: A transformer architecture conductive to blood pressure estimation through solo PPG signal capturing

Recent statistics indicate that approximately billion individuals worldwide are affected by hypertension, a major contributor to premature mortality. Blood pressure (BP) is an important physiological indicator for hypertension assessment and management. Conventional cuff-based methods provide intermittent measurements and may be unsuitable for continuous monitoring. Recent advances in artificial intelligence (AI) and deep neural networks (DNNs) have therefore increased interest in data-driven cuff-less BP estimation. Existing studies frequently combine electrocardiography (ECG) and photoplethysmography (PPG), requiring multiple synchronized sensor types. PPG-only estimation offers a simpler acquisition setup but remains challenging because BP-related morphological information must be extracted in the presence of motion artifacts and high-frequency noise. This paper introduces TransfoRhythm, a Transformer-based DNN framework for PPG-only BP estimation using the MIMIC-IV Waveform Database. The framework uses multi-head attention to model temporal relationships among cycle-level PPG features. Evaluation was performed using patient-independent 5-fold cross-validation, with patient IDs used as the splitting unit and zero patient overlap between training and test folds. Within each fold, feature selection, collinearity filtering, data scaling, hyperparameter selection, and model fitting were performed using only the training patients. Under this protocol, TransfoRhythm achieved mean MAEs of mmHg for systolic BP and mmHg for diastolic BP, with corresponding RMSE values of and mmHg. The mean errors were mmHg and mmHg, while the mean values were and for systolic and diastolic BP, respectively. These findings demonstrate consistent predictive performance across held-out patient groups within the evaluated MIMIC-IV cohort. External and prospective validation remains necessary to establish generalizability beyond this dataset.