Reliable image reconstruction from DNA storage through marker-guided decoding and candidate filtering

Abstract DNA has emerged as a promising medium for long-term digital data storage because of its exceptional information density and durability. However, reliable data recovery remains challenging for Oxford Nanopore sequencing, where insertion, deletion, and substitution errors can disrupt sequence synchronization and increase ambiguity during decoding. In this study, we present a proof-of-concept framework for DNA-based RGB image storage and reconstruction that integrates marker-aware codebook construction, consensus-based marker-guided decoding from a single sequence, and AI-assisted image restoration. The framework systematically evaluates how different codebook construction strategies influence downstream decoding, candidate ambiguity, and final image reconstruction. Under simulated nanopore sequencing conditions, all constrained codebooks achieved high payload recovery (> 91%), but substantial differences were observed in their ability to preserve the correct payload among decoded candidate sequences. The codebook combining sequence-space discrimination, biochemical quality control, and structural compatibility produced the highest candidate-set quality and, when coupled with Random Forest-based reliability filtering, achieved the best image reconstruction performance, with a peak signal-to-noise ratio of 29.82 dB, a structural similarity index of 0.9309, and 80.19% exact pixel recovery. These findings demonstrate that payload recovery alone is insufficient to evaluate DNA storage performance, as downstream candidate ambiguity directly impacts the quality of reconstructed data. The proposed framework provides an end-to-end approach to investigating the relationship among codebook design, decoding robustness, and image reconstruction, thereby supporting the development of more reliable nanopore-based DNA data storage systems.