Identification of Cerebral Hemorrhage Location in Brain Electrical Impedance Tomography

Electrical impedance tomography (EIT) has shown its vast potential for applications in the biomedical engineering. Low spatial resolution is acknowledged as the main problem of EIT. Due to poor imaging quality, accurate location of lesion based on reconstructed images is difficult. It becomes more difficult in the brain imaging with EIT attributing to the fact that head structure is complicated and multilayered. Moreover, highly resistive skull seriously impedes the injection of the current. Such factors generate poorer imaging quality. In this study, a new approach which combines convolutional neural network, slime mould algorithm and random forest is proposed for locating cerebral hemorrhage. Different from other studies that locate lesion with reconstructed images, this study is conducted from a different perspective. Cerebral hemorrhage location is implemented by directly processing and differentiating the measured voltage data with the proposed approach in this work. The effectiveness of the proposed approach is tested under noiseless condition. Also, the impacts of noise interruption, conductivity variation, contact impedance change and head shape variation on the proposed method are studied. Comparing with CNN method and CNN-RF method, the proposed approach exhibits greater competence in identifying the location of cerebral hemorrhage and the healthy. In addition, stronger robustness to noise interruption and less sensitivity to contact impedance change are observed. This study offers a substitute option to traditional image-based cerebral hemorrhage location methods.