Imaging labels from board-certified radiologists.
Imaging models are judged by radiologists, so they should be trained by radiologists. We commission vetted readers to label studies, write and grade reports, and review each other's work.
Radiology, with proof.
Imaging AI tends to fail in ways a radiologist recognizes immediately: a finding attributed to the wrong side, an incidental nodule ignored because it was not the question asked, a measurement taken on the wrong series, or a generated report that contradicts the images. HUMXN commissions credential-checked radiologists to create the labels and judgments that teach models to avoid those errors, working at the study and series level rather than on isolated, context-free slices.
Imaging data is sourced only from consenting sources, with consent bound to each item. Studies are fingerprinted with frame and perceptual hashes at intake to catch duplicates, report text is scanned for personal information and masked, and every label records whether it came from the author, a model suggestion or an expert reviewer. A second radiologist reviews each item before delivery, and you receive a signed receipt and audit trail for the whole dataset.
For radiologists, this is paid, remote reading work in your own subspecialty. You label findings, draw segmentations, grade model-generated reports and review a colleague's reads, at a rate shown before you accept each job. The work creates data for AI systems. It is not clinical reporting, does not contribute to any patient's care and creates no physician-patient relationship. Your identity is not published, and approved work is paid automatically through Stripe.
Built to your specification.
Every engagement starts from a written spec and a pilot batch. These are the most common requests we source for radiology.
Findings and classification labels
Study-level and series-level labels for findings, severity and follow-up recommendations, defined in a written labeling guide.
Segmentations and measurements
Organ, lesion and structure masks with measurements taken on the correct series and phase, exportable as COCO or in your format.
Report annotation
Radiology reports labeled for findings, laterality, negation, uncertainty and recommendations, linked to the images they describe.
Generated report grading
Radiologists score model-written reports for missed findings, hallucinated findings, laterality errors and clinically significant omissions.
Image-report pairs
Consented studies paired with radiologist-written reports for vision-language training, each pair fingerprinted and provenance-signed.
Evaluation sets
Held-out, subspecialty-tagged cases with adjudicated reference labels, built to measure the failures that matter in practice.
01Laterality and location
Left-right swaps and wrong anatomical levels are common model errors and need readers who check every label against the images.
02Incidental findings
Models trained on narrow labels ignore what they were not asked about. Radiologists label the whole study, not just the target finding.
03Series and sequence context
A measurement on the wrong series or contrast phase is a wrong label. Study-level context keeps labels tied to the right images.
04Inter-reader variation
Second-reader review and recorded label provenance show where readers agreed, where they were corrected and by whom.
Paid work in your field.
Remote and flexible, with the rate stated before you accept. Every job is reviewed by a second expert, and approved work is paid automatically.
- A medical degree and completed radiology residency
- Board certification or the equivalent specialist registration
- A current or recent license in your jurisdiction
- Several years of independent reading experience
- Fellowship training in a subspecialty such as neuro, MSK, thoracic or breast
Label studies
Mark findings, severity and laterality at study and series level, following a written labeling guide.
Draw segmentations
Outline lesions and structures and take measurements on the correct series and phase.
Grade generated reports
Compare model-written reports with the images and flag missed, invented or mislocated findings.
Act as second reader
Review another radiologist's labels and adjudicate disagreements before the batch is delivered.
Radiology: asked often.
Do you work with full DICOM studies or single images?
Both, depending on the request. We prefer study-level work, because many labeling errors come from missing context: the wrong series, the wrong contrast phase or a prior comparison. Labels can be delivered against the original series, exported as COCO masks and boxes, or packaged as WebDataset or Parquet for training pipelines, each with a signed receipt.
Where does the imaging data come from?
Imaging is sourced only from consenting sources, with the consent record bound to each item before it enters a dataset. Every study is fingerprinted at intake and checked for duplicates, report text is scanned for personal information and masked, and each item carries signed provenance. If you supply your own images for labeling, the same checks and provenance apply to the labels we add.
Is this work clinical reporting?
No. You are creating labels and judgments for AI training and evaluation. Your reads are not used in any patient's care, do not replace a clinical report and create no physician-patient relationship. Each job explains exactly how the labels will be used, so you can decide whether to accept it.
Do I need board certification to join?
Most radiology jobs require board certification or equivalent specialist registration, and many ask for a specific subspecialty. Some tasks, such as basic anatomy labels or report text annotation, may be open to senior residents or fellows. We verify your identity and credentials before offering work, and each job lists the qualifications it requires.
How do I get paid, and how much time does it take?
Rates are set by field and task and shown before you accept, so you know what a batch pays before starting. You take jobs that fit your schedule and work remotely within the stated deadline. Work approved by a second reader is credited to your balance and paid automatically through Stripe once your identity and tax details are verified.
Can we commission an exclusive imaging dataset?
Yes. Commissioned labels and evaluation sets can be licensed exclusively, with scope, term and exclusivity agreed in writing before work begins. Most teams start with a pilot batch to agree on the labeling guide and acceptance criteria, then move to a standing order so new labeled studies arrive at the same standard each month.
Radiology, done by experts.
Need expert-made data in this field, or have the expertise to make it? Start here.