The high-quality imaging of vascular networks in biological tissue is significant to accurate cancer diagnosis with acoustic-resolution-based photoacoustic microscopy (AR-PAM). So far, many new back-projection (BP) models have been proposed to improve the image quality of AR-PAM in the off-focal regions. However, many essential arguments are still open regarding the effectiveness of these methods. To settle these remaining questions and explore the potential and adaptability of these BP methods in vascular network imaging, we conducted extensive simulations of a complex vascular network based on a GPU-based data generation framework. Results show that the SAFT-CF algorithm effectively improves the reconstructed image but mainly highlights point targets. In contrast, the STR-BP algorithm can effectively balance the computational cost, signal-to-noise ratio (SNR), and consistency of target intensity for both point and line targets. Results proved that data interpolation for more A-line numbers would not improve the image quality due to information lost. Thus, the detector number in the scan should be sufficiently large. Results also showed that the STR-BP method improved the PSNR of the image by 4.7 to 7.5 dB, which helps the image withstand a noise level of higher than 25%. The proposed simulation framework and the intuitive findings will guide the design of AR-PAM systems and image reconstruction.