Abstract
To tackle real-time denoising of noisy laser speckle blood flow images, a novel lightweight denoising speckle contrast image generative adversarial network (LDSCI-GAN) is proposed. In the framework, a lightweight denoiser removes noise from the original image, and a discriminator compares the denoised result with the reference one, enabling efficient learning and optimization of the denoising process. With a multi-scale loss function in the log-transformed domain, the training process significantly improves accuracy and denoising by using only five frames of raw speckle images while well-preserving the overall pixel distribution and vascular contours. Animal and phantom experimental results indicate that the LDSCI-GAN can eliminate vascular artifacts while retaining the accuracy of relative blood flow velocity. In terms of peak signal-to-noise ratio (PSNR), mean structural similarity index (MSSIM), and Pearson correlation coefficient (R), the LDSCI-GAN outperforms other deep-learning methods by 3.07 dB, 0.10 (p < 0.001), and 0.09 (p = 0.023), respectively. It has been successfully applied to the real-time monitoring of laser-induced thrombosis. Through conducting tests on the denoising performance of blood flow images of a moving subject, our proposed method achieved enhancements of 23.6% in PSNR, 30% in MSSIM, and 6.5% in the metric R, respectively, when compared to DRSNet. This means that the LDSCI-GAN also shows possible application in handheld devices, offering a potent tool for investigating blood flow and thrombosis dynamics more efficiently and conveniently.