Skip to content

What TumorQuantAI predicts

TumorQuantAI applies HistoPLUS to H&E whole-slide image tissue tiles. For each detected cell it writes a predicted cell-type class and level-0 pixel coordinates. It then summarizes detected classes per completed slide and across a cohort.

The workflow adds reproducibility and audit controls around the model: deterministic tile sampling, source fingerprints, source/target physical scale, immutable model revision, container identity, per-slide retry/resume, visual overlays, and failure-aware aggregation.

TumorQuantAI does not:

  • diagnose disease or provide clinical advice;
  • establish that a predicted class is pathologist ground truth;
  • provide tumor-region segmentation unless a separately validated workflow does so;
  • infer outcomes, treatment response, or lymphoma subtype;
  • establish physical scale when source metadata/provenance is absent; or
  • validate biological conclusions.

The public tutorial collection has no diagnostic annotations or pathologist ground truth. A technically successful run therefore demonstrates workflow operation, not model performance or clinical validity.

Next: read research limitations and review QC overlays.