Intracerebral Hemorrhage Imaging Based on Hybrid Deep Learning With Electrical Impedance Tomography

Intracerebral hemorrhage (ICH) is a common disease that has characteristics of high disability rate and high mortality rate. Rapid detection and continuous monitoring are crucial for the diagnosis and treatment of ICH. As a potential technique, electrical impedance tomography (EIT) offers an alternative to brain imaging. With this technique, an image representing conductivity distribution variation is reconstructed and physiological change in the intracerebral region can be provided. However, image reconstruction is an inherently nonlinear and severely ill-posed inverse problem. To solve the problem, a convolutional neural network (CNN) and transformer combined hybrid network (CTHN) are proposed to map the nonlinear relationship between boundary measurement and conductivity distribution. With the hybrid network, high-level features of voltage measurement sequence are extracted, and long-range dependencies learning ability can be improved with a few layers. To evaluate the performance of the proposed method, extensive simulations are carried out on a three-layer head model with the inclusion which simulates ICH. Image reconstructions are separately performed under noiseless conditions, different noise level condition, conductivity variation in three layers, and deformed head models. The performance of the proposed method is also tested by the experiments. Both reconstructed images and quantitative evaluations show that ICH can be accurately reconstructed with the proposed approach.