Tasks Overview¶
The UNICORN challenge includes 20 tasks designed to evaluate model performance across both radiology and pathology. Each task has a defined time limit, which sets the maximum allowed runtime for the algorithm Docker container. For vision and vision-language tasks, each case is processed individually, so the time limit applies per case. For language tasks, the algorithm processes all reports in a single run, meaning the time limit applies to the entire dataset.¶
Note: Additional tasks may be introduced in the future, also as additional (hidden) test data.
ID | Task Name | Task Type | Modality | Domain | Metric | Time limit | Task Description |
---|---|---|---|---|---|---|---|
T1 | Classifying HE prostate biopsies into ISUP scores | Classification | Vision | Pathology | quadratic weighted kappa | 600s | |
T2 | Classifying lung nodule malignancy in CT | Classification | Vision | Radiology | AUC | 300s | View Full Task Details |
T3 | Predicting the time to biochemical recurrence in HE prostatectomies | Regression | Vision | Pathology | censored c-index | 1500s | |
T4 | Predicting slide-level tumor proportion score in NSCLC IHC-stained WSI | Classification | Vision | Pathology | quadratic weighted kappa | 600s | View Full Task Details |
T5 | Detecting signet ring cells in HE-stained WSI of gastric cancer | Detection | Vision | Pathology | F1 score | 300s | View Full Task Details |
T6 | Detecting clinically significant cancer in prostate MRI exams | Detection | Vision | Radiology | average of AUROC and AP | 600s | View Full Task Details |
T7 | Detecting lung nodules in thoracic CT | Detection | Vision | Radiology | sensitivity | 300s | View Full Task Details |
T8 | Detecting mitotic figures in breast cancer HE-stained WSIs | Detection | Vision | Pathology | F1 score | 300s | View Full Task Details |
T9 | Segmenting ROIs in breast cancer HE-stained WSIs | Segmentation | Vision | Pathology | DICE | 300s | View Full Task Details |
T10 | Segmenting lesions within ROIs in CT | Segmentation | Vision | Radiology | DICE, long- and short-axis errors | 600s | |
T11 | Segmenting three anatomical structures in lumbar spine MRI | Segmentation | Vision | Radiology | DICE | 600s | |
T12 | Predicting histopathology sample origin | Classification | Language | Pathology | unweighted kappa | 7200s | View Full Task Details |
T13 | Classifying pulmonary nodule presence | Classification | Language | Radiology | AUC | 7200s | View Full Task Details |
T14 | Classifying kidney abnormality | Classification | Language | Radiology | AUC | 7200s | View Full Task Details |
T15 | Predicting Hip Kellgren-Lawrence score | Classification | Language | Radiology | unweighted kappa | 7200s | View Full Task Details |
T16 | Classifying colon histopathology diagnosis | Classification | Language | Pathology | macro AUC | 7200s | View Full Task Details |
T17 | Predicting lesion size measurements | Regression | Language | Radiology | RSMAPE | 7200s | View Full Task Details |
T18 | Predicting prostate volume, PSA, and PSA density | Regression | Language | Radiology | RSMAPE | 7200s | View Full Task Details |
T19 | Anonymizing report | Named Entity Recognition | Language | Radiology + Pathology | weighted F1 | 7200s | View Full Task Details |
T20 | Generating caption from WSI | Generation | Vision-Language | Pathology | BLEU-4, ROUGE-L, METEOR, CIDER | 1500 s | View Full Task Details |