DI Radiological Analyzer: a public-safe educational demos for structured radiology reasoning, separating visible evidence, inference, certainty, and needed confirmation.
Title: Radiological Deterministic Intelligence (RDI): A Rule-Based, Auditable Reasoning System for Radiological Image Interpretation and Diagnostic Support
Inventor: Mark S. Weinstein
Organization: Grounded DI LLC
Priority-chain reference: June 11, 2025 deterministic cross-domain reasoning filing
Status: Public educational summary / patent-rights notice
Clinical status: Not clinical AI. Not a diagnostic tool. Not a medical device.
Radiological Deterministic Intelligence, or RDI, establishes a structured reasoning framework for radiology-image interpretation support.
The goal is not to replace licensed radiologists. The goal is to make image reasoning more:
- visible;
- structured;
- auditable;
- uncertainty-aware;
- human-reviewable;
- resistant to unsupported overclaiming.
RDI is designed around a simple public-safe reasoning chain:
Visible Evidence → Interpretation → Certainty → Needed Confirmation
This structure helps separate what is actually seen from what is inferred.
RDI applies deterministic reasoning structure to radiology workflows.
Instead of producing a black-box answer, RDI organizes analysis into inspectable layers:
| Layer | Purpose |
|---|---|
| Visible evidence | What can be directly observed from the image, report, or metadata |
| Anatomical context | The relevant body region, modality, and structural relationships |
| Interpretation | What the visible findings may suggest |
| Certainty | Whether the interpretation is definite, probable, possible, or unresolved |
| Needed confirmation | What additional imaging, clinical context, or specialist review is required |
| Audit trail | The reasoning path used to reach or withhold a conclusion |
RDI requires outputs to distinguish between:
- direct observations;
- anatomy-supported inferences;
- unresolved findings;
- unsupported claims that should not be made.
Example:
coronary artery label
→ tortuous enlarged branch
→ vascular cluster
→ suspected vascular communication
→ drainage site unresolved unless visible or supplied
RDI uses rule-governed reasoning structures rather than open-ended black-box interpretation.
A public-safe logic path may include:
Modality identified
→ body region identified
→ visible structures listed
→ abnormality described
→ anatomical relationships checked
→ diagnostic category proposed cautiously
→ uncertainty preserved
→ confirmation needs listed
RDI outputs should classify findings using reviewable certainty language:
| Certainty Level | Meaning |
|---|---|
| Definite | Directly supported by supplied image/report data |
| Probable | Strongly supported but still requiring confirmation |
| Possible | Plausible but not established |
| Unresolved | Not determinable from supplied data |
This prevents the system from presenting weak inferences as final diagnoses.
RDI is designed to preserve a reasoning trail.
A public-safe audit trace may include:
- image labels used;
- visible findings extracted;
- anatomical relationships considered;
- diagnostic hypotheses generated;
- uncertainty flags;
- missing inputs;
- confirmation requirements;
- human-review status.
The following formulas are presented as educational reasoning scaffolds only.
They are not validated clinical scoring systems, diagnostic probability tools, medical-device algorithms, or treatment decision rules.
SEDI = (E_s - E_b) / D_norm
Purpose:
Qualitatively asks how much a scan appears to deviate from expected baseline complexity.
| Variable | Public-Safe Meaning |
|---|---|
E_s |
Current scan abnormality or complexity load |
E_b |
Expected baseline complexity |
D_norm |
Normalizing factor for modality/body region |
Guardrail:
Do not assign numeric values without a defined rubric and validation dataset.
MPDR = P_m / max(T_s, ε)
Purpose:
Asks whether malignancy concern is being raised faster than the evidence supports.
| Variable | Public-Safe Meaning |
|---|---|
P_m |
Malignancy concern level |
T_s |
Tissue stability or benign-support evidence |
ε |
Small nonzero value to prevent division by zero |
Guardrail:
Do not present this as a true malignancy probability.
RIV = -abs(R_i - R_consensus) / n
Optional readability companion:
AgreementScore = 1 + RIV
Purpose:
Compares an initial interpretation with later expert consensus, ground truth, or report-based review.
| Variable | Public-Safe Meaning |
|---|---|
R_i |
Initial interpretation or coded interpretation |
R_consensus |
Consensus interpretation or ground truth |
n |
Normalization factor |
Guardrail:
Do not compute without a defined scoring rubric and reference standard.
SPMC = (C_r * V_d) / (1 + E_a)
Purpose:
Checks whether the scan is properly matched to the patient and case record.
| Variable | Public-Safe Meaning |
|---|---|
C_r |
Record/credential match confidence |
V_d |
Visual and metadata consistency |
E_a |
Attribution uncertainty or mismatch risk |
Guardrail:
Never assign high patient-study match confidence from an isolated image alone.
OEL = (F_p * S_d) / (E_f + 1)
Purpose:
Asks whether an interpretation is becoming more severe or specific than the evidence supports.
| Variable | Public-Safe Meaning |
|---|---|
F_p |
False-positive pressure |
S_d |
Severity distortion |
E_f |
Evidence firmness |
Guardrail:
Use this to soften unsupported claims and preserve uncertainty.
RDI may be explored as an educational or research-oriented scaffold for:
- radiology reasoning demonstrations;
- trainee education;
- second-read reasoning audits;
- QA review concepts;
- uncertainty-preserving image analysis;
- report-structure review;
- evidence/inference separation;
- human-review workflow design.
RDI is not:
- autonomous diagnosis;
- a radiologist replacement;
- a clinical triage system;
- a treatment recommendation engine;
- FDA-cleared or FDA-approved software;
- a validated clinical scoring system;
- a medical device;
- a system for making patient-care decisions from isolated images.
All real-world radiology interpretation requires qualified professional review.
A real clinical interpretation depends on:
- full imaging dataset;
- DICOM metadata;
- acquisition protocol;
- clinical indication;
- prior imaging;
- patient history;
- laboratory and clinical context where applicable;
- radiologist interpretation;
- institutional workflow and regulatory requirements.
RDI outputs should be treated as educational, draft, or audit-support material only unless separately validated and authorized through appropriate clinical, legal, and regulatory channels.
This repository documents a public-safe educational summary of Radiological Deterministic Intelligence concepts.
The material is provided to demonstrate a structured approach to radiology reasoning:
Finding → Visible Evidence → Interpretation → Certainty → Needed Confirmation
No license is granted to use this material for clinical diagnosis, medical-device deployment, patient triage, treatment guidance, or regulated healthcare use without separate written authorization and appropriate validation.
Inventor / Author: Mark S. Weinstein
Organization: Grounded DI LLC
Project: Radiological Deterministic Intelligence
Repository purpose: Public educational notice and structured demo documentation
This repository is for educational and technical demonstration purposes only.
It does not provide medical advice, clinical diagnosis, treatment recommendations, or patient-specific interpretation. Do not use this repository to diagnose, treat, triage, or manage any patient.
All clinical decisions must be made by qualified healthcare professionals using complete clinical and imaging information.