Evidence · Endpoints · Reasoning

New scoring systems, new guidelines, new AI. Radiology is only as good as the way we evaluate them.

Independent analysis of the evidence behind what radiology adopts: scoring systems, guidelines, protocols, and AI. What the endpoints actually measure, how the tools affect the radiologist, and whether impressive results hold up in clinical practice.

Written by Melina Pectasides, MD
Academic abdominal radiologist, educator, and builder
RadBrief is a journal club for practicing radiologists.

The paper is evidence. The question is the subject.

New evidence in radiology often arrives with an impressive headline number. The harder questions come afterward.

  • Was the comparison fair?
  • Did the endpoint reflect clinical value?
  • Does the scoring system perform in practice the way it performed in the validation cohort?
  • Does the AI strengthen the radiologist's reasoning, or replace a step that still needs to be learned?

RadBrief examines those questions closely and in public.

About

Written from inside the clinical workflow.

Melina Pectasides, MD

Melina Pectasides, MD, is an academic abdominal radiologist at Cleveland Clinic, focused on gynecologic and prostate MRI. Most of her career has been on the standards side of imaging: body MRI directorships, quality programs, the NCI Ovarian Task Force, and the ACR Appropriateness Criteria panel for gynecologic imaging.

She writes RadBrief to examine what the evidence behind radiology's standards and its newest tools actually establishes, and what it leaves unanswered.

She is also the founder of RadReason, a reasoning-support tool designed for use while a radiologist is actively interpreting a case. Building and studying the tool provides an additional perspective on endpoints, human and AI interaction, and the difference between producing an answer and improving a clinician's reasoning.

The tool behind it
Rradreason

Reasoning support during the read, not a diagnosis after it.

RadReason surfaces structured differentials while findings are being dictated, and asks the questions that separate them. It works during the read, and it never makes the diagnosis.

See RadReason →
Live
Differentials surface while you dictate, not after you sign.
Private
Audio and transcripts are processed in memory and discarded. No patient identifiers.
Teaching
Learn Mode asks the discriminator questions a good attending would.
Part of Modvium

Where this work lives.

RadBrief is one part of Modvium. Modvium is where I keep the tools and the teaching, all aimed at making reasoning clearer. RadReason is there. So are plain guides on building with AI.

Visit Modvium ↗
Subscribe

A journal club for practicing radiologists. Once a month: what holds up, what doesn't, and what I'd actually do. Shorter notes go up on LinkedIn in between, and everything stays in the archive here, with references.

Or read on Substack ↗