The other half, beyond the sachet.
A sachet of powder answers “can it be seen.” The remaining question is “can it be seen the same way every time” — and that takes a procedure and a piece of software.
Four layers, and where each one actually stands.
We publish the status of each layer rather than describing the whole platform in the present tense.
Clinical service
Structured reporting support, health management, follow-up
Analysis support
Organ localization; marking regions for physician review; indicating the type of finding they may represent
Annotation
Physician annotation tools, multi-level quality control, expert review
Acquisition
Standardized acquisition procedure, de-identification, quality management
Writing an examination down as a table.
Three fixed stations (head, body, tail), two planes recorded at each, four time points: baseline, +5, +10 and +20 minutes. Those minutes are not clock minutes — every clip is indexed as τ = t − t₀, where t₀ is the moment the patient finishes drinking.
Station × time point × plane forms an acquisition completeness matrix. Before the examination is over, the software can show which cells are still empty.
It depends on whoever is on duty
- What gets seen is what gets recorded; how much is a matter of feel
- Images from different physicians and hospitals cannot be compared
- Missed clips are discovered after the patient has left
- Data cannot be pooled into a trainable dataset
It depends on a fixed table
- What to acquire, when, and in how many planes is all fixed
- The time axis is anchored to ingestion, so patients are comparable
- Whatever is missing is flagged on the spot, while it can still be recovered
- Every examination becomes one structured record
It marks things for the physician. It does not draw conclusions for them.
The established part scores six properties of the acoustic window — filling, uniformity, gas artifact, attenuation, usable area and boundary integrity — and, with the completeness matrix, answers “is this acquisition adequate, and what is still missing.”
The analysis layer in development (S3) provides decision support: it localizes the organ, marks locations in the image that are worth a physician’s second look, and indicates which type of finding each location may represent. It hands those markers, together with the original images, to the physician — and the physician makes the final judgment.
The output of both layers is markers, not diagnoses. The software does not draw conclusions for the physician, and it does not appear in clinical reports.
The software described here is under development and is used in research data collection and internal validation only. Its established functions assess image quality and acquisition completeness; an analysis layer in development is designed to mark regions for physician review and indicate the type of finding they may represent. No layer produces a diagnosis, none is used in clinical reporting, and no clinical decision should be based on its output.
Try the annotation tools.
Two working demonstrations of the S2 annotation layer. Both run entirely in your browser; nothing is uploaded anywhere. Demonstration only — these tools are not in clinical use. They are in the testing stage and are shown here for demonstration purposes only.
Pancreatic Ultrasound Annotation v1.02
One physician, one case. Upload an image, DICOM file or video; step through frames; grade pancreas visibility and image quality; outline structures; record acquisition details (time point, position, plane); capture key frames; export a case package. Built-in de-identification removes burned-in patient information before frames are captured.
Open the demoMulti-Reader Annotation v2.01
Several physicians, the same case. Each reader’s assessments and outlines are stored separately and compared in a reader-comparison view; a batch mode parses acquisition details from standard filenames; frames can be auto-extracted from video at a fixed interval. Exports multi-reader data and an agreement report.
Open the demoThese demonstrations are the physician-facing annotation tools of the established S2 layer. The “Suspected lesion” option in the image assessment means only that an image shows a focal appearance warranting further attention, for data triage and review; it is not a clinical diagnosis. The analysis layer described above (S3) is not part of these demonstrations.
How the system is used in China.
Everything in this section describes deployments in China. It is not a description of current availability in North America, and not an offer.
Health-screening centers
Examinations offered to people without symptoms, where the priorities are throughput, standardization and repeatability.
- A standardized examination package
- Image quality assessed during acquisition, reducing repeat examinations
- A standardized report structure, so a later examination can be compared with an earlier one
- Follow-up and record-keeping support
Hospital departments
An assistive tool for clinical departments, where the priorities are precision, depth and clinical usefulness.
- Organ-localization support during the examination
- Image quality control, reducing operator-to-operator variation
- Structured report templates and consistent descriptive terms
- Traceable audit logs, so studies can be reviewed retrospectively
Data security
Strict de-identification, encrypted storage and transmission; original data is never overwritten.
Traceability
Every step is logged and the provenance of each record can be traced, which is what makes retrospective review and compliance audit possible.
Physician-led
Software output is for physician review only; physician confirmation is the decision point.
Regulated development
We follow the regulations that apply to us and advance productization in stages, not ahead of the evidence and the approvals.
Interested, or thinking about working together?
Research collaboration, distribution partnerships, or simply a conversation about the technology — all are welcome. We do not currently operate in North America, but we are open to the conversation.