An integrated convolutional neural network with zero-dimensional cardiovascular hemodynamics parameters for early cardiovascular disease detection

This study addresses a critical challenge in cardiovascular disease (CVD) management: late detection, which, often at an advanced stage, can lead to high mortality risk. Conventional approaches to severe CVD cases involve invasive treatments, which can distress patients. To mitigate risk of severe outcomes or sudden death from CVD, this research introduces a novel predictor framework, combining upstream blood pressure waveform analysis with artificial intelligence, specifically integrating a Convolutional Neural Network (CNN) and Rideout's zero-dimensional cardiovascular model parameters. Rideout's model identified 16 significant parameters affecting the aortic wave, which were used to train CNN for predicting CVD from aortic waveforms, derived from radial pulse waveforms using two transfer functions. The study pinpointed two critical parameters, Pulmonary Vein 2 and Systemic Aortic Artery 1, as CVD indicators, proposing a biological correlation where these parameters concurrently relax to facilitate smooth blood flow, thereby reducing blood vessels’ resistance values. Experimental validation involved using the best-performing CNN to obtain parameter values from signals in the PhysioNet MIMIC II database, which included 4 CVD and 19 non-CVD signals, serving as base indicators for classifying cardiovascular and non-cardiovascular diseases. These indicators were then used to verify the classification of 3365 healthy signals from the HaeMod dataset and 40 CVD signals collected from Hospital Sultanah Bahiyah (HSB), Malaysia. The system achieved 80.0% and 82.5% accuracy for EIF and GTF transfer functions respectively, based on HSB data, significantly enhancing early detection and offering timely intervention, while proving the potential for practical application of the system in clinical settings.