ISIC 2019 Challenge [Closed]

Task

The goal for ISIC 2019 is classify dermoscopic images among nine different diagnostic categories:

  1. Melanoma
  2. Melanocytic nevus
  3. Basal cell carcinoma
  4. Actinic keratosis
  5. Benign keratosis (solar lentigo / seborrheic keratosis / lichen planus-like keratosis)
  6. Dermatofibroma
  7. Vascular lesion
  8. Squamous cell carcinoma
  9. None of the others

25,331 images are available for training across 8 different categories. Additionally, the test
dataset (planned release August 2nd) will contain an additional outlier class not represented in
the training data, which developed systems must be able to identify.

Two tasks will be available for participation: 1) classify dermoscopic images without meta-data,
and 2) classify images with additional available meta-data. Task 1’s deadline will be August 16th.
Task 2 will be August 23th, after release of test meta-data on August 16th. Participants of Task 2
must submit to Task 1 as well, though participants can submit to Task 1 alone.

In addition to submitting predictions, each competitor is required to submit a link to a manuscript
describing the methods used to generate predictions.

Submission

Submissions are made to the
ISIC Challenge submission system, which provides automated
format validation, pre-scoring, metadata editing capabilties.

Evaluation

Goal Metric

Submissions are scored using a normalized multi-class accuracy metric (balanced across categories). Tied positions will be broken using the area under the receiver operating characteristic curve (AUC) metric.

Definition

Normalized (or balanced) multi-class accuracy is defined as the accuracies of each category, weighted by the category prevalence. Specifically, it is the arithmetic mean of the (<category>_true_positives / <category>_positives) across each of the diagnostic categories. This metric is semantically equivalent to the average recall score.

Rationale

Clinical application on skin lesion classification has two goals eventually: Giving specific information and treatment options for a lesion, and detecting skin cancer with a reasonable sensitivity and specificity. The first task needs a correct specific diagnosis out of multiple classes, whereas the second demands a binary decision "biopsy" versus "don’t biopsy". In the former ISIC Challenges, focus was on the second task, therefore this year we want to rank for the more ambitious metric of normalized multiclass accuracy, as it is also closer to real evaluation of a dermatologist. This is also important for the extending reader study, where the winning algorithm(s) will be compared to physicians performance in classification of digital images.

Other Metrics

Participants will be ranked and awards granted based only on the balanced multi-class accuracy metric. However, for scientific completeness, predicted responses will also have the following metrics computed (comparing prediction vs. ground truth) for each image:

Category Metrics
Aggregate Metrics
  • average AUC across all diagnoses
  • malignant vs. benign diagnoses category AUC

Validation Scoring

All submissions to the ISIC Challenge are immediately issued a validation score. This validation score is not intended to be used for algorithm ranking or evaluation, but is provided for a sanity check of submission data (e.g. to guard against instances where prediction labels are mismatched).

The validation score is computed with the goal metric (balanced multi-class accuracy), taken against a small (~100), non-representative, pre-determined subset of images.

For reference, a random submission generates a validation score of about 0.3.

Final Score Release

Final scores and a public leaderboard are released shortly after the conclusion of the ISIC Challenge submission period.

Transparency Statement

The code of the isic-challenge-scoring package is used for actual score computation. This code is open source, permissively licensed, and published, to facilitate external auditing.

Awards

Cash prizes of $4,000, $2,000, and $1,000 will be awarded to the first, second, and third place
participants of image-only and meta-data tasks. USD will be awarded to winners of each of the
tasks. The monetary prizes for the winners of the challenge will be awarded after the MICCAI
Workshop in Shenzhen, China, in October 2019. The prizes are being provided by Canfield Scientific,
Inc., a US company, and are subject to any restrictions incumbent on the sponsor. Winners will be
asked to identify a recipient individual or entity who will be required to provide tax documentation
(U.S. citizens- IRS form W-9, non-U.S. citizens Form W-8 BEN).

The results are presented at the ISIC Skin Image Analysis Workshop @ CVPR 2019.

Sponsors

  • Canfield Scientific
  • IBM
  • MetaOptima

Clinical Chairs

  • Josep Malvehy, M.D. ;
    University Hospital Clinic of Barcelona, Barcelona, Spain
  • Allan Halpern, M.D. ;
    Memorial Sloan Kettering Cancer Center, New York City, NY, USA

Computer Vision Chairs

  • Noel C. F. Codella, Ph.D. ;
    IBM Research, Yorktown Heights, NY, USA

Challenge Co-Chairs

  • M. Emre Celebi, Ph.D. ;
    University of Central Arkansas, Conway, AR, USA
  • Marc Combalia, M.S. ;
    Fundació Clínic per a la Recerca Biomèdica, Barcelona, Spain
  • David Gutman, M.D., Ph.D. ;
    Emory University, Atlanta, GA, USA
  • Brian Helba ;
    Kitware, Inc., Clifton Park, NY, USA
  • Harald Kittler, M.D. ;
    Medical University of Vienna, Vienna, Austria
  • Veronica Rotemberg, M.D., Ph.D. ;
    Memorial Sloan Kettering Cancer Center, New York City, NY, USA
  • Philipp Tschandl, M.D., Ph.D. ;
    Medical University of Vienna, Vienna, Austria
Note: Any organizations/companies affiliated with members of the organizing committee are not excluded from participation in the Challenge, but must assure that their submissions are completely independent of the members of the organizing committee. {# intentionally omit task-listing for 2019, since both tasks are so similar they can be described with just this landing page #}