SpikeBP: Efficient Spike-Driven Transformer for Blood Pressure Waveform Generation with Frequency Knowledge Distillation

First spike-driven framework for BP waveform generation from ECG/PPG: lightweight spiking Transformer plus Spike Frequency Knowledge Distillation (FFT of hidden states, no extra parameters). On MIMIC-III: claimed SOTA BP estimation and competitive waveform generation at ~12× less energy than the lowest-energy ANN baseline. No numbers in abstract.

• Form factor: clinical_finger

Hypertension is a leading controllable risk factor for cardiovascular disease and premature mortality, and continuous blood pressure (BP) monitoring is critical for capturing dynamic variations that intermittent single-point measurements fail to reveal. While non-invasive methods using physiological signals such as electrocardiography (ECG) and photoplethysmography (PPG) have emerged as promising alternatives, existing artificial neural networks (ANNs)–based approaches impose substantial computational and energy burdens, limiting their deployment in wearable settings. Spiking neural networks (SNNs) offer energy-efficient event-driven processing but suffer from degraded accuracy due to sparsity. To overcome these limitations, we propose SpikeBP, the first spike-driven framework for BP waveform generation. SpikeBP employs a lightweight transformer architecture to achieve efficient long-term waveform prediction. To further enhance accuracy, we introduce Spike Frequency Knowledge Distillation (SFKD), a frequency-domain strategy that transforms hidden representations via FFT to capture the temporal periodicity and global dynamics of BP waveforms without introducing extra parameters. Empirical results on the MIMIC-III dataset show that SpikeBP achieves SOTA performance in BP estimation and competitive results in waveform generation, while consuming nearly 12× less energy than the lowest-energy ANN method.

Verbatim from OpenAlex (opens in a new tab) · Abstract via OpenAlex (publisher-deposited); © the authors/publisher

No results from this paper on ChatPPG leaderboards yet. Browse the leaderboards.