Acceleration of high-quality Raman imagingviaa locality enhanced transformer network

• DOI: 10.1039/d3an01543b
• Publication Date: 2023-11-04T07:02:20Z
• Authors: Shizhuang Weng; Rui Zhu; Yehang Wu; Cong Wang; Pan Li; Ling Zheng; Dong Liang; Zhangling Duan
• Keywords: Diagnostic Imaging; Brain Neoplasms; Acceleration; Image Processing, Computer-Assisted; Humans; Neural Networks, Computer; Cytology

Raman imaging (RI) is an outstanding technique that enables molecular-level medical diagnostics and therapy assessment by providing characteristic fingerprint morphological information about molecules. However, obtaining high-quality images generally requires a long acquisition time, up to hours, which prohibitive for RI applications of timely cytopathology histopathology analyses. To address this issue, image super-resolution (SR) based on deep learning, including convolutional neural networks transformers, has been widely recognized as effective solution reduce the time required achieving RI. In study, locality enhanced transformer network (LETNet) proposed perform SR. Specifically, general architecture adopted with replacement self-attention convolution generate high-fidelity detailed SR images. Additionally, in LETNet further optimized utilizing depth-wise improve computational efficiency model. Experiments hyperspectral breast cancer cells few channels brain tumor tissues demonstrate achieves superior 2×, 4×, 8× fewer parameters compared other methods. Consequently, can be obtained significant reduction ranging from 4 64 times. Overall, method provides novel, efficient, reliable expedite promote its application real-time diagnosis therapy.