验这个主张,必须在同一批数据上同时面对多个目标脏腑,看位置特异模型是否对”自己该候的脏腑”判别力最高。

这就是跨疾病泛化检验:把心、肺、肝胆、脾胃、肾五类脏腑状态作为独立金标准,在统一的六点脉搏数据上,逐脏腑建立”藏医指派点位模型”与”中医指派点位模型”,得到一张 5×5 判别 AUC 矩阵。检验的核心是矩阵的对角线优势:如果位置映射是真实生理信息,矩阵应对角占优——藏医左手甘位候脾,它在判别脾胃障碍时应胜过判别肝胆障碍;如果位置分配只是操作习惯,矩阵会近乎均匀。

先看全局证据:单点脉波确实能判多病

“多目标”不是天方夜谭,同一类桡动脉脉波信号已在多个器官系统分别取得战绩。心血管方向证据最密:2024 年 608 人研究用桡动脉脉波判别健康/高血压/冠心病,Extra Trees 测试集 AUC 0.936(PMID 39285363);2025 年 LSTM 研究在健康、冠心病、心衰三态间判别,平均准确率 0.86(PMID 40866986);2012 年 531 人前瞻性 ARM-CAD 研究证实桡动脉脉波可独立预测冠脉狭窄,AUC 0.876(PMID 22483418)。

跳出心血管,肾与肝也有直接证据:2023 年一项前瞻研究用腕式血压计脉波提取 156 个谐波特征判别肾功能不全,随机森林准确率 0.99、Kappa 0.94(PMID 37969185);1996 年双盲研究用脉谱谐波分析判别工人肝功能异常(SGOT/SGPT 等血检),与肝经脉诊一致性 Kappa 0.61(PMID 8874678)。肾功能不全患者中脉波还能判别冠脉钙化严重度——2025 年 58 名透析患者 GBDT 模型宏 AUC 0.962(PMID 40883745)。

共同点值得注意:这些研究大多只采集了单一位置的脉波,却判出了多个系统的疾病。这直接证明”全局脉波判别力”很强——也让位置维度的价值更需要单独证明:如果随便在桡动脉上放个传感器就能判出肝病肾病,”必须在甘位才能候脾胃”凭什么成立?

对角线优势:把问题变成可检验的假说

答案藏在矩阵的形状里。第一步,入组五类目标脏器的功能障碍队列(冠心病/慢阻肺/肝胆病/脾胃病/肾病各一组,健康对照),金标准用影像与生化。第二步,统一六点采集协议,逐脏腑、逐位置分别建模(藏医映射:冲候肺/大肠、甘候脾胃/肝胆、恰候肾与膀胱;中医映射:寸关尺按各自脏腑表),得到判别 AUC 矩阵。第三步,检验对角线元素是否显著高于同行/同列其余元素,计算”对角线强度指数”(对角线均值 − 非对角线均值)。

三种结局对应三种结论:矩阵强对角占优 → 位置-脏腑映射是真实生理信息;矩阵均匀 → 位置是冗余编码,”按哪里”不如”按多重”或波形细节重要;矩阵交叉占优(藏医甘位反而更判肝胆、中医关位更判脾胃)→ 两套映射在竞争同一解剖底座时各有强弱,值得逐位对照。

混杂与边界

多目标设计有一个天然混杂:疾病叠加。冠心病患者常合并肾功能不全,判”心”时”肾”的信号可能渗入。对策是逐脏腑做多病共存协变量建模,或用”仅该系统异常”的纯组做验证。另一个边界是样本量——5×5 矩阵每格都要足够样本,通常需数百人。还有”镜像陷阱”:若所有位置判所有疾病的 AUC 都接近 0.9,只说明全局脉波很强,不能说明位置映射弱——必须用增量检验(叠加位置信息后 ΔAUC/NRI)区分”位置独立贡献”与”全局波形共享信息”。

证据与局限

已验证: 单一桡动脉位置的脉波可判别多种系统疾病——冠心病(AUC 0.936,39285363;AUC 0.876,22483418)、心衰(0.86,40866986)、肾功能不全(准确率 0.99,37969185)、肝功能异常(Kappa 0.61,8874678)、冠脉钙化(宏 AUC 0.962,40883745)。

部分验证: “位置特异模型”在单一目标上的判别增量已被间接支持,但从未有人做多目标、两体系、逐格对照的 5×5 矩阵检验;对角线优势假说尚无直接实证。

待验证: 两套脉位映射在五类目标脏腑上的判别 AUC 矩阵形状;对角线强度指数是否显著 >0;分歧位(甘/关)的交叉占优方向;多病共存的混杂量化;位置判别增量与全局波形共享信息的分离。

相关阅读

证据等级: B(”全局脉波判别多病”有大量直接证据,但”位置特异映射的跨脏腑对角线优势”尚无直接实证,属实验设计提案)
参考文献: PMID 39285363, PMID 40866986, PMID 22483418, PMID 37969185, PMID 8874678, PMID 40883745


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。

One Radial Artery Can Diagnose Multiple Organs: Cross-Disease Generalization Validation Based on Dual Pulse Position Mapping

Evidence level: B

Modern research has demonstrated that machine learning algorithms can discriminate various diseases—including coronary heart disease, renal insufficiency, and hepatic dysfunction—solely from the pulse wave of a single radial artery. This raises a critical question: If a “global pulse wave” can identify multiple conditions, does the exclusive allocation of specific organs to specific pulse positions retain independent value?

From “Single-Disease” to “Multi-Target Matrix”

Previous discriminative increment tests focused on single targets. However, the fundamental claim of traditional pulse diagnosis is not merely “the pulse can diagnose disease”—a concept no longer novel in modern contexts—but rather “which position corresponds to which organ.” To validate this claim, one must simultaneously evaluate multiple target organs using the same dataset to determine whether position-specific models achieve highest discriminative power for “their assigned organs.”

This constitutes a cross-disease generalization test: treating the status of five organ categories (heart, lung, liver/gallbladder, spleen/stomach, kidney) as independent gold standards on unified six-point pulse data. We establish a “Tibetan medicine assigned position model” and a “Traditional Chinese Medicine (TCM) assigned position model” for each organ, generating a 5×5 discrimination AUC matrix. The core of the validation lies in the diagonal dominance of this matrix: if position mapping reflects genuine physiological information, the matrix should exhibit diagonal superiority—e.g., the Tibetan “Gan” position (assigned to Spleen/Stomach) should outperform in discriminating Spleen/Stomach disorders compared to Liver/Gallbladder disorders. Conversely, if position allocation is merely an operational habit, the matrix will appear nearly uniform.

Global Evidence First: Single-Point Pulse Waves Do Discriminate Multiple Diseases

The concept of “multi-target” is not unfounded; pulse wave signals from the same radial artery have already demonstrated efficacy across multiple organ systems. Evidence is densest in the cardiovascular domain: a 2024 study involving 608 participants used radial pulse waves to discriminate health, hypertension, and coronary heart disease, achieving an Extra Trees test set AUC of 0.936 (PMID 39285363); a 2025 LSTM study discriminated between three states (health, coronary heart disease, heart failure) with an average accuracy of 0.86 (PMID 40866986); and a 2012 prospective ARM-CAD study with 531 participants confirmed that radial pulse waves can independently predict coronary stenosis with an AUC of 0.876 (PMID 22483418).

Moving beyond cardiovascular systems, direct evidence exists for kidney and liver function: a 2023 prospective study extracted 156 harmonic features from wrist-based sphygmomanometer pulse waves to discriminate renal insufficiency, achieving a Random Forest accuracy of 0.99 and Kappa of 0.94 (PMID 37969185); a 1996 double-blind study used pulse spectral harmonic analysis to discriminate hepatic dysfunction in workers (via blood tests such as SGOT/SGPT), showing consistency with liver meridian pulse diagnosis with a Kappa of 0.61 (PMID 8874678). Furthermore, pulse waves can discriminate the severity of coronary artery calcification in patients with renal insufficiency—a 2025 study of 58 dialysis patients found a GBDT model macro AUC of 0.962 (PMID 40883745).

A notable commonality: these studies predominantly collected pulse waves from a single position yet discriminated diseases across multiple systems. This directly proves the strong discriminative power of “global pulse waves,” thereby necessitating separate validation for positional value. If a sensor placed arbitrarily on the radial artery can identify liver or kidney disease, what justifies the assertion that “Spleen/Stomach must be diagnosed at the Gan position”?

Diagonal Dominance: Transforming the Problem into a Testable Hypothesis

The answer lies within the shape of the matrix. Step 1: Enroll cohorts with functional disorders in five target organ categories (coronary heart disease/COPD/liver/gallbladder diseases/spleen/stomach diseases/kidney disease, each with a healthy control group), using imaging and biochemical markers as gold standards. Step 2: Adopt a unified six-point collection protocol to build models separately for each organ and position (Tibetan mapping: Chong corresponds to Lung/Colon, Gan to Spleen/Stomach/Liver/Gallbladder, Cha to Kidney/Bladder; TCM mapping: Cun/Guan/Chi correspond to respective organ tables), yielding the discrimination AUC matrix. Step 3: Test whether diagonal elements are significantly higher than other elements in the same row or column, calculating a “Diagonal Strength Index” (mean of diagonal elements minus mean of off-diagonal elements).

Three outcomes correspond to three conclusions: Strong diagonal dominance implies position-organ mapping reflects genuine physiological information; a uniform matrix suggests positions are redundant encodings, where “where to press” is less critical than “pressing multiple points” or waveform details; and cross-diagonal dominance (e.g., Tibetan Gan position performs better for Liver/Gallbladder, TCM Guan position better for Spleen/Stomach) implies that both mapping systems compete for the same anatomical foundation with varying strengths, warranting position-by-position comparison.

Confounding and Boundaries

Multi-target design introduces a natural confounder: comorbid disease. Patients with coronary heart disease often coexist with renal insufficiency, potentially causing “kidney” signals to infiltrate when diagnosing “heart.” The countermeasure is to build covariate models for comorbidities per organ or use “pure groups” where only the specific system is abnormal for validation. Another boundary is sample size—the 5×5 matrix requires sufficient samples per cell, typically necessitating hundreds of participants. There is also the “mirror trap”: if all positions discriminate all diseases with AUCs near 0.9, it only indicates strong global pulse wave capability, not weak position mapping. Incremental testing (calculating ΔAUC/NRI after adding positional information) must be used to distinguish “independent contribution of position” from “shared waveform information.”

Evidence and Limitations

Verified: Pulse waves from a single radial artery position can discriminate diseases across multiple systems—coronary heart disease (AUC 0.936, PMID 39285363; AUC 0.876, PMID 22483418), heart failure (0.86, PMID 40866986), renal insufficiency (accuracy 0.99, PMID 37969185), hepatic dysfunction (Kappa 0.61, PMID 8874678), and coronary artery calcification (macro AUC 0.962, PMID 40883745).

Partially Verified: Incremental discriminative power of “position-specific models” on single targets has been indirectly supported, but no one has conducted a 5×5 matrix comparison across multiple targets and two systems; the hypothesis of diagonal dominance lacks direct empirical evidence.

To Be Verified: The shape of the discrimination AUC matrix for both pulse position mapping systems across five target organ categories; whether the Diagonal Strength Index is significantly >0; the direction of cross-diagonal dominance at divergent positions (Gan/Guan); quantification of confounding due to comorbidities; and the separation of positional discriminative increments from global waveform shared information.

Related Reading

Evidence Level: B (“Global pulse wave discrimination of multiple diseases” is supported by substantial direct evidence, but “cross-organ diagonal dominance of position-specific mappings” lacks direct empirical evidence and represents an experimental design proposal)
References: PMID 39285363, PMID 40866986, PMID 22483418, PMID 37969185, PMID 8874678, PMID 40883745


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 *