Papers & technical notes.
We publish method and reasoning before results, and results only once they survive validation on data from institutions that contributed nothing to training. Nothing here will outrun what we can defend.
Nothing published yet.
Project HealthX is at Phase 0 — building the data foundation. We have not published papers because we do not yet have results that would survive external review, and we would rather have an empty page than a padded one.
Below is what Phase 0 is expected to produce, listed in advance so the standard we intend to be held to is on the record before any results exist.
What we intend to publish.
Titles and scope are provisional. Each moves onto this page as a full document when it is written and cleared — not before.
Dataset governance and de-identification protocol
How imaging data is received, de-identified against DICOM PS3.15 Annex E, verified, audited, and licensed under a hospital data-sharing agreement. Includes the DPDP Act 2023 basis and the certification a contributing institution receives.
Validation protocol
The evaluation standard we intend to be judged against: patient-level splits, external validation on institutions that contributed nothing to training, sensitivity at fixed specificity, confidence intervals, and the prevalence assumptions used.
DICOM conformance in Indian public hospital equipment
What actually comes off ageing CR and DR units in district hospitals, which parts of the standard get violated, and what has to be repaired at ingest before any model sees a pixel.
Annotation vocabulary and inter-reader agreement
The controlled vocabulary radiologists label against, how it was arrived at with clinical partners, and how disagreement between readers is measured and handled.
If you are a radiologist, hospital, or health department and want to review a protocol before it is published, or contribute data under a sharing agreement, we would rather hear from you early.
info@triomavtech.com