Objective Recently, spatial frequency domain imaging (SFDI) has been widely used to diagnose and treat skin and other thin tissue diseases owing to its superiority in fast non-contact wide-field imaging. Additionally, the development of single-pixel imaging (SPI) technique has brought significant performance advantages of sensitivity in the invisible band and multispectral detection; therefore, it is applied to the SFDI system. However, since the SPI technique requires hundreds of measurements to capture a single image, the present methods can only achieve low frame rate in dynamic applications. This makes it difficult to meet real-time data acquisition (more than ten frames per second) required for clinical applications. Additionally, SFDI has been challenging in real-time applications in improving the speed of optical property extractions. The look-up table method is widely used in the SFDI system. However, the method is time-consuming for high-resolution image reconstruction, especially in multispectral applications. This is because the method is processed in a pixel-wise manner. In this study, we develop a dynamic SPI method that effectively improves the frame rate of data acquisition. We also propose a fast look-up table algorithm that greatly speeds up the optical property extraction. Methods The dynamic SPI method uses overlapping of the imaging window to speed up the reconstruction of the new frames. In this scheme, only part of the measuring values is updated to form an imaging window and reconstruct a new frame, while other measuring values share those of the previous frame. This study adopts the SPI reconstruction method based on discrete cosine transform (DCT). The DCT coefficients of the reflected images at the planar (DC) and sinusoidal modulation (AC) frequency bands are sampled in a circular path to maximize the sampling efficiency for the reflected images. The measuring priority of the DCT coefficients to be sampled is well-designed to ensure that the DC and AC components are measured synchronously during dynamic imaging. The fast look-up table algorithm works by employing the region and pixel optimizations to reduce the redundant computing processes of the traditional look-up table methods. Region optimization narrows the database region to be searched by utilizing the monotonicity of the diffuse reflectance with respect to the optical property, whereas pixel optimization reduces the number of pixels to be reconstructed by utilizing the change in rate of diffuse reflectance between the adjacent frames. Then, the phantom experiments, including the moving object and dynamic optical property experiments are conducted to access to capability of the dynamic SPI method. Additionally, the speed and error analyses are performed to evaluate the real-time performance of the fast look-up table algorithm. Results and Discussions The results of the moving object experiment show that the dynamic SPI method can achieve a frame rate of more than 10 frames per second to capture the moving process of the object (Fig. 5). The DC and AC components of the reflected images are updated synchronously during dynamic imaging using the optimized measuring priority. The absorption curves of the two wavelengths reconstructed in the dynamic optical property experiment reflect the dynamic changes of the absorption coefficient in the target region; thus, demonstrating the real-time capability of the method in multispectral imaging applications (Fig. 6). The speed and error analyses of the fast look-up table method show that the region optimization improves the speed of the look-up table process five times (from 0.762 s to 0.162 s). Combined with the pixel optimization ( threshold is 0.03), the time required to reconstruct a single image (128 pixel x 128 pixel) can be reduced to 0.031 s, with the mean errors of the reconstructed absorption and reduced scattering coefficient less than 2% and 1 , respectively (Table 1). Conclusions This study proposes a dynamic SPI method for real-time data acquisition and a fast look-up table method for real-time data processing. The proposed dynamic SPI method effectively improves the frame rate without relying on the hardware performance of the imaging system. Combined with the circular sampling strategy and the optimization of measuring priority, the multi-wavelength spatial frequency diffuse images are dynamically measured at a sampling rate less than 1 . Moreover, a fast look-up table algorithm is developed, which performs region and pixel optimizations for the redundancy calculation of traditional look-up table methods. It effectively improves the efficiency of the look-up table method for optical property reconstruction. The proposed method realizes real-time acquisition and reconstruction of the multi-wavelength optical property images for more than 10 frames per second. This provides an effective technical scheme for real-time implementation of single-pixel SFDI.
