Planned research
SLZ Trace
Planned research · Synthetic prototype reviewed · Clinical study planned
Study whether original, time-valid clinical reports improve image-grounded liver-lesion segmentation on routine, non-emergency adult contrast abdominal CT. Structured imaging assessment remains separate from histological grading.
Public summary updated
What the synthetic experiments found
The synthetic attribute comparison produced false lesions even on training negatives. Correct, absent, wrong and shuffled presence attributes did not meaningfully change masks. These attributes disclose the generated target class; they are not independent clinical reports.
Next question
Original clinical reports are the selected research track. Report provenance, prediction-time chronology, complete lesion references and affirmatively verified negative examinations are specified in the planning package. The search did not establish whether a suitable matching source exists. Next: establish a permitted source combining CT, its time-valid original report and independently complete lesion references. Useful report guidance remains unproven; the empty-mask experiment has not started.