Information-Theoretic Medical Image Encryption via LLE-Verified Chaotic Keystreams and DNA Diffusion

We propose an information-theoretic encryption scheme consisting of a four-dimensional chaotic map driver in combination with a prediction model using an LSTM neural net to generate a keystream, which was limited only after passing a test based on the largest Lyapunov exponent (LLE). Our security analysis used a permutation phase to remove spatial redundancy, which was followed by an invertible DNA cross-diffusion procedure based on RGB channels. The removal of uncertainty and redundancy was measured using Shannon's entropy (7.99-8.00 bits per channel), pixel intercorrelation, and differential analysis (NPCR ≈ 99.6%, UACI ≈ 33.3%). In key space analysis (order ≈ 2³⁸⁴), self-right veneering with complete encryption validity was demonstrated in perfect decryptability. We explain how chaos verification enhances the statistical goodness of keystreams and provide ablations that separate each element's influence on entropy and decorrelation.