Automated Universal Classification 2023¶

Welcome to AUC23, a challenge focused on developing a universal algorithm capable of managing a wide range of 3D medical image classification tasks.

As a participant, your task is to create a unified method that can train an AI model to handle any 3D medical image classification task. The model should be robust, capable of dealing with different image and dataset sizes, 3D imaging modalities, and a variety of tasks. These tasks range from identifying organ abnormalities to estimating cancer risk and type.

To test the generalizability of your training method, you'll be given datasets relating to eight specific tasks:

• Identifying abnormalities like tumors or cysts in kidney CT images
• Detecting clinically significant cancer in prostate MRI images
• Grading primary open-angle glaucoma from retina OCT scans
• Determining the molecular type of cancer from breast MRI images
• Determining the WHO glioma grade from brain MRI scans
• Estimating the malignancy risk of pulmonary nodules from lung CT images
• Classifying rib fractures in CT scans
• Diagnosing COVID-19 from lung CT scans

Throughout your model's development, you will submit task-specific algorithms on https://grand-challenge.org/algorithms/. By monitoring your performance on the development leaderboards, which include only a portion of the test data for each task, you can continue to refine your model. The end goal is to design a single codebase that can tackle all tasks.

After the model development phase, you will have one more opportunity to submit models based on your unified method, this time on the full test sets. The team that performs best across all tasks will emerge as the winner. Join us in this exciting journey towards universalizing medical image classification.