Evidence level: B
藏医断言左右手腕六个触诊点各自”候”不同的脏腑——现代脉搏波机器学习和多通道阵列测量技术,已经可以把这句断言翻译成一道可以统计检验的题目:用影像和生化检查作金标准,看六点信号到底能不能盲法判别出对应的脏腑功能障碍。
从”按哪里”到”候什么”:一张可以检验的脏腑分配表
藏医脉诊在左右腕桡动脉各定”冲、甘、恰”三个触诊点,共六点,每点候两个脏腑:左手冲候心与小肠、甘候脾与胃、恰候左肾与生殖器;右手冲候肺与大肠、甘候肝与胆、恰候右肾与膀胱。三指按压力度也随点位不同——冲轻按、甘中按、恰重按。这个体系的独特之处,不在于”脉搏波能反映疾病”(这一点中医和现代医学都同意),而在于“位置-脏腑”的固定分配:肝出问题,应当优先在右甘位暴露出来;肾出问题,应当在恰位暴露出来。
这个断言在机器学习时代变成了可操作的问题:如果位置映射是真的,那么”用右甘位一个点的信号判别肝功能异常”应该显著优于”随便用哪个点”;如果映射是假的,六点信号不过是对同一血流动力学状态的重复采样,位置特异性就带不来任何判别增量。俄罗斯学者 Pupysheva 与 Boronoev 早在 2020 年就论证过藏医脉诊计算机化的可行性——用脉诊设备在可控刺激下采集波形特征,确认”把主观触诊变成客观测量”在原理上成立(PMID 33303385)。
已验证的半座桥:单点脉搏波确实能判别疾病
先看基础事实:桡动脉脉搏波携带疾病信息,且机器学习能把它提取出来。2024 年一项 608 人的研究用桡动脉脉搏波判别冠心病,Extra Trees 分类器在测试集上准确率 85.8%、AUC 0.936(PMID 39285363)。肾病患者同样成立——2025 年一项研究用 58 名血液透析患者的桡动脉脉搏波判别冠状动脉钙化严重度(以低剂量 CT 的 Agatston 评分为金标准),GBDT 模型五折交叉验证平均准确率 84.1%、宏 AUC 0.962(PMID 40883745)。更早的病例对照研究也确认,桡动脉脉搏波特征可以识别冠状动脉病变(PMID 35005017)。
另一条线是”谐波-脏腑”对应。王唯工学派的核心假说认为,脉搏波频谱的各次谐波与不同器官的血流动力学共振相关。一项大鼠实验提供了最直接的证据:给予大剂量对乙酰氨基酚造成急性肝损伤后,血压脉搏波第一谐波幅度 A1 与血清转氨酶(SGPT/SGOT)显著相关——A1 低于阈值且 SGPT 升高时 kappa 值约 0.6,统计学检验 p>0.999(PMID 20021158)。2023 年一篇纳入 50 项研究的系统综述进一步总结:心血管病患者第一谐波(对应肝经)升高、糖尿病患者第三谐波(对应脾经)减弱、肝硬化患者呈现特征性谐波模式(PMID 38481553)。
技术底座:六点多通道同步采集已经成熟
藏医”三指同时按六点”在物理上有一个天然难点——手指不可能完全同步、等压地覆盖六个位置。多通道阵列传感器正好解决这个问题。2023 年发表在 ACS Nano 的一项研究展示了 27 通道(3×9)高密度柔性脉搏感知阵列,可以一次提取脉搏的空间-时间分布信息,自动识别率高达 97.8%(PMID 36716225)。这种”一块贴片覆盖整个诊脉区”的技术,让左右各三点同时采集、记录每点的按压深度和空间梯度成为工程上可行的常规操作。
待验证的半座桥:盲法判别模型的落地设计
把两块桥墩接起来,需要一套严格的三步实验设计:
第一步,定金标准。 脏腑功能障碍必须用独立客观指标定义,且金标准判定者不知道脉诊结果。肝功能取血清 ALT/AST 或超声/弹性成像;肾功能取 eGFR/肌酐或影像;心功能取超声心动。注意:金标准不能与脉诊共享任何测量通道,否则”判别成功”可能是仪器间的虚假关联。
第二步,建两套模型、做头对头对比。 同一批六点数据,分别构建”位置特异模型”(只用藏医指派给该脏腑的那个点位,如肝功能只用右甘位)与”全局模型”(六点全量特征)。关键统计检验是:位置特异模型的判别 AUC 是否显著高于”随机选一个点”的基准线。如果右甘位对肝的判别力确实优于右冲位,藏医的”位置-脏腑”映射就拿到了第一块直接证据;如果六个点判别力无差异,则”位置映射”更像文化约定而非生理事实——这同样是有价值的结论。
第三步,控住三个坑。 一是标签噪声:脏腑功能障碍有程度差异,灰色地带样本应单独建模或按亚组分析,避免”判别力被边缘病例稀释”;二是按压深度:藏医冲轻按、恰重按,深度本身改变波形谱(PMID 30105076),必须把每点实际按压压力记录为协变量;三是左右手系统差异:左右本就有系统性偏差,必须分开建模再报告。
证据与局限
已验证: 桡动脉脉搏波经机器学习可判别冠心病(AUC 0.936)、肾病钙化(AUC 0.962)等疾病;脉搏波第一谐波与急性肝损伤的生化指标显著相关(动物实验);多通道阵列技术可同步采集六点以上脉搏信号;藏医脉诊计算机化原理上可行。
部分验证: “谐波-脏腑”对应有系统综述支持但方法学异质性强;中医体系(寸关尺)的位置特异判别有零星证据,藏医体系(冲甘恰)的脏腑分配尚未被任何判别研究直接检验。
待验证: 藏医六点”位置-脏腑”映射是否能转化为盲法判别增量;位置特异模型 vs 全局模型的 AUC 对比;按压深度梯度作为判别特征的独立价值。
相关阅读
证据等级: B(”脉搏波+ML判别疾病”有直接证据,”谐波-脏腑”对应有间接证据,但”藏医六点位置-脏腑映射的盲法判别”整体尚无直接实证)
参考文献: PMID 39285363, PMID 40883745, PMID 20021158, PMID 36716225, PMID 33303385, PMID 38481553, PMID 35005017, PMID 30105076
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。
Blind Discrimination Model for the Six-Pulse Diagnosis in Traditional Tibetan Medicine: Statistical Validation of \”Where to Press Corresponds to What\
Evidence level: B
Tibetan medicine asserts that each of the six palpation points on the wrists corresponds to specific organs. Modern pulse wave machine learning and multi-channel array measurement technologies have translated this assertion into a statistically testable question: using imaging and biochemical tests as the gold standard, can the six-point signals blindly discriminate corresponding organ dysfunction?
From “Where to Press” to “What to Await”: A Verifiable Organ Allocation Table
In Tibetan medicine, three palpation points—”Chong,” “Gan,” and “Cha”—are defined on the radial artery of each wrist, totaling six points. Each point corresponds to two organs: the left Chong awaits the heart and small intestine; Gan awaits the spleen and stomach; Cha awaits the left kidney and reproductive organs. The right Chong awaits the lung and large intestine; Gan awaits the liver and gallbladder; Cha awaits the right kidney and bladder. The pressure applied by the three fingers also varies by location—light for Chong, medium for Gan, and heavy for Cha. The uniqueness of this system lies not in the premise that “pulse waves reflect disease” (a point agreed upon by both Traditional Chinese Medicine and modern medicine), but in the fixed allocation of location to organs: liver dysfunction should manifest primarily at the right Gan position; kidney dysfunction should appear at the Cha position.
In the era of machine learning, this assertion becomes an operational problem: if the positional mapping is true, then “discriminating liver function abnormalities using a single signal from the right Gan point” should yield significantly better results than “randomly selecting any point”; if the mapping is false, the six-point signals merely represent repeated sampling of the same hemodynamic state, offering no discriminative increment through positional specificity. Russian scholars Pupysheva and Boronoev argued early in 2020 regarding the feasibility of computerizing Tibetan pulse diagnosis—demonstrating that capturing waveform features under controlled stimuli with pulse diagnosis equipment confirms the principle that “subjective palpation can be transformed into objective measurement” (PMID 33303385).
A Verified Half-Bridge: Single-Point Pulse Waves Indeed Discriminate Disease
First, establish the baseline fact: radial artery pulse waves carry disease information, and machine learning can extract it. A 2024 study involving 608 individuals used radial artery pulse waves to discriminate coronary heart disease, where an Extra Trees classifier achieved 85.8% accuracy and an AUC of 0.936 on the test set (PMID 39285363). Similarly, in kidney disease patients—a 2025 study utilized radial artery pulse waves from 58 hemodialysis patients to discriminate the severity of coronary artery calcification (using low-dose CT Agatston scores as the gold standard), where a GBDT model showed 84.1% average accuracy across five-fold cross-validation and a macro AUC of 0.962 (PMID 40883745). Earlier case-control studies also confirmed that radial artery pulse wave features can identify coronary lesions (PMID 35005017).
Another line of inquiry is the “Harmonic-Organ” correspondence. The Wang Wei engineering school’s core hypothesis posits that various harmonics of the pulse wave spectrum correlate with hemodynamic resonance in different organs. A rat experiment provided the most direct evidence: after inducing acute liver injury with a high dose of acetaminophen, the amplitude of the first harmonic (A1) of the blood pressure pulse wave showed a significant correlation with serum transaminases (SGPT/SGOT)—when A1 fell below the threshold and SGPT rose, the kappa value was approximately 0.6 with a statistical significance of p>0.999 (PMID 20021158). A 2023 systematic review encompassing 50 studies further summarized that the first harmonic (corresponding to the liver meridian) is elevated in cardiovascular disease patients, the third harmonic (corresponding to the spleen meridian) is weakened in diabetic patients, and liver cirrhosis patients exhibit characteristic harmonic patterns (PMID 38481553).
Technical Foundation: Multi-Channel Synchronous Acquisition of Six Points is Mature
The Tibetan medicine practice of “pressing six points simultaneously with three fingers” presents a physical challenge—fingers cannot perfectly synchronize and apply equal pressure across six locations. Multi-channel array sensors address this issue directly. A 2023 study published in ACS Nano demonstrated a high-density flexible pulse sensing array with 27 channels (3×9), capable of extracting spatial-temporal distribution information of pulse waves in a single operation, achieving an automatic recognition rate as high as 97.8% (PMID 36716225). This technology, involving “a single patch covering the entire diagnosis area,” makes it an engineering routine to simultaneously collect data from three points on each side, recording the pressing depth and spatial gradient for each point.
A Pending Half-Bridge: Designing the Blind Discrimination Model
Connecting the two bridge pillars requires a rigorous three-step experimental design:
Step 1: Define the Gold Standard. Organ dysfunction must be defined by independent objective indicators, and the gold standard adjudicator must be unaware of the pulse diagnosis results. For liver function, use serum ALT/AST or ultrasound/elastography; for kidney function, use eGFR/creatinine or imaging; for heart function, use echocardiography. Note: The gold standard must not share any measurement channels with the pulse diagnosis, otherwise “discrimination success” may reflect spurious correlation between instruments.
Step 2: Build Two Models and Conduct Head-to-Head Comparison. Using the same set of six-point data, construct a “Position-Specific Model” (using only the point assigned to that organ according to Tibetan medicine, e.g., right Gan for liver function) and a “Global Model” (using all six points). The key statistical test is whether the discriminative AUC of the position-specific model is significantly higher than the baseline of “randomly selecting one point.” If the discriminative power of the right Gan position for the liver is indeed superior to the right Chong position, Tibetan medicine’s “location-organ” mapping receives direct evidence; if there is no difference in discriminative power across the six points, then “positional mapping” resembles a cultural convention rather than physiological fact—which is also a valuable conclusion.
Step 3: Control for Three Pitfalls. First, label noise: organ dysfunction varies in degree; samples in the gray zone should be modeled separately or analyzed by subgroup to avoid “dilution of discriminative power by edge cases.” Second, pressing depth: Tibetan medicine uses light pressure for Chong and heavy pressure for Cha; depth itself alters the waveform spectrum (PMID 30105076), so actual pressing pressure for each point must be recorded as a covariate. Third, inter-hand system differences: there are inherent systematic biases between left and right hands; models must be built separately for each hand before reporting.
Evidence and Limitations
Verified: Radial artery pulse waves can discriminate diseases like coronary heart disease (AUC 0.936) and kidney calcification (AUC 0.962) via machine learning; the first harmonic of pulse waves shows significant correlation with biochemical markers of acute liver injury (animal experiment); multi-channel array technology enables synchronous acquisition of pulse signals from six or more points; the principle of computerizing Tibetan pulse diagnosis is feasible.
Partially Verified: The “Harmonic-Organ” correspondence has support from systematic reviews but suffers from high methodological heterogeneity; there is sparse evidence for position-specific diagnosis in the TCM system (Cun-Guan-Chi), whereas the organ allocation in the Tibetan medicine system (Chong-Gan-Cha) has not been directly tested by any discrimination study.
Pending Verification: Whether the Tibetan medicine six-point “location-organ” mapping can translate into a blind discrimination increment; comparison of AUC between position-specific models and global models; the independent value of pressing depth gradients as discriminative features.
Related Reading
- Signal Differences and Organ Attribution in Tibetan Six-Point Pulse Diagnosis
- Blind Discrimination Model for Pulse Harmonic Spectra in Tibetan Tridosha Constitutions
- Quantitative Mapping Models Between Arterial Geometry Parameters and Pulse Harmonic Spectra
- Longitudinal Stability and Plasticity of Individual Pulse Spectrum Signatures
- Research Pool: Biological Frequency Codes
Evidence Level: B (“Pulse Wave + ML Discrimination of Disease” has direct evidence; “Harmonic-Organ Correspondence” has indirect evidence; but “Blind Discrimination of Tibetan Six-Point Location-Organ Mapping” lacks direct empirical evidence overall)
References: PMID 39285363, PMID 40883745, PMID 20021158, PMID 36716225, PMID 33303385, PMID 38481553, PMID 35005017, PMID 30105076
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