UNICORN Banner

Tasks Descriptions

Below are detailed descriptions of all 20 tasks, including a brief summary, the expected inputs and outputs, and the inference time limit on the Grand Challenge platform.

Note: Additional tasks may be introduced in the future, also as additional (hidden) test data.

T1: Classifying HE prostate biopsies into ISUP scores

  • Description: This task aims to classify prostate biopsies and resections into one of the International Society of Urological Pathology (ISUP) grade groups.
  • Algorithm docker input:
    • Whole-Slide Image (.tif): H&E-stained prostate biopsy.
    • Tissue Segmentation Mask (.tif): Binary mask indicating tissue regions (0: background, 1: tissue).
    • Task description (.json): A JSON file containing metadata about the task, including a short description, the domain, modality, and task type.
  • Algorithm docker output:
    • image-neural-representation.json: A JSON file containing a slide-level feature representations extracted from the WSI.
  • Adaptor input:
    • image-neural-representation.json: cases features
    • image-neural-representation.json: few-shots features + isup-grade.json: ground truths
  • Adaptor output: ISUP score (integer between 0 and 5)
  • Time limit: 600s per case.

T2: Classifying lung nodule malignancy in CT

  • Description: The main goal is to predict a risk score for a lung nodule candidate, indicating either a low or high risk of malignancy.
  • Algorithm docker input:
    • Nodule blocks of size (128, 128, 64) (.mha).
    • Task description (.json): A JSON file containing metadata about the task, including a short description, the domain, modality, and task type.
  • Algorithm docker output:
    • image-neural-representation.json: A JSON file containing image-level feature representations.
  • Adaptor input:
    • image-neural-representation.json: cases features
    • image-neural-representation.json: few-shots features + lung-nodule-malignancy-risk.json: ground truths
  • Adaptor output: Malignancy risk score (float between 1 and 0)
  • Time limit: 300s per case.

T3: Predicting the time to biochemical recurrence in HE prostatectomies

  • Description: This task aims to estimate the biochemical recurrence risk of patients undergoing radical prostatectomy surgery.
  • Algorithm docker input:
    • Image-level neural representation
    • Task description (.json): A JSON file containing metadata about the task, including a short description, the domain, modality, and task type.
  • Algorithm docker output:
    • image-neural-representation.json: A JSON file containing image-level feature representations.
  • Adaptor input:
    • image-neural-representation.json: cases features
    • image-neural-representation.json: few-shots features + overall-survival-years.json: ground truths
  • Adaptor output: Time to recurrence (positive float)
  • Time limit: 1500s per case.

T4: Predicting slide-level tumor proportion score in NSCLC IHC-stained WSI

  • Description: This task aims to assess the Tumor Proportion Score, which is computed as the total amount of PD-L1-positive tumor cells divided by the total number of tumor cells in the histology slide. For the purposes of this challenge, the continuous TPS value is discretized into three categories: TPS < 1%; 1% ≤ TPS < 50%; TPS ≥ 50%.
  • Algorithm docker input:
    • Digital Pathology Whole-Slide Image (.tif)
    • Tissue segmentation mask (.tif)
    • Task description (.json): A JSON file containing metadata about the task, including a short description, the domain, modality, and task type.
  • Algorithm docker output:
    • image-neural-representation.json: A JSON file containing a slide-level feature representation extracted from the WSI.
  • Adaptor input:
    • image-neural-representation.json: cases features
    • image-neural-representation.json: few-shots features + pd-l1-tps-binned.json: ground truths
  • Adaptor output: Binned tumor proportion score (integer)
  • Time limit: 600s per case.

T5: Cell detection of signet ring cells in HE-stained WSI of gastric cancer

  • Description: This task aims to accurately predict the (x,y) coordinates of signet ring cells in variable-sized ROIs extracted from whole-slide images of gastric cancer.
  • Algorithm docker input:
    • Annotated ROI extracted from WSI (.tif)
    • Task description (.json): A JSON file containing metadata about the task, including a short description, the domain, modality, and task type.
  • Algorithm docker output:
    • patch-neural-representation.json: A JSON file containing patch-level embeddings.
  • Adaptor input:
    • patch-neural-representation.json: cases features
    • patch-neural-representation.json: few-shots features + cell-classification.json: ground truths
  • Adaptor output: Coordinates (x, y, 0.5) of the detected cell
  • Time limit: 300s per case.

T6: Detecting clinically significant cancer in prostate MRI exams

  • Description: The main goal of this task is the 3D detection of clinically significant cancerous lesions (ISUP score greater than or equal to 2).
  • Algorithm docker input:
    • Transverse T2 Prostate MRI (Transverse T2 MRI of the Prostate)
    • Transverse HBV Prostate MRI (Transverse High B-Value Prostate MRI)
    • Transverse ADC Prostate MRI (Transverse Apparent Diffusion Coefficient Prostate MRI)
    • Task description (.json): A JSON file containing metadata about the task, including a short description, the domain, modality, and task type.
  • Algorithm docker output:
    • patch-neural-representation.json: A JSON file containing patch-level embeddings.
  • Adaptor input:
    • patch-neural-representation.json: cases features
    • patch-neural-representation.json: few-shots features + prostate-cancer-likelihood.json: ground truths + cspca-detection-map.json: ground truths
  • Adaptor output:
    • Case-level Cancer Likelihood Prostate MRI (Case-level likelihood of harboring clinically significant prostate cancer, in range [0,1].)
    • Transverse Cancer Detection Map Prostate MRI (Single-class, detection map of clinically significant prostate cancer lesions in 3D, where each voxel represents a floating point in range [0,1].)
  • Time limit: 600s per case.

T7: Detecting lung nodules in thoracic CT

  • Description: The main goal of this task is to accurately detect pulmonary nodules in both clinical routine chest CT scans and screening settings.
  • Algorithm docker input:
    • Thorax abdomen scan (.mha)
    • Task description (.json): A JSON file containing metadata about the task, including a short description, the domain, modality, and task type.
  • Algorithm docker output:
    • patch-neural-representation.json: A JSON file containing patch-level embeddings.
  • Adaptor input:
    • patch-neural-representation.json: cases features
    • patch-neural-representation.json: few-shots features + nodule-locations.json: ground truths
  • Adaptor output: Prior nodule locations with AnnotationID, CoordX, CoordY, and CoordZ, and Probability (float) between 0 and 1
  • Time limit: 300s per case.

T8: Cell detection of mitotic figures in breast cancer HE-stained WSIs

  • Description: The main goal of this task is to accurately predict the (x,y) coordinates of each mitotic figure in a region of interest extracted from a whole-slide image.
  • Algorithm docker input:
    • Annotated ROI extracted from WSI (.tif)
    • Task description (.json): A JSON file containing metadata about the task, including a short description, the domain, modality, and task type.
  • Algorithm docker output:
    • patch-neural-representation.json: A JSON file containing patch-level embeddings.
  • Adaptor input:
    • patch-neural-representation.json: cases features
    • patch-neural-representation.json: few-shots features + mitotic-figures.json: ground truths
  • Adaptor output: Coordinates (x, y) of the detected mitotic cells
  • Time limit: 300s per case.

T9: Segmenting ROIs in breast cancer HE-stained WSIs

  • Description: This task aims to segment tumor and stroma tissue in breast cancer histopathological images.
  • Algorithm docker input:
    • Annotated ROI extracted from WSI (.tif)
    • Task description (.json): A JSON file containing metadata about the task, including a short description, the domain, modality, and task type.
  • Algorithm docker output:
    • patch-neural-representation.json: A JSON file containing patch-level embeddings.
  • Adaptor input:
    • patch-neural-representation.json: cases features
    • patch-neural-representation.json: few-shots features + tumor-stroma-and-other/<uuid>.tif: ground truths
  • Adaptor output: A multi-class segmentation mask, matching the input shape, with pixel values indicating class labels — 0 for background, 1 for tumor, 2 for stroma, and 3 for other.
  • Time limit: 300s per case.

T10: Segmenting lesions within ROIs in CT

  • Description: The goal of this task is to segment 3D masks of lesions in CT scans.
  • Algorithm docker input:
    • Stacked 3D CT lesion volumes (.mha)
    • Task description (.json): A JSON file containing metadata about the task, including a short description, the domain, modality, and task type.
  • Algorithm docker output:
    • patch-neural-representation.json: A JSON file containing patch-level embeddings.
  • Adaptor input:
    • patch-neural-representation.json: cases features
    • patch-neural-representation.json: few-shots features + ct-binary-uls.json: ground truths
  • Adaptor output: Segmentation masks (.mha)
  • Time limit: 600s per case.

T11: Segmenting three anatomical structures in lumbar spine MRI

  • Description: This task aims to segment vertebrae, intervertebral discs, and spinal canal in lumbar MRI.
  • Algorithm docker input:
    • Sagittal T1 and T2 MRI (.mha)
    • Task description (.json): A JSON file containing metadata about the task, including a short description, the domain, modality, and task type.
  • Algorithm docker output:
    • patch-neural-representation.json: A JSON file containing patch-level embeddings.
  • Adaptor input:
    • patch-neural-representation.json: cases features
    • patch-neural-representation.json: few-shots features + sagittal-spine-mr-segmentation.json: ground truths
  • Adaptor output: Singular segmentation mask with a similar size and direction as the input image (.mha)
  • Time limit: 600s per case.

Language Tasks

T12: Predicting histopathology sample origin

  • Description: This task aims to classify the type of histopathology material described in the report as either a biopsy, resection, or excision.
  • Algorithm docker input:
    • NLP Task Configuration (JSON object containing a configuration for NLP Tasks. The object must include the following properties: "jobid," "task_name," "input_name," "label_name," "recommended_truncation_side," and "version.")
    • NLP training dataset
  • Algorithm docker output: NLP predictions dataset
  • *Time limit: 7200s per dataset.

T13: Classifying pulmonary nodule presence

  • Description: This task aims to determine if a pulmonary nodule is mentioned in the radiology report.
  • Algorithm docker input:
    • NLP Task Configuration (JSON object containing a configuration for NLP Tasks. The object must include the following properties: "jobid," "task_name," "input_name," "label_name," "recommended_truncation_side," and "version.")
    • NLP training dataset
  • Algorithm docker output: NLP predictions dataset
  • Time limit: 7200s per dataset.

T14: Classifying kidney abnormality

  • Description: This task aims to identify the presence of significant kidney abnormalities based on the radiology report.
  • Algorithm docker input:
    • NLP Task Configuration (JSON object containing a configuration for NLP Tasks. The object must include the following properties: "jobid," "task_name," "input_name," "label_name," "recommended_truncation_side," and "version.")
    • NLP training dataset
  • Algorithm docker output: NLP predictions dataset
  • Time limit: 7200s per dataset.

T15: Predicting Hip Kellgren-Lawrence score

  • Description: This task aims to classify the radiology report by the Hip Kellgren-Lawrence score, ranging from 0 to 4.
  • Algorithm docker input:
    • NLP Task Configuration (JSON object containing a configuration for NLP Tasks. The object must include the following properties: "jobid," "task_name," "input_name," "label_name," "recommended_truncation_side," and "version.")
    • NLP training dataset
  • Algorithm docker output: NLP predictions dataset
  • Time limit: 7200s per dataset.

T16: Classifying colon histopathology diagnosis

  • Description: This task requires predicting the sample type (biopsy or polypectomy) and diagnosis type, with options including hyperplastic polyps, low-grade dysplasia, high-grade dysplasia, cancer, serrated polyps, and non-informative.
  • Algorithm docker input:
    • NLP Task Configuration (JSON object containing a configuration for NLP Tasks. The object must include the following properties: "jobid," "task_name," "input_name," "label_name," "recommended_truncation_side," and "version.")
    • NLP training dataset
  • Algorithm docker output: NLP predictions dataset
  • Time limit: 7200s per dataset.

T17: Predicting lesion size measurements

  • Description: The goal of this task is to predict the size of all lesions presented in the report in a standardized format.
  • Algorithm docker input:
    • NLP Task Configuration (JSON object containing a configuration for NLP Tasks. The object must include the following properties: "jobid," "task_name," "input_name," "label_name," "recommended_truncation_side," and "version.")
    • NLP training dataset
  • Algorithm docker output: NLP predictions dataset
  • Time limit: 7200s per dataset.

T18: Predicting prostate volume, PSA, and PSA density

  • Description: The goal of this task is to predict the PSA level, the prostate volume, and the PSA density based on the radiology report.
  • Algorithm docker input:
    • NLP Task Configuration (JSON object containing a configuration for NLP Tasks. The object must include the following properties: "jobid," "task_name," "input_name," "label_name," "recommended_truncation_side," and "version.")
    • NLP training dataset
  • Algorithm docker output: NLP predictions dataset
  • Time limit: 7200s per dataset.

T19: Anonymizing report

  • Description: This task requires identifying and tagging personally identifiable information (PII) within reports, including dates, personal identifiers, report identifiers, locations, clinical trial names, times, and ages.
  • Algorithm docker input:
    • NLP Task Configuration (JSON object containing a configuration for NLP Tasks. The object must include the following properties: "jobid," "task_name," "input_name," "label_name," "recommended_truncation_side," and "version.")
    • NLP training dataset
  • Algorithm docker output: NLP predictions dataset
  • Time limit: 7200s per dataset.

Vision-Language Task

T20: generating a caption from a WSI

  • Description: In this task, participants are asked to generate a descriptive caption from a whole-slide pathology image (WSI) using a pre-trained vision-language model.
  • Algorithm docker input:
    • HE-stained whole-slide image (.tif)
    • Tissue segmentation mask (.tif)
    • Task description (.json): A JSON file containing metadata about the task, including a short description, the domain, modality, and task type.
  • Algorithm docker output: Generated caption per image (.json)
  • Time limit: 1500s per case.