A Deep Learning-Based TCM Deficiency-Excess Syndrome Differentiation Framework for Spectrum Analysis of PPG Pulse Wave

Traditional Chinese Medicine (TCM) deficiency-excess syndrome differentiation remains highly subjective. This study proposes a deep-learning framework alongside the Pulse Waveform Spectral Parameter Estimation Model (PWSPEM), that analyses PPG pulse-wave spectra to establish an objective mapping between PPG signals and TCM deficiency-excess patterns. Compared with conventional Fourier decomposition (MSE = 0.12), PWSPEM integrates multi-cycle harmonic superposition, time-varying attenuation and linear skip connections, achieving stable fitting (MSE = 0.002–0.005). Twelve PPG spectral features were quantified by the QSS algorithm. In 6,032 healthy participants, C3 and C4 showed significant positive correlation with deficiency-type syndromes (C3: P = 0.003; C4: P = 0.021) and negative correlation with age (C3: r = –0.160; C4: r = –0.330, both P < 0.001). In classification, PWSPEM outperformed all comparators: accuracy = 0.8182, AUC = 0.9036 versus WaveNet (0.8010/0.8866), CNN-BiLSTM (0.7961/0.8781), CBAM (0.8059/0.8893) and Random Forest (0.7985/0.8847). Preliminary pre-/post-intervention tests further demonstrated clinical utility. PWSPEM thus provides a technical means to support objective TCM deficiency-excess syndrome differentiation.