Question 1

What is RadReason?

Accepted Answer

A real-time reasoning partner for the reading room. While you describe a finding the way you'd describe it to a colleague, RadReason returns a calibrated differential, the can't-miss diagnosis, and a recommended next step — fast enough to consume between dictated sentences. Built for the dictation moment, not for tab-switching.

Question 2

How is this different from ChatGPT or other AI tools?

Accepted Answer

Three things. It's mic-first and tuned to the cadence of dictation, not chat. The prompts and reasoning are radiology-specific — the model is steered toward calibrated differentials, can't-miss entities, and discriminators rather than generic conversational drift. And your case content stays scoped to the active session, treated as clinical data rather than training fodder.

Question 3

Is this meant to replace radiologists?

Accepted Answer

No. RadReason is decision support — a fast colleague that thinks alongside you. The final read, the recommendation, and the report are yours. The goal is to make your reasoning faster and your blind spots smaller, not to push you out of the loop.

Question 4

Can I use this during live dictation?

Accepted Answer

Yes — that's exactly what it's built for. The interface is mic-first and designed to keep up with real dictation cadence: tap, speak the finding, read the differential while your cursor is still on the next sentence. No tab-switching, no app-context loss.

Question 5

What modalities and subspecialties does it support?

Accepted Answer

RadReason reasons about findings you describe in text, so it's modality-agnostic — CT, MRI, ultrasound, plain film, fluoroscopy, mammography, nuclear medicine. Subspecialty reach spans neuro, chest, abdominal, MSK, breast, peds, and emergency reads. It doesn't read pixels itself; it thinks about what you describe.

Question 6

Why does the differential sometimes feel broad?

Accepted Answer

Because the input was. A calibrated model widens the differential when the description is non-specific — that's a feature, not a bug. The alternative is a confident wrong answer. Add a discriminator (location, enhancement, demographics, prior history) and the list narrows fast.

Question 7

How accurate is it?

Accepted Answer

The reasoning models we use are top-tier and tuned to be calibrated — when they're uncertain, they widen the list rather than fabricate confidence. We don't quote a single accuracy number because performance depends on what you describe and how. Treat the output as a knowledgeable colleague's opinion: usually useful, occasionally wrong, always cross-checkable. Output is generated from de-identified clinical descriptions and should be used as decision support, not a final diagnosis.

Question 8

What are its limitations, and how should I rely on it?

Accepted Answer

RadReason is decision support, not interpretation. It can miss rare entities, weight unusual presentations oddly, or anchor on a discriminator you didn't intend. Use it the way you'd use a curbside consult: a check and a teacher, never the final voice. Cross-check every output against the imaging and the clinical picture.

Question 9

What does Private Session mean?

Accepted Answer

Private Session is enabled by default. In this mode, dictated audio and transcripts are processed in memory and are not stored. No case text is saved unless you explicitly choose to save a case.

Question 10

Does it store my dictation or patient data?

Accepted Answer

RadReason does not store dictated audio or full transcripts when Private Session is on (the default). Dictation is processed in memory to generate results and is not persisted to the database. Transcript content is also redacted from logs. We recommend avoiding direct patient identifiers (names, MRNs, dates of birth) as an additional safeguard. See the Privacy and Security pages for full details.

Question 11

Is this HIPAA compliant?

Accepted Answer

RadReason is designed with privacy in mind, including in-memory processing of dictation and no storage of transcripts in Private Session. However, it is not a HIPAA-covered service unless you have a signed Business Associate Agreement (BAA). For protected workflows, consult your institution's compliance team.

Question 12

How does it generate differentials?

Accepted Answer

Your description is parsed for the imaging features that actually matter — location, modality, signal and enhancement, demographics, clinical context — then run through a reasoning step that ranks plausible diagnoses, flags can't-miss entities, and surfaces what would discriminate between them. The output isn't a list of statistics; it's a structured argument you can audit.

Question 13

Does it adapt to different regions or disease prevalence?

Accepted Answer

The reasoning step weighs prevalence as part of clinical context, so adding a cue like "tropical region", "endemic for TB", or "elderly outpatient" shifts the ranking accordingly. There's no per-region model today, but you can steer it explicitly through what you describe. Region-aware tuning is on the roadmap.

Question 14

Can it help with board prep or learning?

Accepted Answer

Yes — that's what Learn Mode is for. Instead of jumping straight to an answer, it walks through the imaging features, the discriminators, and the reasoning chain the way an attending would on rounds. Especially useful for case review, board-style differentials, and pattern-recognition drills.

Question 15

Why is there a membership fee? Aren't there free AI tools available?

Accepted Answer

Free AI tools are designed for general use — short answers, quick tasks, broad conversations. RadReason is built differently: it is designed for real-time clinical reasoning during radiology workflow. It listens to your findings, structures the problem, and generates a differential instantly — without prompting, rephrasing, or back-and-forth — while you are still interpreting the case. Each case runs a focused reasoning workflow (constructing a differential, testing alternatives, synthesizing an impression) that requires substantially more compute per case than a typical AI interaction. Rather than limiting usage unpredictably or introducing ads, the membership model funds reliable performance during real clinical use, a purpose-built radiology interface, real-time dictation integration with structured outputs, teaching features like guided questions and board-style review, and no ads or sponsored content.

Question 16

How can I get the best results from RadReason?

Accepted Answer

Describe findings the way you'd describe them to a colleague. Lead with anatomic location and modality, then add the features that actually move the needle (size, signal, enhancement, time course) and the clinical context (demographics, history, presentation). Concrete beats vague every time — and use Learn Mode when you want the reasoning, not just the answer.

Questions, answered.

Frequently asked questions

Core understanding

Clinical use

Accuracy & safety

Data & privacy

How it works

Membership & pricing

Getting the best results