Electrical Impedance Tomography (EIT) image reconstruction is the process of generating visual images that represent the electrical conductivity or impedance distribution within a body or object, based on external electrical measurements. Existing EIT image reconstruction methods suffer from low image resolution and unclear boundary problems due to its nonlinear, under-determined and ill-posed nature. To address the limitations of existing works, we propose a novel MSARM-based EIT image reconstruction method, which aims to improve the quality of reconstructed image. Our method consists of three modules: pre-mapping, encoder and decoder modules. Furthermore, we introduce multi-scale feature extraction, Sim-AM attention mechanism and residual blocks, which can extract more features and fuse contextual information discriminatively. We evaluate our proposed method on a simulated dataset generated by EIDORS. The experimental results demonstrate that our proposed MSARM method have achieved state-of-the-art performance on image reconstruction tasks. To be specific, compared with existing multi-scale U-Net module, our method has the \(1.43\%\) improvement in CC and \(13.43\%\) reduction in RE, showing the effectiveness of our proposed method.