Secure and Robust Machine Learning for Healthcare: A Survey.

Area of Science:

• Healthcare AI
• Machine Learning in Medicine
• Deep Learning for Medical Imaging

Background:

• Machine learning (ML) and deep learning (DL) are increasingly used in healthcare for tasks like cardiac arrest prediction and computer-aided diagnosis (CADx).
• Despite high performance, the robustness of ML/DL in healthcare is questioned due to security and privacy concerns.
• ML/DL models are vulnerable to adversarial attacks, posing risks in sensitive medical applications.

Purpose of the Study:

• To provide an overview of healthcare applications utilizing ML/DL from a security and privacy perspective.
• To identify and discuss the challenges associated with secure and privacy-preserving ML in healthcare.
• To explore potential methods and future research directions for enhancing ML/DL security and privacy in medical contexts.

Main Methods:

• Literature review of ML/DL applications in healthcare.
• Analysis of security and privacy challenges, including adversarial attacks.
• Exploration of proposed methods for secure and privacy-preserving ML.

Main Results:

• ML/DL techniques are widely adopted across diverse healthcare applications, demonstrating significant potential.
• Key challenges include ensuring data security, patient privacy, and model robustness against sophisticated attacks.
• Various strategies are being developed to mitigate risks and ensure trustworthy AI in healthcare.

Conclusions:

• Secure and privacy-preserving ML is crucial for the responsible adoption of AI in healthcare.
• Addressing adversarial vulnerabilities and privacy concerns is essential for building trust in medical AI.
• Continued research is needed to develop robust, secure, and ethical ML/DL solutions for healthcare.