Enhanced ResNet-based super-resolution method for two-photon microscopy image

Two-photon microscopy (TPM) image is composed of two-photon excitation fluorescence (TPEF) and second harmonic generation (SHG) signals, which play a vital role in detecting lesions of biological tissues. However, low contrast and signal-to-noise ratio (SNR) appear in TPM image due to the complex imaging process. Image super-resolution (SR) aims to reconstruct high-resolution (HR) images from low-resolution (LR) ones. According to the working principle of TPM imaging system, the degradation model of TPM image is first analyzed. Then, exploiting the Retinex theory, an enhanced ResNet-based super-resolution (ERNSR) method for TPM image is proposed, which consists of residual blocks, discrete cosine transform (DCT) and sub-pixel convolution. ERNSR is trained on a TPM dataset, which contains 113 TPM images of liver samples collected through Zeiss LSM510. Comparisons with several state-of-the-art methods, the experimental results demonstrate that our proposed approach achieves a notable improvement in terms of both quantitative and qualitative measurements. It holds the potential to improve the precision of some computer-aided diagnosis systems after SR reconstruction by our method.