Radiology AI That
Actually Works.
Complete, structured, clinically signable draft reports across 25+ exam types: chest X-ray, plain radiographs of 20 body parts, four ultrasound modules, and non-contrast head CT. The draft is ready when the radiologist opens the study, inside the tools they already use. No hallucinated findings. No fabricated measurements.
How Voxel Vision Works
Most AI in radiology produces probability scores or bounding boxes that radiologists cannot use. Voxel Vision takes a fundamentally different approach: it produces complete, structured reports that follow the same clinical scoring standards your radiologists use, with built-in safeguards against hallucinated findings. Reports you can actually sign.
The Process
From medical images to clinically signable reports.
Step 1
Image Analysis
Processes complex multi-image studies, handling real-world imaging variations across dozens of images per exam. Organizes and attributes every image to the correct anatomical finding.
Step 2
Structured Reporting
Produces structured, per-finding scoring that follows established clinical standards (such as ACR TI-RADS). Generates EHR-compatible structured output with compliant follow-up recommendations.
Step 3
Continuous Improvement
The system adapts to your radiologists' preferences and reading style over time. The more your practice uses Voxel Vision, the more accurately it reflects your clinical standards.
Thyroid ultrasound study - bilateral anatomy, two planes, burned-in annotations
Every score grounded in image evidence
ACR TI-RADS compliant, mapped to your EHR schema
Illustrative example output
In Daily Clinical Use
25+ exam types across X-ray, ultrasound, and CT, drafting reports on real studies every day. Every draft is reviewed and signed by a physician.
Chest X-ray + Plain Radiographs
Full draft reads for chest X-ray, plus a generic radiograph reader covering 20 body parts: spine, pelvis, hip, knee, ankle, foot, shoulder, elbow, wrist, hand, abdomen, and more. Each body part gets its own subspecialist prompting, so a knee is read like a knee and a cervical spine like a cervical spine.
Every finding is grounded in image evidence with region-level localization, and the reader observes before it concludes.
Code-drawn illustration · no patient data
Ultrasound: Four Modules
- Thyroid: full ACR TI-RADS scoring per nodule
- Pelvic: O-RADS v2022 risk stratification
- Renal: kidneys, cysts, and stone assessment
- Abdominal: liver, gallbladder, CBD, and AAA criteria
Technologist worksheet measurements are ingested as an input to the read, and guideline criteria are implemented in code, not left to the model.
Deep-dive: Thyroid Ultrasound ReaderCode-drawn illustration · no patient data
Head CT (Non-Contrast)
The newest module: a non-contrast head CT reader built by a practicing neuroradiologist. It reads the pixel data with per-series calibration and pulls prior neuro report context into every read, so the draft knows what the last head CT and brain MR said.
Cross-sectional reading is where the coverage expands next; see the roadmap below.
Prior CT head matched
impression on hand for comparison
Prior MR brain impression pulled
same anatomy, different modality
HU calibration per series
density read on calibrated values
Impression
Code-drawn illustration · no patient data
Context Arrives With the Draft
Voxel Vision works alongside Voxel Suite, a workflow layer that gathers the clinical picture while the study is still on the worklist: matched prior reports and impressions, the reason for exam, clinical history, and CT radiation dose capture. By the time the radiologist opens the case, the draft and its context are already waiting in their own dictation software.
Explore Voxel SuiteWorking Prototypes
Concept modules with working prototype pipelines. These are not in clinical use; the outputs below are illustrative, and each module follows the same validation path as the roadmap in Expanding Coverage.
CT Perfusion (CTP) Reporting
Structured radiology reports drafted from perfusion maps generated by clinical perfusion software. The system interprets CBF, CBV, MTT, and Tmax parameters, evaluates core infarct versus penumbra mismatch, and produces a structured stroke perfusion report ready for your EHR.
Illustrative example output
Cardiac Nuclear Medicine (SPECT MPI)
Structured reporting for myocardial perfusion imaging. The system interprets rest and stress SPECT perfusion maps, evaluates regional wall motion, calculates summed stress/rest/difference scores (SSS/SRS/SDS), and produces a structured cardiac perfusion report with risk stratification.
Illustrative example output
DEXA Bone Densitometry
Automated structured reports from dual-energy X-ray absorptiometry (DEXA) scans. Evaluates T-scores and Z-scores across lumbar spine and hip sites, classifies osteoporosis and osteopenia per WHO criteria, flags fracture risk, and generates EHR-compatible structured output with follow-up recommendations.
Illustrative example output
Expanding to New Modalities
For imaging studies that require analysis across dozens or hundreds of slices, we are developing advanced capabilities that handle true 3D spatial reasoning. Same proven approach, applied to increasingly complex imaging. The working prototypes above (CT perfusion, SPECT MPI, DEXA) are on this same path.
Radiologist-Validated Training
Each new modality is trained on expert-verified imaging studies. The AI learns disease-specific visual patterns directly from practicing radiologists.
3D Volumetric Analysis
Advanced processing that preserves spatial relationships across slices, enabling the system to reason about anatomy in three dimensions rather than isolated 2D images.
Disease-by-Disease Expansion
New modalities are added systematically, starting with common findings and progressing to rare pathology. Each expansion is clinically validated before deployment.
Target Modalities
Compute Infrastructure
GPU-accelerated medical image processing, optimized inference for real-time analysis, advanced compute infrastructure
Cloud AI platform for model training, scalable inference APIs, secure cloud storage and processing
GPU processing, secure medical image storage, scalable compute infrastructure
The Compounding Advantage
Voxel Vision diffs its own draft against the report the radiologist actually signs, automatically, on every study. Corrections become standing rules that apply to future reads at that institution. Radiologists can also teach it directly, with plain-language notes and annotated image snips. No retraining cycle, no waiting for a model update.
Impression
Stable 1.2 cm hepatic lesion, likely hemangioma; follow-up as clinically indicated.
Institution style rules
Code-drawn illustration · no patient data
Institution-Specific
Standing rules are scoped to each practice: its preferences, terminology, and clinical standards.
Teach It Directly
A note or an annotated image snip on any read becomes a rule the system applies going forward.
Audit Trail Built In
Every correction is logged with the original output and the updated recommendation.
Deploy Your Way
Two deployment architectures. One system. Your security requirements dictate the choice.
On-Premise
NVIDIA GPU-Accelerated
- Air-gapped - zero PHI leaves your premises
- Purpose-built AI models for radiology
- Full data sovereignty and compliance
- Runs on standard NVIDIA GPU hardware
HIPAA Cloud
Vertex AI + BAA Coverage
- De-identified medical images via secure cloud
- Business Associate Agreement (BAA) covered
- Customer-Managed Encryption Keys (CMEK)
- No GPU hardware investment required
VoxelMD products are in active research and development. Nothing on this website constitutes a cleared or approved medical device, clinical decision support system, or diagnostic tool. No VoxelMD system is intended to replace physician judgment, and all outputs require review and sign-off by a licensed physician before any clinical use. Performance metrics referenced are based on internal testing and have not been independently validated. Silicon Health Solutions LLC (d/b/a VoxelMD) is a California-registered company.
See It In Action
Request early access to explore Voxel Vision's capabilities with your imaging data.