AI-DRIVEN PACS SECURITY FRAMEWORK: THREAT MODEL, ATTACK VECTORS, AND DEFENSE STRATEGY

THREAT MODEL & SUPPLY-CHAIN COMPROMISE

WSUS Server (Compromised)

Malicious Updates (CVE-2025-58287)

COORDINATED INTEGRITY ATTACKS

IMAGE INTEGRITY ATTACK (DISPLAY-TIME)

Rendering Engine (Display-Time)

Image Manipulation (During Idle Screen)

In-Memory Display (Altered Representation)

Workflow Timing Inference

REPORT INTEGRITY ATTACK (CREATION-TIME)

Report Assembly & Editing

Report Manipulation (Subtle Alterations)

Image Features & Workflow Context

Report Content (Structured & Free-Text, Clinically Incorrect)

Digital Signing (Pre-Signing)

RIS/EHR Systems (Propagated Incorrect Output)

AI-DRIVEN DETECTION, DEFENSE & MITIGATION FRAMEWORK

Self-Supervised Vision Transformers (e.g., ResNet, EfficientNet) Temporal Consistency Analysis Diffusion-Based Reconstruction Models

Large Language Models (LLMs) (e.g., ClinicalBERT, BioBERT) Cross-Modal Vision-Language Networks

Semantic Validation of Report Content

MITIGATION & REVIEW Halt Image Display & Report Signing Regenerate Trusted Content Flag for Secondary Clinical Review

Verified Clinical Decision Points

Side-Channel Observations (Timing, Resource Usage)

Image Integrity Verification

Semantic Validation of Report Content

picture1