About Radex
Radex is an educational reference tool designed for radiology residents and medical professionals to quickly access imaging appropriateness criteria and MRI protocol information.
Built By
Patrick Matulich, DO
Radiology Resident | CoreGRAI
Radex was developed as a capstone project in radiology AI, motivated by a recurring challenge in clinical training: imaging appropriateness guidelines exist, but accessing them quickly at the point of care remains cumbersome. Built on the ACR Appropriateness Criteria and validated through systematic testing against 202 clinical queries, Radex is designed to reduce time-to-answer from minutes to seconds without sacrificing evidence fidelity.
Key Statistics
Data Sources & Methodology
Appropriateness Criteria: Based on the ACR Appropriateness Criteria, evidence-based guidelines developed by the American College of Radiology to assist referring physicians and other providers in making the most appropriate imaging decisions. The database covers 3,226 clinical scenarios across 10 body regions.
MRI Protocols: Sample protocols based on common clinical practice. Your institution may have different sequences, parameters, or protocols based on scanner hardware, software versions, radiologist preferences, and institutional policies.
Consensus Algorithm: Each of the 74 quick-answer topic cards aggregates all related ACR scenarios for that topic, scores imaging procedures by how often they appear as the top-rated option, and classifies the result: Strong Consensus (≥70% agreement), Conditional (40-69%), High Variance (<40%), or Clinical Assessment First. This transforms 3,226 individual scenarios into quick-answer summaries without losing the underlying evidence base.
Manual Validation: All 74 topic cards were individually reviewed by a clinical domain expert prior to publication. This audit step corrects cases where frequency counting produces a mathematically correct but clinically incorrect primary recommendation (for example, when surveillance scenarios dilute consensus for initial workup).
AI Methodology: The search system uses a hybrid architecture combining a 708 KB intent classifier (a custom 2-layer transformer quantized to INT8 for browser deployment), explicit rule-based patterns, and a 375-term radiology semantic dictionary. The hybrid approach achieves 82.2% phase accuracy and 86.6% urgency accuracy across 202 test queries, compared to 52% for rules alone.
Browse by Region
Important Notice
This tool is provided for educational purposes only. It is not a substitute for clinical judgment, institutional protocols, or direct consultation with radiologists. Imaging decisions should always be made in the context of individual patient circumstances.
Also by CoreGRAI
GRAi is an AI-powered radiology study and clinical reference platform built for residents preparing for the ABR Core Exam. It provides AI-assisted Q&A across radiology topics, radiology corpus search powered by a proprietary knowledge base and custom retrieval model, differential diagnosis support, and structured lesson content - all through a conversational interface. Also built by Patrick Matulich, DO.
Contact
Questions or suggestions? support@coregrai.com
Try Radex - free, browser-based Open Radex →