The increased dependence on digital technologies across the healthcare domain accentuates the demand for strong security protocols in effect for the securing of sensitive medical image information. Traditional steganography methods often involve trade-offs between image quality and security, which may influence the resilience of concealed information against detection and potential adversarial analysis. To overcome the issue, this paper introduces a new adversarial attack-resistant steganography framework that securely conceals medical information in non-medical images without compromising visual coherence or diagnostic validity. Based on Generative Adversarial Networks (GANs) and Self-Generated Supervision, the framework offers both detection and tampering resistance as well as reduced annotation data dependence. Moreover, semantic-aware usage of design leads to highly improved consistency and lowers visual distortions while concealing the data. Experiments and evaluation show that the proposed technique produces high-quality steganographic images that are resistant to state-of-the-art steganalysis methods while possessing high quality and security. This work promises an efficient, secure solution towards securing medical information and overcoming necessary challenges inherent to healthcare data security.