How PhotonMagnet works
From a folder of images and a patient's chart to a structured second read.
Your organization signs up
The person who signs up becomes the organization's administrator. They invite colleagues by email as clinicians (upload, add records, run analyses) or viewers (read only), and can change roles or disable an account at any time.
Add a patient
Patients are identified by a reference your organization chooses (PT-0001…) and an optional label. Age band and sex help the AI; names, dates of birth and record numbers are not needed and should not be entered. Animals too: set the species to Auto-detect and the AI identifies it from the first images you upload (and checks it again in every analysis), or choose it yourself - human, dog, cat, horse and more. If the images disagree with a species chosen by hand, PhotonMagnet tells you.
Upload a study
Pick the modality, body part and the clinical question, then drop the files. DICOM (single or multi-frame, uncompressed) is opened in your browser: each frame is rendered with its window settings and only technical tags - modality, slice thickness, pixel spacing - are kept. JPEG, PNG and WebP are re-encoded, which drops EXIF data. PDFs are stored as they are.
Add the history - now or later
Records are typed notes, lab results, medication lists, prior reports, PDFs, and images: earlier X-rays or scans, photos of films or documents - several at once, each becoming its own record. Earlier images are compared with the current study in the report's Comparison section. They belong to the patient and can be tied to one study. A study page tells you when records were added after its last analysis, so you can run it again.
Run the analysis
PhotonMagnet sends up to 16 images (a long series is sampled evenly from top to bottom) and the records to the AI engine, which writes a radiology report for the clinician - patient information and clinical history, examination and technique, comparison with the patient's earlier studies, findings region by region (normal structures as well as abnormalities), impression and recommendations - plus the key findings marked on the images, a differential, measurements to verify and limitations, and a plain-language view for the patient.
Review, print, decide
Flagged regions are drawn on the images in the viewer. Switch between the clinician report and the patient view, print either one (the printed report has a line for the reviewing clinician's signature), or export it as a PDF or as DICOM (an encapsulated PDF or a structured report) for a PACS or viewer, and compare with earlier analyses. The decision always stays with the clinician.
What the AI does not do
It does not diagnose, it does not see images it was not sent, and it does not replace a radiologist's report. It is decision support: a structured second look that a qualified clinician reviews.
Want to see what de-identification does? Download the sample DICOM (a synthetic phantom whose header holds a fake name and record number) and upload it: only the pixels and technical tags arrive.
Ready to try it?
Your organization starts with a sample patient to analyse.