Deep learning-based microcirculation performance classification using multispectral photoacoustic technology

Investigation of microcirculation is the key to diagnose circulatory dysfunction. Tissue circulation monitoring is a crucial part of the care of patients with severe chronic illnesses because it affects oxygen delivery to tissue. Recent technology, such as hyperspectral imaging, has allowed visualization of microcirculation at the price of high computation resources. Meanwhile, pulse oximeter performance varies with factors like the subject’s skin colour. This study explores the feasibility of using an in-house assembled multispectral photoacoustic (PA) system to investigate microcirculation performance in human subjects. We used pretrained Alexnet, Long Short-Term Memory (LSTM), and a hybrid Alexnet-LSTM network for the prediction task. This research included thirty-seven healthy participants in this cross-sectional study. The ultrasonic waves collected from their posterior left arm under two experimental settings, namely at rest (i.e., control) and with arterial blood flow occlusions, were used to predict the microcirculation changes in tissue using the deep networks. Our findings showed the superiority of the hybrid model over the Alexnet and LSTM, with an average testing accuracy of 95.7 % and precision of 98.2 %, making it an ideal deep learning model for the task. This study concluded that the proposed deep learning incorporated photoacoustic system has a promising future for diagnosing and treating patients with compromised microcirculatory conditions.