Secure and fault tolerant cloud based framework for medical image storage and retrieval in a distributed environment

Area of Science:

• Medical Imaging
• Cloud Computing
• Cybersecurity

Background:

• Centralized cloud-based medical image retrieval systems face significant security and availability challenges.
• Existing deep learning solutions are vulnerable to adversarial attacks and emerging quantum threats.
• There is a critical need for secure, distributed solutions in medical image management.

Purpose of the Study:

• To propose SFMedIR, a secure and fault-tolerant framework for medical image retrieval.
• To enhance the resilience of cloud-based medical image systems against adversarial and quantum threats.
• To improve the accuracy and generalizability of medical image retrieval.

Main Methods:

• Developed an adversarial attack-resistant federated learning approach for hashcode generation using a ConvNeXt model.
• Integrated quantum-chaos-based encryption for robust data security.
• Implemented dynamic threshold-based shadow storage and a distributed cloud architecture for fault tolerance and to mitigate single points of failure.

Main Results:

• Achieved a 60-70% improvement in retrieval accuracy for adversarial queries on Brain MRI and Kidney CT datasets.
• Demonstrated an overall retrieval accuracy of 90%, outperforming existing models by 5-10%.
• Showcased superior performance in both security and retrieval efficiency compared to conventional methods.

Conclusions:

• SFMedIR offers a resilient and efficient solution for secure cloud-based medical image retrieval.
• The framework significantly enhances security and availability in healthcare applications.
• This approach represents a valuable advancement for secure medical image management.