Fig. 3: Deep learning-based networks allow parameter-free encoding and decoding of the OPMR.

a, Examples of medical information that can be encoded on an OPMR MNP. b, Information data are converted to an encoded binary string before ECC. c, Binary information data are encoded following a 2D template. d, The 2D array becomes an encoded pattern. e, An encryption mask is applied for patient privacy. f, The encrypted pattern is generated for MNP encoding. g, Encoded MNP is fabricated. h, Decoding phase begins with raw image acquisition. i, Raw image is initially rectified via a deep learning-based rectification network. j, Rectified image is in a black-and-white square format. k, Bits are recognized by a deep learning-based recognition network. l, Recognition network outputs a binary array. m, Encryption step is reversed by removing the encryption mask. n, Error bits are identified. o, Error bits are corrected. p, Encoded binary string is translated back to the original information and output on a screen. q, Signal retention analysis quantifies the number of detected NIR bits for 96-bit MNPs. r, Pattern decodability analysis decodes patterned MNPs and evaluates whether they were decoded successfully or not.