Deep learning–based ROI Localization and Integrity Assurance Using a Keyed Hash Function with a 4D Hyperchaotic System and DWT/DCT Transforms

Ensuring the integrity of medical images has recently become a significant concern, particularly with the increasing use of telemedicine.These images are vulnerable to both intentional and unintentional alterations during transmission and storage, which may degrade image quality and the accuracy of subsequent clinical diagnoses.Although traditional cryptographic approaches offer basic authentication, they fall short when applied to medical images due to their inability to accommodate the unique spatial and frequency-domain characteristics of these images, especially within diagnostically vital regions of interest (ROIs).To address this limitation, the proposed method focuses on protecting diagnostically important ROIs, unlike traditional hash functions that treat all image areas equally.This is achieved through a new approach for hashing that employs DWT, DCT, and a four-dimensional hyperchaotic map."ROI localization was performed with a standard 3D U-Net trained on BraTS 2020, achieving a validation Intersection-over-Union of ≈0.85 and accuracy >0.99, thereby producing anatomically coherent tumor masks that stabilize downstream DWT-DCT feature extraction and the initialization of the 4D-hyperchaotic core.Some attributes of images in the form of their spatial and frequency-domain features are used to initialize the chaotic system alongside a 512-bit key, which is broken into four separable parts.A series of chaotic maps followed by nonlinear iterations to increase diffusion and robustness yields a final hash through multiple steps.Across 512 bit outputs, the hash exhibits near ideal diffusion: in N = 2048 one bit flip trials it averages 253.1 changed bits (49.44%, range 219-289), confirming a stable avalanche.Under routine tampering for example, JPEG Q = 15/20, ±5° rotation, noise σ = 8, blur, 10% crop, 90% resize, and ±15% brightness/contrast Hamming distances remain ≈254-272 bits (BER≈0.50),so manipulations are reliably detected.Hex digit frequencies are uniform, evidencing high randomness.With a 512 bit key space (2 512 ) and ≈0.0275 Gbps median throughput (~16 ms/ROI), the scheme is robust and practical.