AI4S · Organ-on-a-chip · Live research prototype

CellScope AI

A working image-quality review assistant for bright-field organ-on-a-chip microscopy. Upload your own authorized image and see a real prediction from our frozen DINOv2 model.

Analyze an image

Upload one image or up to five images from the same acquisition session. PNG, JPEG, TIFF and BMP are supported; 8 MB per image.

What this shows

The model compares full-frame and local DINOv2 image features. When several views from one session are provided, it also uses within-session context.

Development evidence: 81.03% balanced accuracy and 0.882 ROC-AUC on one session-grouped split of 3,072 public development images. A second split of those same images reached 79.81% and 0.868. Architecture choices were informed by this cohort; performance on new laboratory sessions is not established.

Research only. A low score is not a verified biological explanation, and a high score is not clearance for an experiment or a diagnosis. A qualified reviewer decides what to accept.

Uploads are processed in temporary storage and removed after analysis. We do not add them to training. Please submit only images you are allowed to share with this service.

Dataset: Movčana et al., Organ-on-a-Chip Image Dataset. Backbone: Meta DINOv2. No source microscopy images are redistributed by this demo.

CellScope AI

Analyze an image

Model output