⚠ For research and demonstration purposes only. Not for clinical use.
Galacticos
biopsy intelligence
System Active
--:--:--
1
Upload
2
Validate
3
Classify
4
Segment
5
Results
Image Input
Upload biopsy image
Drag & drop or click to browse
PNGJPGBMP
Analysis Mode
Classification
LoRA Fine-tuned
Base ModelDINOv2-ViT-L
Fine-tuningLoRA (Low-Rank Adaptation)
Input Size224 × 224 × 3
Classes12 (labels 0–11)
Ensemble5-fold cross-validation
Loss FunctionFocal Loss (γ=2)
SchedulerCosineAnnealing
Segmentation
nnUNet
ArchitecturennUNet 2D
ConfigurationAuto-configured pipeline
Input SizeDynamic (auto-adapted)
OutputBinary mask (sigmoid)
Post-processingConnected component analysis
InferenceFold 0
1
Resize
Bilinear interpolation to model input size, preserving aspect ratio with padding
2
Normalize
Scale [0,1] then subtract ImageNet mean/std — stabilizes training convergence
3
Stain Normalization
Macenko method — corrects H&E staining variation across scanners/labs
4
Denoise
Gaussian blur (σ=0.5) — suppresses scanning artifacts without losing structure
Scan Viewer
No image loaded
Upload a biopsy image to begin analysis
History
No analyses performed yet
Analysis Results
Awaiting analysis
Upload an image and run analysis to see results
System Log