A Joint Multimodal User Authentication-based Privacy Preservation with Disease Prediction Framework in Modern Healthcare System Using Multi-Scale Cross Attention-based ResNet.

Background/introduction: The disease prediction process plays a crucial part in a person's life "to lead a healthy life." The sudden spread of the data mining approach has generated the disease forecasting system. Secure transfer of medical data and effective storage is the major difficulty faced by recent healthcare management. Moreover, there is significant attention towards privacy preservation, especially for medical information, which is highly sensitive. For disease prediction, several prevailing privacy preservation approaches have been developed. "Moreover, although the disease prediction system is auspicious, its complexity may limit practical use, including information security and prediction efficiency."

Methods: Multimodal user authentication is performed by a Multi-scale Cross Attention-based Residual Network (MCARNet) to prevent unauthorized access to the healthcare system. Images and signals are converted into 2D images for performing the encryption using the Optimal Rossler Hyper Chaotic Encryption (ORHCE). The decrypted images are given to the same MCARNet for predicting the disease.

Results: The precision of the developed model was enhanced by 7.3% of DNN, 12.3% of RNN, 3.6% of LSTM, and 4.3% of GRU when taking the k fold value as 5.

Conclusion: The multimodal user authentication and disease detection using the proposed heuristic-based hybrid deep learning model enhanced its authentication and detection performance.