Competition Rules

We are pleased to invite submissions to the OCT-Cyst Challenge.

Topics of Interest

We invite submissions addressing, but not limited to, the following topics:

• Development of segmentation algorithms for cysts in OCT images.

• Application of deep learning, machine learning, and hybrid techniques for the Cyst segmentation problem.

• Transfer learning and domain adaptation strategies for OCT segmentation.

• Quantitative analysis and validation of segmentation performance.

• Interpretability and explainability of segmentation models in clinical settings.

• Novel preprocessing and data augmentation techniques for OCT cyst datasets.

Dataset

Participants will be provided with a carefully curated OCT dataset containing images with cyst boundaries annotated by domain experts.

The dataset consist of images from 7 different patient with cysts of different degrees of severity. In particular, the images are divided as follow:
| Patient | No. of Images |
| 001        | 1024                    |
| 002        |   144                     |
| 003        |   144                     |
| 004        |     96                      |
| 005        |     48                      |
| 006        |     48                      |
| 007        |   321                     |
| Tot.         | 1825                    |

For each image containing cysts, a corresponding binary segmentation mask annotated by domain experts is supplied, where cyst pixels are white and background pixels are black. 

The dataset is avaiable upon request at https://forms.gle/e9ZarcZoEadfpkoy8 

Use of other data and models

Participants are allowed to use additional publicly available annotated datasets and publicly available pre-trained models.

Evaluation

The submission process involves the following steps: 

1) Participants submit their algorithms and models using their prefferred platform, sendind an email to the organizators. Submissions should include all necessary files to run the method on the test set, packaged in a convenient format (e.g., Docker container, public/private GitHub repository, compressed Pickle files). 

2) The organizers will execute the submitted methods on a hidden test set.

3) Concurrently, participants must submit a paper describing their model and approach, which will undergo peer review. 

Evaluation of the submitted masks will be based on the Dice Coefficient metric. 

Dice coefficient = 2 * |A ∩ B| / (|A| + |B|)

For the final ranking, only submissions associated with papers accepted after the peer-review process will be considered.

The organizers are available to assist participants during the submission phase. 

The organizers of the challenge reserve the right to disqualify any participant or participating team, at any point in time, on grounds of unfair or dishonest practices.

Submission Guidelines

Researchers interested in contributing to this competition are invited to submit original work that aligns with the topics of interest using easychair . Submissions should adhere to the following guidelines:

1. Paper Format: Papers must be written in English and formatted according to the provided template

2. Length: Manuscripts should not exceed 12 pages, including figures, tables, and references. 

3. Code and Models: Authors are encouraged to publicly share their code and trained models to promote reproducibility.

Authors can find complete instructions of how to format their papers here.

Contacts

Dr. Domenico Amato - domenico.amato01@unipa.it

Ethical Approval 

Retrospective data collection was approved in 10 Dec. 2024 by the local ethics committee CEL Palermo1 protocol n.28