Photoacoustic Image Reconstruction Using a Two-Phase Approach with Lightweight LNEformer

Medical imaging plays a crucial role in early disease diagnosis, with photoacoustic imaging (PAI) gaining attention due to its high-resolution imaging capabilities. However, reconstructed images often suffer from artifacts, affecting clinical reliability. This paper proposes a novel two-phase approach for enhancing PAI reconstruction, focusing on the efficiency and effectiveness of the Local Neighborhood-Enhanced Transformer (LNEformer) model. The first phase involves preliminary image reconstruction using the Exponential Filtering Method, which effectively suppresses artifacts while preserving critical structural details. The second phase refines the reconstruction quality using LNEformer, a lightweight Transformer-based model that incorporates a neighborhood attention (NA) mechanism to capture both local and long-range dependencies while maintaining computational efficiency. Unlike standard self-attention, NA selectively attends to spatially relevant features, enhancing fine structural preservation in reconstructed images. The proposed framework is evaluated on a breast tissue dataset. Experimental results demonstrate that LNEformer outperforms widely used deep learning models such as U-Net and Restoration Transformer (Restormer) in terms of both structural similarity index measure (SSIM) and peak signal-to-noise ratio (PSNR), achieving an SSIM of 0.91 and PSNR of 25.3 dB, while requiring significantly fewer parameters, with LNEformer utilizing only 0.2M parameters compared to 19M for Restormer and 3.6M for U-Net. The proposed framework provides an efficient and effective solution for improving PAI reconstruction, making it suitable for real-time medical imaging applications.