Overview Digital pathology and whole-slide imaging (WSI) present unique evaluation challenges due to their massive scale (gigapixel images), hierarchical viewing patterns, and complex annotation requirements. Rubric provides specialized support for pathology AI evaluation.
SVS Aperio ScanScope format
DICOM WSI DICOM Supplement 145
Additional formats: .mrxs (MIRAX), .vsi (Olympus), .qptiff (PerkinElmer), .czi (Zeiss)
Logging Pathology AI Output client.pathology.log( project = "breast-cancer-detection" , # Slide reference slide_id = "slide_2024_001234" , slide_url = "https://pathology-store.example.com/slides/001234.svs" , # Slide metadata metadata = { "stain" : "H&E" , "tissue_type" : "breast" , "specimen_type" : "core_biopsy" , "scanner" : "Aperio AT2" , "magnification" : "40x" }, # AI analysis output ai_analysis = { "classification" : { "diagnosis" : "invasive_ductal_carcinoma" , "confidence" : 0.94 , "grade" : "2" , "subtype" : "luminal_a" }, "regions_of_interest" : [ { "id" : "roi_001" , "type" : "tumor" , "geometry" : { "type" : "Polygon" , "coordinates" : [[[ 100 , 200 ], [ 150 , 200 ], [ 150 , 250 ], [ 100 , 250 ]]] }, "level" : 0 , # Pyramid level "confidence" : 0.91 , "features" : { "mitotic_count" : 12 , "nuclear_grade" : "intermediate" } } ], "quantitative_metrics" : { "tumor_percentage" : 45.2 , "ki67_index" : 22.5 , "til_score" : 15 } }, # Ground truth expected = { "diagnosis" : "invasive_ductal_carcinoma" , "grade" : "2" , "pathologist_annotations" : [ ... ] } )
Pathology Evaluators Classification Accuracy For slide-level or region-level diagnosis:
{ "type" : "pathology_classification" , "config" : { "classes" : [ "benign" , "atypical" , "dcis" , "invasive_ductal_carcinoma" , "invasive_lobular_carcinoma" ], "hierarchical" : True , # Malignant > IDC/ILC "severity_weighting" : { "benign_as_malignant" : 1.0 , # False positive "malignant_as_benign" : 5.0 # False negative - critical } } }
Detection & Localization For cell detection, mitosis counting, or region identification:
{ "type" : "pathology_detection" , "config" : { "target_objects" : [ "mitosis" , "tumor_cell" , "lymphocyte" ], "iou_threshold" : 0.3 , # Lower threshold for small objects "evaluate_at_levels" : [ 0 , 2 ], # Pyramid levels "metrics" : [ "precision" , "recall" , "f1" , "froc" ] } }
Segmentation Quality For tissue segmentation and tumor delineation:
{ "type" : "pathology_segmentation" , "config" : { "regions" : [ "tumor" , "stroma" , "necrosis" , "normal" ], "metrics" : [ "dice" , "hausdorff" , "surface_distance" ], "dice_threshold" : 0.7 } }
Multi-Resolution Handling WSI pyramids require evaluation at appropriate magnification levels:
# Specify evaluation at specific magnification client.pathology.log( ai_analysis = { "regions_of_interest" : [ { "geometry" : { ... }, "level" : 0 , # Native resolution "effective_mpp" : 0.25 , # Microns per pixel "magnification" : "40x" } ] } )
Rubric automatically handles coordinate transformation between pyramid levels for evaluation.
GeoJSON (Recommended) { "type" : "Feature" , "geometry" : { "type" : "Polygon" , "coordinates" : [[[ 100 , 200 ], [ 150 , 200 ], [ 150 , 250 ], [ 100 , 250 ], [ 100 , 200 ]]] }, "properties" : { "classification" : "tumor" , "confidence" : 0.91 } }
QuPath Export # Import QuPath annotations client.pathology.import_annotations( slide_id = "slide_001" , format = "qupath" , file_url = "https://storage.example.com/annotations/slide_001.geojson" )
ASAP XML client.pathology.import_annotations( slide_id = "slide_001" , format = "asap" , file_url = "https://storage.example.com/annotations/slide_001.xml" )
Quality Control Evaluate pre-analytical factors that affect AI performance:
{ "type" : "slide_quality" , "config" : { "check_focus" : True , "check_staining" : True , "check_artifacts" : True , "check_tissue_folds" : True , "min_tissue_percentage" : 20 } }
Biomarker Quantification For IHC and special stains:
client.pathology.log( metadata = { "stain" : "HER2" , "antibody" : "4B5" }, ai_analysis = { "her2_score" : "2+" , "membrane_completeness" : 0.75 , "intensity_distribution" : { "0" : 0.15 , "1+" : 0.25 , "2+" : 0.45 , "3+" : 0.15 } }, expected = { "her2_score" : "2+" , "fish_result" : "amplified" # Ground truth from reflex testing } )
Next Steps 