Inspection efficiency
Inspection time · noise · resolution · defect information
Janus Sphere Innovations adaptive computational imaging
Adaptive X-ray acquisition
Imaging that decides what to measure next. Janus Sphere Innovations is developing RBYRCT: adaptive imaging systems that use information from intermediate reconstructions to guide subsequent measurements.Janus Sphere Innovations
Imaging that decides what to measure next.
JSI is developing adaptive X-ray acquisition and Ray-by-Ray Computed Tomography (RBYRCT), where information from an intermediate reconstruction can influence which measurement is acquired next.
Current work includes computational validation, acquisition-policy research, and development toward a physical closed-loop prototype.
Measure→
Reconstruct→
Analyze→
Select next measurement→
Repeat
Industrial CT / NDT
Spend the acquisition budget where the inspection needs information.
In industrial CT, adaptive measurement selection could be useful where inspection time, image noise, defect detectability, resolution, and production throughput compete for the same acquisition budget. JSI is investigating whether closed-loop acquisition can make those tradeoffs more deliberate.
Patient and diagnostic value
Dose · diagnostic information · scan experience · time
Same closed-loop architecture. Different objectives, constraints, and evidence requirements.
Focus
Adaptive X-ray acquisition
Prototype
Computational validation → physical prototype
Baselines
Fixed, random, and adaptive strategies
JSI field
A sphere, a signal, and a decision surface.
The Janus Sphere mark becomes a moving field: layered blue glass, silver light, and ray paths bending around the idea at the center.
Translation roadmap
From computational evidence to real-world imaging systems.
Janus Sphere advances RBYRCT through staged evidence: computational validation, physical closed-loop acquisition, application-specific testing, and then responsible translation into industrial or medical imaging systems.
Vision
Define sparse CT as a ray-by-ray decision problem.
Scientific Questions
Identify where adaptive rays can preserve information with fewer measurements.
Engineering Questions
Turn reconstruction theory into controllable acquisition and feedback systems.
Simulation
Compare policies, phantoms, noise models, and ray budgets in software.
Prototype
Build research instruments that expose decisions, uncertainty, and evidence.
Bench Validation
Measure performance against controlled targets, baselines, and repeatable tests.
Preclinical Models
Evaluate biological relevance where the evidence and ethics support it.
Clinical Translation
Design studies, safety cases, and workflows for real medical settings.
Commercialization
Package validated capability into products, partnerships, and deployment paths.
Regulatory and quality foundations.
A credible imaging technology has to grow inside the language of medical device evidence, safety, and repeatable engineering practice.
- FDA pathways
- ISO 13485 quality systems
- IEC 60601 electrical safety
- ISO 14971 risk management
- Design controls
- Verification and validation
Reading beyond imaging.
RBYRCT is an imaging problem, but also a control, sensing, and decision problem. The strongest ideas may come from neighboring fields.
- Robotics
- Aerospace systems engineering
- Control theory
- Experimental design
- Cybernetics
- Human factors
- Systems engineering
Research program
A scientific program, not just a demo.
The WebGL instrument is one visible surface of a larger research arc: turn ray selection into a measurable, auditable control problem for sparse CT reconstruction.
Every ray can be treated as a decision with measurable value.
If acquisition policy, reconstruction state, and target uncertainty are linked tightly enough, RBYRCT may preserve useful information while reducing unnecessary measurements.
Which ray-selection policies maximize diagnostic information under fixed budgets?
The live instrument shows policy comparisons, ray budgets, and reconstruction telemetry.
Noise, motion, limited-angle behavior, dose claims, and clinical utility remain open.
Stress-test phantoms, lesion targets, adaptive policies, repeatability, and failure modes.
Open technical questions
The questions that decide whether RBYRCT becomes real.
- Which ray-selection policies maximize diagnostic information?
- Can adaptive acquisition reduce dose or scan time?
- How does RBYRCT behave under noise, motion, and limited-angle constraints?
- What validation would make the method clinically credible?
Compare fan, random, and adaptive policies across controlled phantoms.
Move from image examples to repeatable target and noise experiments.
Test performance against measured hardware, calibration, and acquisition limits.
Define biological relevance, safety rationale, and ethical study gates.
Translate validated evidence into study protocols and medical workflows.
Engineering Design Reference
The discipline that has to make the invention real.
RBYRCT is being organized as a medical-technology evidence program: intended use, risk, design inputs, verification, validation, and requirements traceability are part of the work from the beginning.
Founder operating thesis
Our job is not to claim a better CT system prematurely. Our job is to build the evidence chain that would let such a claim survive physics, software, regulatory, clinical, and human-factors scrutiny.
The company thesis is that RBYRCT is a control problem as much as an imaging problem: choose the next ray, update the reconstruction, quantify uncertainty, log the decision, and trace every result back to a repeatable test.
Research-stage adaptive breast CT.
We are studying whether ray-level acquisition policies can preserve diagnostic information while avoiding unnecessary measurements.
- Initial context: breast-equivalent phantoms and sparse-view simulation.
- Users in view: researchers, medical physicists, radiologists, and technologists.
- Boundary: not for patient diagnosis or treatment decisions.
Scout, adapt, reconstruct, repeat.
The internal research program treats acquisition as a closed loop: an initial scout estimate, adaptive ray selection, and MART-family reconstruction updates.
- Compare sequential, stochastic, and residual-prioritized policies.
- Measure dose-detectability tradeoffs under fixed ray budgets.
- Keep acquisition decisions replayable and auditable.
Safety starts before hardware.
The risk file has to cover the ways an adaptive imaging system can be wrong, overconfident, poorly calibrated, or hard to use.
- Missed lesions, false positives, artifacts, and undersampling.
- Dose concentration, scatter, detector limits, motion, and drift.
- Software faults, cybersecurity, workflow delay, and use error.
Every claim needs a test behind it.
Simulation is the beginning. The real program is to convert promising behavior into repeatable phantom, bench, software, and workflow evidence.
- Phantom CNR/SNR, MTF, dose maps, repeatability, and robustness.
- Bench steering maps, detector modeling, calibration, noise, and scatter.
- Human review of uncertainty, acquisition maps, and stopping criteria.
Simulation motivates the research thesis.
Internal RBYRCT studies suggest that stochastic and adaptive ray policies can reduce coherent artifacts, prioritize high-residual regions, and improve sparse reconstruction behavior in controlled phantoms.
No clinical claim without clinical evidence.
Dose reduction, lesion detectability, scan time, and clinical utility claims must survive physical phantom work, bench performance testing, regulatory review, and appropriately designed clinical studies.
Medical-device discipline from day one.
The roadmap includes FDA pathway analysis, ISO 13485 quality-system thinking, ISO 14971 risk management, design controls, verification, validation, and human-factors engineering.
Clinical need: detect small targets with bounded exposure
Requirement: adaptive policy preserves global coverage
Verification: phantom CNR, dose maps, repeatability
Evidence: replayable logs, test report, locked dataset
Research and regulatory basis
Published patent application
FDA device pathway overview
FDA human factors guidance
FDA SaMD framework
ISO 13485
ISO 14971
Current status: research simulation and engineering development. RBYRCT is not cleared or approved for clinical diagnosis, patient care, or treatment decisions.
RBYRCT
A reconstruction method built around individual ray decisions.
Conventional CT workflows often assume dense or structured ray acquisition. RBYRCT explores a different control surface: choose, score, and use each ray as part of an adaptive reconstruction loop.
The current work compares fixed fan sweeps, random sparse rays, and targeted policies against visual and quantitative reconstruction feedback.
Live RBYRCT scanner
Steer a ray. Measure attenuation. Reconstruct in real time.
This browser instrument uses the public RBYRCT control-loop model: individual ray selection, line-integral attenuation, Beer–Lambert transmission, and a MART-family multiplicative update. Drag inside the phantom to steer the next ray, then fire it into the reconstruction.
RECONSTRUCTION
initial estimate
Tumor Hunt / Battleship analogy
A game-like way to explain what adaptive ray selection is trying to do.
This section turns RBYRCT into a Battleship-style metaphor: each shot is a measurement, the hidden board stands in for a breast-equivalent phantom, and the goal is not to sweep the whole board blindly but to discover where suspicious structure is hiding with as few shots as possible.
Hidden board
Probe the phantom
Click a square to fire a measurement. Hits suggest a lesion location; near misses leave a weaker signal.
Adaptive belief state
Suspicion map
The browser updates a crude probability field from every shot, showing how RBYRCT-style feedback can guide the next measurement.
Select
Choose each ray from a policy: fan, random, lesion-aware, or uncertainty-driven.
Project
Trace through the phantom and update reconstruction state from sparse evidence.
Adapt
Use coverage, error, and target-region feedback to decide where the next rays go.
Visual research atlas
Making the ray-level instrument legible.
These public-safe 3D studies translate the RBYRCT thesis into geometry: source, steering array, target, detector, feedback loop, and reconstruction evidence in one visual language.
A clean no-text render for explaining the basic physical layout without locking the invention to a single hardware implementation.
The system story becomes explicit: aim, measure, reconstruct, update the policy, and preserve the decision trail.
Non-ionizing optics can help explain steering, sensing, and feedback before the project enters x-ray hardware questions.
A compact bridge from visual explanation to technical diligence: policy comparisons need quantitative evidence behind them.
Breast cancer in context
A technology sits inside a century of medicine, culture, policy, and belief.
RBYRCT is not only a reconstruction question. Breast screening is a social institution: it joins x-ray physics to public-health recommendations, regulation, anxiety, advocacy, access, cost, and the philosophical problem of deciding what is worth measuring.
From radiography to adaptive acquisition.
Breast imaging developed from early x-ray practice into dedicated mammography, digital systems, tomosynthesis, and increasingly computational workflows. Each generation changed not only what could be seen, but what counted as an acceptable tradeoff among dose, resolution, time, and uncertainty.
Screening became part of everyday life.
Breast cancer acquired an unusual public vocabulary: screening reminders, survivorship, pink-ribbon advocacy, family stories, celebrity diagnoses, dense-breast notices, and the emotional weight of a callback after an image. Imaging therefore acts in a cultural world, not only a technical one.
What should we measure next?
Every imaging system allocates attention. Conventional geometries encode that choice in advance. RBYRCT asks whether measurement itself can become responsive: observe, update belief, and spend the next ray where it can reduce uncertainty or test a clinically meaningful hypothesis.
Screening rules are evidence translated into institutions.
Recommendations, quality standards, insurance coverage, device regulation, hospital procurement, and state and federal notification rules determine how imaging reaches people. In the United States, the USPSTF currently recommends biennial mammography from ages 40–74 for average-risk women, while FDA MQSA rules set federal mammography quality standards and now require breast-density notification.
Earlier is valuable, but screening is never consequence-free.
Screening can find cancers before symptoms and reduce breast-cancer mortality, but it can also create false positives, additional procedures, overdiagnosis, and uncertainty. A better imaging technology has to improve the information tradeoff rather than pretend those tensions do not exist.
An invention matters only if it can become infrastructure.
Translation means more than a reconstruction metric: reproducible software, physical steering, calibration, risk management, regulatory evidence, manufacturing, reimbursement, clinician workflow, capital equipment economics, service, and ultimately trust. JSI is building toward that whole chain.
A compressed history
Seeing the breast became progressively more computational.
X-rays open a new way to inspect internal anatomy without surgery.
Dedicated mammographic techniques evolve around breast-specific imaging and lower-dose practice.
The U.S. Mammography Quality Standards Act establishes a national quality framework for mammography facilities.
Digital mammography moves acquisition, storage, display, and analysis into a computational pipeline.
Digital breast tomosynthesis expands quasi-3D screening and makes acquisition geometry more visible to patients and clinicians.
USPSTF screening guidance moves routine average-risk screening to age 40; federal breast-density notification requirements take effect.
Adaptive acquisition asks whether the imaging geometry itself can respond to information as the scan unfolds.
Janus Sphere mission
Make every measurement earn its place.
Our long-range mission is to develop imaging systems that use exposure and acquisition time deliberately: enough information to answer the diagnostic question, fewer measurements that add little, and an auditable record of why the system chose what to measure next. The ambition is large; the development standard is evidence first.
Read the public evidence behind this context
NCI: Breast Cancer Screening
NCI: Mammograms
USPSTF: Breast Cancer Screening
FDA: MQSA Regulations
Book companion
The End of Breast Cancer: Steerable Ray-by-Ray X-Ray Computed Tomography.
The title names the ambition. The public research claim is narrower and testable: RBYRCT studies whether dose-limited imaging can be treated as an information-allocation problem, ray by ray.
Central argument
Scout
Prioritize
Adapt
Under dose constraints, every ray should earn its place.
The book develops Ray-by-Ray Computed Tomography as a proposed adaptive CT framework: after an initial low-dose scout scan, subsequent rays can be directed toward uncertain regions, high-gradient structure, or candidate lesions rather than spent uniformly across the field.
Build an initial uncertainty map with a bounded opening scan.
Score candidate rays by structure, residual error, and local risk.
Spend the next rays where they can reduce ambiguity the most.
CancerZap book assistant
A simulator for the book's measurement-selection idea.
CancerZap turns the book's argument into a public, adjustable research sketch. Move the controls to see how scout dose, ray budget, structural uncertainty, and candidate-lesion focus shift the simulated acquisition policy.
- Focused rays
- 58%
- Coverage floor
- 73%
- Info index
- 79
A scout scan gives the system an initial map of anatomy, gradients, and unresolved regions. CancerZap then treats later rays as a limited budget to allocate toward the parts of the reconstruction still asking the hardest questions.
CancerZap is a research and education simulator for the book companion. It is not a medical device, not a diagnostic tool, not medical advice, and not cleared or approved for patient care.
Evidence panel
Sparse-ray behavior you can inspect.
These images show sparse-ray reconstruction behavior across phantoms, policies, and ray budgets.
Ray budget
100-20k
Experiment ranges across phantom families.
Comparison
RBYRCT vs Fan
Direct reconstruction comparison across matched 1,000, 5,000, and 10,000 ray budgets.
Status
Research prototype
Not a clinical device; built for scientific exploration and collaboration.
Living lab
A place where the instrument starts thinking back.
RBYRCT asks a concrete question at every step: which ray should be measured next, and how much does that decision improve the reconstruction?
Strategy arena
Sparse policies racing under one ray budget.
- Coverage
- 84%
- Target score
- 1.42x
- Ray budget
- 12.5k
Adaptive policy concentrates rays around high-information regions while preserving a global coverage floor.
Ask the reconstruction
Local explanations for what the image is doing.
Adaptive is the current lead because it spends fewer rays on already-covered
regions and keeps sampling near edges, lesions, and high-uncertainty texture.
WebGL demo
A live instrument for inspecting ray decisions.
The demo lets visitors compare adaptive, random, and fan-beam sampling, watch reconstruction telemetry, and preserve a shareable run state at demo.janus-sphere.com.
Contact