BB
BenchBoard

Decision cockpit for pathology foundation-model benchmarking, built to move from broad coverage scans to task-level evidence.

Task Evidence

IMP · cervical_dysplasia_grade (IMP__grade)

Diagnostic Pathology / Histologic Grading & Dysplasia Severity

Inspect benchmark evidence for one task at a time: who leads, how stable the split-level results are, and which neighbouring tasks are likely to transfer intuition.

clsL1 · Diagnostic PathologyL2 · Histologic Grading & Dysplasia SeveritycolorectalH&EInformative evidence

Task metadata

Fields sourced from benchmark_main. Missing values are shown as —.

cls
Identity
task_key
IMP__grade
dataset_name
IMP
task_name
cervical_dysplasia_grade
task_type
cls
Taxonomy
task_family_l1
Diagnostic Pathology
task_family_l2
Histologic Grading & Dysplasia Severity
label_structure
ordinal
num_classes
3
labels
{"0": "0 (non-neoplastic)", "1": "1 (low-grade)", "2": "2 (high-grade)"}
Biology
organ
colorectal
stain
H&E
specimen_type
mixed
timepoint
diagnosis
disease
结直肠癌
Scale & Labels
patient_n
5320
slide_n
5320
label_source
pathology
label_visibility
direct
Result Evidence
result_tier
Informative
result_quality_score
78.8
result_flags
ceilingcrowded_head
result_tier_reason
Stable split behavior: median std 0.005.Clear model separation: significant-win density 0.476.Best observed mean is near the metric ceiling: 0.989.Many contenders sit near the best score: 44.4% within 0.01.
loading…

Similar tasks

Ranked by shared L2 subtype first, then L1 family, organ, type and label structure.