{
  "fetched_at": "2026-10-19T06:30:59Z",
  "source_url": "https://www.kaggle.com/competitions/eurecom-aml-2021-challenge-2/data",
  "pages": [
    {
      "name": "data-description",
      "content": " This challenge has been adapted from the first edition of the “Retinal Fundus Glaucoma Challenge” (a.k.a. REFUGE) which was included in the MICCAI conference of 2018. A report describing the original challenge and the top performing solutions for this edition is available [[2]](https://arxiv.org/abs/1910.03667). The dataset at your disposal consists of 1200 retinal fundus images (400 for training, 400 for validation, 400 for testing), issued from different cameras, and annotated by human experts. The annotations are only provided for the train and validation splits, and are described below. Among the provided samples, 10% represent positive diagnosis of glaucoma, therefore the dataset is unbalanced.\n\nThe annotations in `index.json` contain for each sample :\n\n1. **The Glaucoma diagnosis:** a `1` for a glaucomatous eye, `0` for a healthy eye.\n\n2. **A segmentation map for the OD and OC:** its size is the same as the input image.\n\n3. **Fovea localization coordinates:** (X,Y) coordinates. The fovea is the sharp central vision point. \n\nTo give you an idea of the Optic Disc and the Optic Cup area you can find illustrations like Figures 2 and 3 in **Overview/Retinal Fundus Images** for training samples in the illustration folder.\n",
      "isPublished": false,
      "hasIsPublished": false,
      "postTitle": "",
      "hasPostTitle": false,
      "mimeType": "",
      "hasMimeType": false
    },
    {
      "name": "Description",
      "content": "## Context\n\n       **Glaucoma** is a prevalent condition related to an abnormal fluid balance in the eye that causes an increase of internal ocular pressure. The increase of pressure gradually damages the eye optic nerve. If undiagnosed, these damages can result in permanent vision loss. Approximately **2.86% of the global population over 40 years old** is affected by glaucoma [[1]](https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1856963/). Patients affected by glaucoma usually do not present symptoms in the early stages of the disease, while an early diagnosis is critical to prevent irreversible damages. It is thus important to develop inexpensive detection methods, in order to massively and systematically control patients, before the symptoms appear.\n\n       One way to diagnose glaucoma is to perform a visual examination of the inside back surface of the eye (fundus). The images are obtained by special cameras through a dilated pupil. The main advantage of this technique, called **\"Retinal Fundus Imaging\"**, is to be brief (usually a minute long) and painless to the patient, making it suited for simple routine checks. However, establishing an accurate diagnosis from these images is particularly difficult. This task is currently performed by human experts, which represents a significant financial burden. Therefore, automatic glaucoma detection methods from fundus imaging are both highly desirable and challenging.\n\n       As most of the applications in medical imaging, this research topic essentially suffers from two limitations:\n        1. **The lack of data**: such data is costly to annotate, essentially due to the human experts involved.\n        2. **The disparity of data**: the data can vary a lot depending on the imaging device involved and its calibration. Therefore, one solution developed in one particular medical center may perform poorly in another medical center. It is thus crucial to design solutions that can achieve good generalization performance with few data.\n\n## Your objective\n\n        Your task in this challenge is to participate in the research of a machine learning solution to automatically detect glaucoma from retinal fundus images. You will be provided with a baseline method based on both a Convolutional Neural Network and some useful human experts knowledge, which you are free to use as a basis. Previous works exist in this application domain [[2]](https://arxiv.org/abs/1910.03667), and may eventually inspire you. Your goal will be to improve our baseline. As usual, model performance will not be the only evaluation criterion, and serendipity will also be taken into account.\n\n       In the various sections we introduce some knowledge on glaucoma detection from retinal fundus images, describe the dataset, present the provided baseline method, and detail the rules of the challenge.\n",
      "isPublished": false,
      "hasIsPublished": false,
      "postTitle": "",
      "hasPostTitle": false,
      "mimeType": "",
      "hasMimeType": false
    },
    {
      "name": "Evaluation",
      "content": "## Metric\nThe metric for the evaluation of the results is the **Area Under Curve** (AUC) which is the area under the [**Receiver Operating Characteristic** (ROC)](https://en.wikipedia.org/wiki/Receiver_operating_characteristic) curve. \n\n## Submission format\nTo evaluate your results on the leaderboard, you should submit a single csv file reporting the results of your model on the test images. Your csv submission file should report the image ID and the prediction of your model between `0` (healthy) and `1` (glaucoma) in the following format:\n\n```\nId, Predicted\nT0001, 0.9738\nT0002, 0.1625\nT0003, 0.7638\n...\n```\n\nYou can download an example submission file (sample_submission.csv) on the **Data** page.",
      "isPublished": false,
      "hasIsPublished": false,
      "postTitle": "",
      "hasPostTitle": false,
      "mimeType": "",
      "hasMimeType": false
    },
    {
      "name": "rules",
      "content": "- Only the provided data shall be used to train models, and the participants are expected to respect the provided data splits (e.g. “train” for training and “val” for validation only).\n\n- Your group ideas are expected to be original and not inspired from other groups.\n\n- The leaderboard rank is not the only evaluation criterion. The quality of the research approach and the originality of your work will also be taken into account.\n\n- Although your main objective is the glaucoma detection task, you are free to investigate the OD/OC segmentation and the fovea focalization tasks, for which we provide you metrics in our notebook. Every effort in this direction will be taken into account, even if it shouldn’t be prioritized.\n\n- You can use the segmentation masks of the test set provided in the data folder to investigate the OD/OC segmentation task, and present some bonus results in your report. However, **any** use of these segmentations masks in the pipeline of your method or for training your model is **strictly prohibited**.",
      "isPublished": false,
      "hasIsPublished": false,
      "postTitle": "",
      "hasPostTitle": false,
      "mimeType": "",
      "hasMimeType": false
    },
    {
      "name": "Retinal Fundus Images",
      "content": "On the ocular anatomy diagram shown in Figure 1, we can identify the different parts of the eye related to the retinal fundus images.![Figure 1: Eye anatomy with a focus on the fundus part of the eye](https://i.postimg.cc/1XDy8XRc/scheme-eye.png)\n                            Figure 1: Eye anatomy with a focus on the fundus part of the eye\n\nDamages on the optic nerve can be spotted by professional examination of the Optic Disc (OD) and the Optic Cup (OC) located inside the OD [[3]](https://www.hindawi.com/journals/joph/2015/180972/).\n\nFigures 2 and 3 respectively show a healthy eye and a glaucomatous eye, taken from the dataset.\n![Figure 2: Optic Disc, Optic Cup and Fovea in fundus healthy eye fundus image](https://i.postimg.cc/C5NYYtxw/OC-OD-F.png)\n                                          Figure 2: Optic Disc, Optic Cup and Fovea in fundus image\n\n![cup_h_disc_h.png](https://i.postimg.cc/CMmkcKw6/cup-h-disc-h.png)\nFigure 3: Disc height (OD height) and Cup height (OC height) used in vCDR are represented on a glaucomatous eye fundus image.\n\nAs shown in Figures 4 and 5, one relevant feature for glaucoma detection is the vertical Cup-to-Disc Ratio (vCDR) which is the ratio between OC and OD heights, both represented in Figure 3. \n\n![vCDR](https://i.postimg.cc/KztCSpmJ/Capture-d-cran-2021-04-14-10-25-09.png)\n\nAlthough other relevant markers exist (Figure 5), the vCDR is the easiest to exploit. The baseline method presented in the following section is thus based on the computation of the vCDR.\n\n![Figure 4: Comparison of Optic Disc (OD) and Cup (OC) for healthy and Glaucomatous eyes. A cupping effect is observed in the second case.](https://i.postimg.cc/GpNKXJGK/g-vs-h.png)\nFigure 4: Comparison of Optic Disc (OD) and Cup (OC) for healthy and Glaucomatous eyes. A cupping effect is observed in the second case.\n\n![Figure 5: Three relevant markers for glaucoma.](https://i.postimg.cc/W3GZ0Mnh/g-signs.png)\n                                             Figure 5: Three relevant markers for glaucoma.",
      "isPublished": false,
      "hasIsPublished": false,
      "postTitle": "",
      "hasPostTitle": false,
      "mimeType": "",
      "hasMimeType": false
    },
    {
      "name": "Baseline",
      "content": "Our pipeline (Figure 6) is made of four main steps, and aims at diagnosing glaucoma based on the vCDR of the eye, which is obtained from OD and OC segmentations. The four steps are described below.\n\n\n**Pre-processing:** Input images are of high resolution, which would considerably increase the processing time of a deep convolutional network. We thus systematically resize (without crop) the input data to 256*256.\n\n**Deep CNN:** We use a UNet [[4]](https://towardsdatascience.com/understanding-semantic-segmentation-with-unet-6be4f42d4b47) , which is a CNN encoder-decoder designed for image segmentation, and well known in medical imaging. Its main specificity is to combine the encoder feature maps with those of the same scale in the decoder, in order to restitute more accurate spatial information (it creates a sort of \"U\" shape of the network). The network outputs a prediction maps for both OD and OC segmentation, of the size of its input (256*256 in our case), and is trained with the segmentation ground truth. We monitor the [Dice Coefficient](https://towardsdatascience.com/metrics-to-evaluate-your-semantic-segmentation-model-6bcb99639aa2) during training to monitor the segmentation network performances.\n\n\n**Post-processing:** In order to accurately determine the OD and OC heights, we only retain the largest connected component of the prediction. This will filter potential noise in the prediction.\n\n\n**Linear classifier:** We use a simple logistic regression model, which predicts the final classification probability from the vertical Cup-to-Disk Ratio (vCDR). The classifier is trained with the classification labels.\n\n![Pipeline](https://i.postimg.cc/02xPZhYt/pipeline.png)\n                                                        Figure 6: Baseline solution pipeline\n\n\nAlthough this baseline produces good performance, there are many ways you could think of to improve it, either by modifying some of the four processing steps, or by improving/replacing the pipeline itself.",
      "isPublished": false,
      "hasIsPublished": false,
      "postTitle": "",
      "hasPostTitle": false,
      "mimeType": "",
      "hasMimeType": false
    }
  ]
}