A Hybrid Deep Learning Method for Lesion Identification With Electrical Impedance Tomography

Identification of lesion in the lung through sensors is of great importance. In this letter, a new method based on a hybrid TCN-BiLSTM model is proposed to identify the lesion with electrical impedance tomography (EIT). Unlike traditional methods that rely on reconstructed images, the process of image reconstruction is avoided in the proposed method. To differentiate subtle voltage variations in the boundary measurement by sensors between different types, the measured voltage data is processed by multi-scale feature extraction and bidirectional temporal modeling. The performance of the proposed method is compared to that of TCN-LSTM and TCN models. The results show that the identification accuracy reaches 99% under noise-free conditions and is higher than 97% at SNR of 40dB outperforming the comparison models. This approach provides an alternative for lesion detection in the lung with EIT.