MICCAI 2022 MELA Challenge:¶

Mediastinal Lesion Analysis¶

1. Dataset Overview¶

An illustration of a CT scan from the axial, coronal, and sagittal view.
Inside the 3D bounding box is a mediastinal lesion.

This challenge aims to automatically detect mediastinal lesions from computed tomography (CT) scans. In this challenge, we build a large-scale dataset termed MELA, which contains 1100 CT scans collected from patients with one or more lesions in the mediastinum. The MELA dataset is split into a subset of 770 CT scans for training, a subset of 110 CT scans for validation, and a test set of 220 CT scans for evaluation. Experienced radiologists annotated each mediastinal lesion in each CT scan by drawing a bounding box wrapping around the lesion from the axial, coronal, and sagittal direction as close as possible. Each mediastinal lesion corresponds to an annotation consisting of the coordinates and lengths of the ground truth bounding box in three-dimension. The CT scans are in ".nii" format, and the annotations are in ".csv" format.

Data split of MELA dataset.

2. Task: Detection¶

In this task, the participants are expected to detect the mediastinal lesions from CT scans.  The evaluation is based on FROC analysis for detection. Training and validation cases contain annotations with 3D bounding boxes, which were drawn as close as possible to wrap around the mediastinal lesions from axial, coronal, and sagittal direction by experienced radiologists.

Evaluation of Task: Detection¶

Please refer to the evaluation page for details.