Evidence level: B
中医”望闻问切”四诊合参,与现代多模态生物特征融合,本质上都是同一件事:把多个互有重叠、又各不相同的信号通道合并成一个更可靠的判断。
正文
一、四诊是什么:四个独立的信息通道
中医看病从不靠单一信息。望诊看舌象、面色(视觉通道),闻诊听声音、嗅气味(听觉+嗅觉通道),问诊问症状、病史(语言/文本通道),切诊摸脉搏(触觉/压力通道)。四者采集的物理介质完全不同,各自的噪声源也完全不同:舌象受光照影响,声音受环境干扰,问诊受患者表达偏差影响,脉诊受医生指压主观性影响。
关键之处在于:四路信号的噪声互不重叠,而它们携带的”病人状态信息”有大量重叠又有互补。这正是信息论里多通道融合能提升判读准确性的前提条件。
二、现代证据一:四诊信息融合确实提升证候判别
现代研究已经在做”把四诊数字化再融合”这件事。Kang 等人(2015,PMID 25797918)对 634 例慢性乙肝患者,把中医四诊信息与现代临床检验指标共 222 个参数整合成”双视图分层分类模型”来预测中医证候(肝胆湿热/肝郁脾虚等),结果融合模型的判别性能显著高于只用任一单视图参数——四诊和现代指标各自贡献了对方没有的信息,整合后证候分类更准。
更进一步,数字化的单通道本身已经能干活:Lim 等人(2023,PMID 37957660)仅用桡动脉脉搏波参数 + 机器学习,就在 459 人里以 AUC 0.715-0.722 分类多囊卵巢综合征(PCOS),证明”切诊”这一路信息单独就能携带疾病判别力;Zhao 等人(2024,PMID 38875060)仅凭 1149 张舌象照片 + 集成学习,中医体质识别准确率达 0.81-0.83;Liu 等人(2025,PMID 40506765)用舌象传统特征 + 深度学习特征融合,在 1374 人双中心队列上把体质识别推高到 ACC 0.837、AUC 0.898——融合特征显著优于单特征。每一条通道都能独立判读,融合又比单通道更强,这是四诊合参经验法则在现代数据上的直接复现。
三、现代证据二:多模态信号融合的同一规律
在生物特征识别与情感计算领域,同样的规律被反复验证:单模态系统有上限(噪声、个体差异、伪装),多模态融合系统性优于最优单模态。Hou 等人(2023,PMID 37862275)提出判别相关融合 + 时间对齐机制,把中枢神经(EEG)与自主神经(ECG/皮肤电等)信号融合做情绪识别,融合后性能显著超过任何单模态;且他们特别强调”时间对齐”——各模态信号必须先对齐到同一时间轴,融合才有意义。
这与四诊的逻辑惊人地同构:望闻问切也是四个”测量探头”,必须在同一时刻、同一病人身上采集,才有”合参”的价值。中医历代强调”四诊合参”而非”四诊单参”,与现代信息论”融合增益来自通道互补性”的结论,说的是同一件事。
四、信息论等价性:一个可检验的命题
把四诊与现代多模态融合放在同一框架下,可以形式化成一个可检验命题:
中医”四诊合参”的判读增益 ≈ 现代多模态融合的增益规律——两者的共同结构是:多通道各自独立噪声 + 通道间信息冗余与互补并存 → 最优融合策略(而非简单平均)→ 判读准确性提升。
具体而言,可以测量四诊四个数字化通道(舌象特征、声音特征、症状文本特征、脉搏波特征)之间的互信息矩阵与条件互信息,与现代多模态生理信号融合(EEG/ECG/脉搏)的同一指标做同构性比较;再检验”融合增益是否服从同一信息论不等式”。若成立,”四诊合参”就不再只是经验教条,而是一条可以量化的信息论准则。
证据与局限
已验证: 四诊信息+现代指标融合显著提升证候分类(PMID 25797918);单通道脉诊(AUC 0.72)/舌诊(ACC 0.83)独立可判别,融合特征优于单特征(PMID 38875060、40506765);多模态生理信号融合显著优于单模态且需时间对齐(PMID 37862275)。
部分验证: 四诊各通道的互补性在慢性乙肝、PCOS、体质识别中均有体现,但尚未有人测量过四诊通道间的互信息结构;多模态融合的增益规律(互补性→增益)在情感计算领域稳定复现,但未与四诊做直接对照。
待验证: 四诊四通道的互信息矩阵与现代多模态信号融合的同构性;融合增益是否服从同一信息论不等式;单通道缺失时融合系统的衰减曲线(鲁棒性边界)。
相关阅读
证据等级: B(四诊融合与多模态融合各自的增益各有 A 级直接支撑,”两套系统的信息论等价性”本身无直接研究,属待补空白)
参考文献: PMID 25797918, PMID 37957660, PMID 38875060, PMID 40506765, PMID 37862275
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。
Four Diagnostics Combined Equals Multimodal Information Fusion: Information-Theoretic Equivalence Between Traditional Observation, Hearing, Inquiry, and Touch with Modern Biomet…
Evidence level: B
Traditional Chinese Medicine’s “Four Diagnostics Combined” (Wang-Wen-Wen-Qie) and modern multimodal biometric fusion are fundamentally the same process: merging multiple overlapping yet distinct signal channels into a more reliable judgment.
Main Text
I. What Are the Four Diagnostics: Four Independent Information Channels
TCM diagnosis never relies on a single information source. Wang (inspection) examines the tongue and complexion (visual channel); Wen (auscultation/olfaction) listens to sounds and smells odors (auditory + olfactory channels); Wen (inquiry) asks about symptoms and history (language/text channel); Qie (palpation) feels the pulse (tactile/pressure channel). The physical media collected by each differ significantly, and their respective noise sources are also distinct: tongue images are affected by lighting, sounds by environmental interference, inquiries by patient expression bias, and pulse diagnosis by the physician’s subjective palpation.
The key lies in that: the noise across the four signals does not overlap, while they carry overlapping yet complementary “patient state information.” This is the prerequisite condition for multi-channel fusion to improve diagnostic accuracy in information theory.
II. Modern Evidence I: Four-Diagnostic Information Fusion Indeed Enhances Syndrome Discrimination
Modern research is already conducting the task of “digitizing and fusing the four diagnostics.” Kang et al. (2015, PMID 25797918) integrated TCM four-diagnostic information with modern clinical indicators (222 parameters total) into a “dual-view stratified classification model” to predict TCM syndromes (e.g., liver-gallbladder damp-heat, liver stagnation-spleen deficiency). The results showed that the discriminative performance of the fused model was significantly higher than using any single-view parameter alone—each diagnostic channel contributed information absent in the other, leading to more accurate syndrome classification upon integration.
Furthermore, single-channel digitalization itself is functional: Lim et al. (2023, PMID 37957660) used only radial artery pulse wave parameters + machine learning to classify Polycystic Ovary Syndrome (PCOS) in 459 individuals with an AUC of 0.715–0.722, proving that the Qie (palpation) channel alone carries disease discriminative power; Zhao et al. (2024, PMID 38875060) achieved TCM constitution identification accuracy rates of 0.81–0.83 using only 1,149 tongue image photos + ensemble learning; Liu et al. (2025, PMID 40506765) pushed constitution identification to ACC 0.837 and AUC 0.898 in a dual-center cohort of 1,374 individuals by fusing traditional tongue features with deep learning features—fused features were significantly superior to single features. Each channel can independently diagnose, yet fusion is stronger than single-channel performance, which is a direct replication of the “Four Diagnostics Combined” empirical rule on modern data.
III. The Same Law in Multimodal Signal Fusion
In the fields of biometric recognition and affective computing, the same law is repeatedly verified: unimodal systems have limits (noise, individual differences, spoofing), while multimodal fusion systematically outperforms the optimal unimodal approach. Hou et al. (2023, PMID 37862275) proposed discriminative correlation fusion + time alignment mechanisms, fusing central nervous system signals (EEG) with autonomic nervous system signals (ECG, skin conductance, etc.) for emotion recognition; performance significantly exceeded any unimodal approach after fusion. They particularly emphasized “time alignment”—multi-modal signals must first be aligned to the same time axis for fusion to be meaningful.
This logic is strikingly isomorphic to Wang-Wen-Wen-Qie: the four diagnostics are also four “measurement probes” that must be collected from the same patient at the same moment to realize the value of “combined reference.” Traditional TCM has always emphasized “Four Diagnostics Combined” rather than “Four Diagnostics Single-Reference,” mirroring the modern information theory conclusion that “fusion gain arises from channel complementarity.” They describe the same underlying phenomenon.
IV. Information-Theoretic Equivalence: A Testable Proposition
Placing the Four Diagnostics and modern multimodal fusion under the same framework allows formalizing a testable proposition:
The diagnostic gain of TCM “Four Diagnostics Combined” ≈ the gain law of modern multimodal fusion—their common structure is: independent noise across multiple channels + coexistence of inter-channel information redundancy and complementarity → optimal fusion strategy (not simple averaging) → improved diagnostic accuracy.
Specifically, one can measure the mutual information matrix and conditional mutual information among the four digitized diagnostic channels (tongue features, sound features, symptom text features, pulse wave features) and compare them with the same metrics used in modern multimodal physiological signal fusion (EEG/ECG/pulse). One should then verify whether “fusion gain follows the same information-theoretic inequality.” If established, “Four Diagnostics Combined” ceases to be merely an empirical dogma and becomes a quantifiable information-theoretic criterion.
Evidence and Limitations
Verified: Fusion of four-diagnostic information with modern indicators significantly enhances syndrome classification (PMID 25797918); single-channel pulse diagnosis (AUC 0.72) / tongue diagnosis (ACC 0.83) independently discriminates, while fused features outperform single features (PMID 38875060, 40506765); multimodal physiological signal fusion significantly outperforms unimodal approaches and requires time alignment (PMID 37862275).
Partially Verified: Complementarity across the four diagnostic channels is evident in chronic hepatitis B, PCOS, and constitution identification, but no one has yet measured the mutual information structure between four-diagnostic channels; the gain law of multimodal fusion (complementarity → gain) is stably replicated in affective computing but has not been directly contrasted with the Four Diagnostics.
To Be Verified: Isomorphism between the mutual information matrix of the four diagnostic channels and modern multimodal signal fusion; whether fusion gain adheres to the same information-theoretic inequality; the attenuation curve of the fusion system when a single channel is missing (robustness boundaries).
Related Reading
- Multimodal Frequency Signature Biometric System
- Individual Frequency Signature as a Biometric Feature
- Temporal Rhythmic Stability of Pulse Spectrum Individual Signature
- Can Quantitative Mapping Models Replicate Harmonic-Viscera Correspondence
- Research Pool: Bio-Frequency Codes
Evidence Level: B (Each system’s gain—Four Diagnostics fusion and Multimodal fusion respectively—has direct A-level support; the “information-theoretic equivalence between two systems” itself lacks direct research, representing a gap to be filled)
References: PMID 25797918, PMID 37957660, PMID 38875060, PMID 40506765, PMID 37862275
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