Cervical cytology · Research workspace
Review the field.
See the evidence.
Upload a field, choose a model, and explore its classification alongside a spatial attention map.
Research use only · Not for diagnosis or patient care
New field analysis
Reference fields
Start with a sample
Select a field, then run your chosen analysis.
Sample labels describe the reference fields, not a new model prediction.
Method note
From field image to evidence map
The model learns from image-level labels while exposing the relative contribution of individual regions.
- 01
Tile
The field is divided into overlapping 1280 px regions.
- 02
Filter
Low-information background glass is excluded.
- 03
Encode
ResNet-50 converts each retained region into features.
- 04
Attend
Gated MIL weighs regions and produces one field-level result.
Research context and limitations
Dataset
Trained on 2,253 conventional Pap-smear field-of-view images from PapSmear-BD. The application does not distribute the dataset.
Intended input
High-power conventional Pap smear fields. Whole-slide images, liquid-based preparations, and phone photographs are outside the validated input domain.
Interpretation
Attention reflects model focus and should not be interpreted as a segmentation mask, a calibrated risk estimate, or a clinical diagnosis.