A Tumor and Liver Automatic Segmentation

The ATLAS challenge at a glance

Keywords: MRI; liver; liver tumors; TARE; hepatocellular carcinoma

The ATLAS challenge is organized in collaboration with the MICCAI Workshop on 2nd Resource-Efficient Medical Image Analysis (REMIA) which will take place online during the MICCAI 2023 edition in Vancouver, Canada.

Liver cancer is the sixth most common cancer in the world and the fourth leading cause of cancer mortality. In unresectable liver cancers, especially hepatocellular carcinoma (HCC), transarterial radioembolization (TARE) can be considered for treatment. TARE treatment involves a contrast-enhanced magnetic resonance imaging (CE-MRI) exam performed beforehand to delineate the liver and tumor(s) which is used to perform dosimetry calculation. Due to the significant amount of time and expertise required by the delineation process, there is a high degree of desirability for automation.

The objective of the "A Tumor and Liver Automatic Segmentation" (ATLAS) challenge is to produce automatic segmentation maps of liver and the tumor(s) using the provided CE-MRI. The use of external dataset or pre-trained models during the training process are authorized if the used datasets/models are publicly available.

Dataset

The dataset comprises 90 T1 CE-MRI scans of the liver from 90 patients with unresectable HCC, along with 90 liver and liver tumor segmentation masks divided into train and test sets with 60 and 30 patients per set, respectively. All cases in the dataset are pathological.

The dataset includes variations in image resolution, contrast phase, MRI sequence, and MRI manufacturer, making it more representative of real-world clinical scenarios. Thus, the image resolution, contrast phase, MRI sequence, and MRI manufacturer for each image are also provided.

Further details regarding the dataset are available in the Dataset section and in the following paper (link to the paper tbd).

Registration

To participate in the ATLAS challenge, participants must create an account using a valid institutional email address.

Once their address is validated, participants can access and download the train set in the Dataset section.

Submission

Submissions to the ATLAS challenge will be done using Docker containers. The container should allow to take a CE-MRI image as input and return a segmentation map of the liver and tumor(s) in the initial resolution of the image as output.

Participants should also provide a mandatory technical report detailing their methods and are encouraged to submit an article to the REMIA workshop before the fixed deadline.

Each participant is authorized to submit twice their algorithms. The best performance over the two submissions will be retained. The details of the submission's conditions are provided in the How to participate section along with a docker container example.

Evaluation

Evaluation of the algorithms will be done on the test set using the provided docker container by the participants. Five different metrics will be used:

• The average dice per structure
• The 5mm surface Dice
• The Average symmetric surface distance
• The Hausdorff distance
• The Root Mean Square Error on tumor burden calculation

Every metric except for the tumor burden will be calculated twice, once for the whole liver healthy (liver + unhealthy liver) and once for the tumor. The code that allows metric calculations is provided with the dataset.

Rank-then-aggregate voting system over the nine scores obtained by the participants will be used to compute the ranking.