Deep Domain Adaptation for Predicting Intra‐Abdominal Pressure with Multichannel Attention Fusion Radar Chip

Abstract

Intra‐abdominal hypertension (IAH) has gained increasing attention worldwide because of its prevalence and high mortality rate among intensive care unit (ICU) patients. Most current approaches of measuring intra‐abdominal pressure (IAP) involve the use of sensors inside or attached to the body, which may, however, make daily monitoring inconvenient. This paper proposes a noninvasive and contactless system to learn the relationship between the passive mechanical behavior of the abdominal wall with millimeter‐wave (mm‐wave) frequency‐modulated continuous wave (FMCW) radar and an IAP sensor via a deep learning approach. We correlated the IAP variance with the mobility measures of the abdominal wall and proposed Pearson‐coefficient‐guided domain adversarial neural network (PCG‐DANN) to learn the mapping relationship. To validate the efficacy of our proposed method, a stable intra‐abdominal hypertension/abdominal compartment syndrome (IAH/ACS) model using swine was established to evaluate the mobility of the abdominal wall under different intra‐abdominal pressures with multichannel mm‐wave FMCW radar. The superiority of the proposed method is demonstrated by comparing with other neural network structures and mostly adopted sensor‐based methods. These preliminary results confirm that the new methodology for evaluating IAP nonlinearity is promising and that it can serve as an important diagnostic and treatment reference for patients with IAH.

Keywords

• abdominal wall
• data processing
• intra‐abdominal pressure
• machine learning
• material informatics
• passive mechanical behavior