GBMS : ICPR 2024 Competition on Segmentation of Pre-Operative and Post-Operative Glioblastoma from MRI

News


The training data will be available shortly.

Overview


Glioblastoma (GBM), a formidable form of brain cancer, is characterized by a median survival of 15–16 months and stands as the most prevalent and deadliest malignant brain tumour among adults.
The GBMS Challenge 2024 is a worshop that introduces both pre and post-operative MRI scans of glioblastoma for segmentation. The researchers are invited to propose strategies that are trained on pre-operative and post-operative cases and segments post-operatvie cases with acceptable accuracy.

Dates


May 30, 2024 Release of training data
June 15, 2024 Release of validation data
June 27, 2024 First deadline for submission of methods
July 10, 2024 Final deadline for submission of methods

About


The GBMS Challenge

The GBMS Challenge is a workshop dedicated to pre and post-operative glioblastoma segmentation from four MRI sequences including i) pre-contrast and ii) contrast-enhanced T1-weighted, iii) T2-weighted and iv) T2-weighted Fluid Attenuated Inversion Recovery (FLAIR) MRI. While pre-operative MRI data for glioblastoma are abundantly available, there's a scarcity of acquired MRI data for post-operative glioblastoma and its follow-ups. Therefore, models trained on pre-operative data and on a few post-operative cases must possess the capability to accurately segment post-operative glioblastoma and avoid the following false detections:

  1. The segmentation of post-operative cavity should not be mistaken for necrosis as seen in pre-operative MRI, since the intensity values of pre-operative necrosis and post-operative cavity are closely aligned.
  2. It is essential for the trained model to accurately distinguish between true progression and pseudoprogression of residual tumor to prevent misinterpretation.
The number of cases for training data is 60, which includes both pre and post-operative cases, with each post-operative case having at least one follow-up. The data, along with corresponding groundtruths, will be provided. Validation data: 30 cases that includes both pre and post-operative cases, with each post-operative case having at least one follow-up. Groundtruth of validation data will not be provided. Testing data: 20 cases which includes only post-operative case. Participants are permitted to utilize additional data from publicly available datasets to enhance the training. However, if they choose to do so, they are required to address the potential variations in their results when using only GBMS data. Our aim is to address the specific segmentation problem while ensuring a fair comparison among the participating methods.


How to participate in GBMS challenge?

If you are interested in participating, you are invited to download the training set in data section, which includes the four MRI sequences of 60 brain tumor patients. During the validation phases, participants must submit the predicted output of their methods to thee valuation platform for scoring. Once the validation phase concludes, participants are required to select the method they wish to evaluate in the final testing/ranking phase. Participants will have to submit their final method that is to be evaluated on the testing data as a Docker container capable of processing NIfTI images as input and producing corresponding segmentations in NIfTI format as output. The organizing team will automatically calculate evaluation metrics on the test dataset using the segmentation generated by the Docker container. Results will be published on the website's leaderboard section after the conference. However, participants will receive their individual results privately after each submission. We intend to publish a challenge paper that will summarize the challenge results and explore future research directions. The paper will feature contributions from authors representing the top 5 teams on the leaderboard.


Prizes

The 1 st place team will be awarded $300, the 2 nd place team will receive $200, and the 3 rd place team will receive $100.

Motivation


Till date, several groups have developed fully automatic segmentation approaches for pre-operative glioblastoma segmentation, but there have been very few works on post-operative glioblastoma segmentation. While there are numerous publicly accessible pre-operative glioblastoma datasets, the availability of post-operative glioblastoma datasets is notably lacking. We aim to release the first public dataset for both pre-operative and post-operative glioblastoma.

Publicly available datasets of annotated pre-operative GBM MR images have facilitated numerous studies on segmentation, yielding results comparable to human grader. However, there is a dearth of research on annotated post-operative GBM MR images raising questions about the transferability of pre-operative methods. Moreover, it is unclear if the methods that were developed for pre-operative GBM work well on post-operative datasets.

Data


Data Centre: Postgraduate Institute of Medical Education and Research is a public medical university in Chandigarh (PGIMER), INDIA.

Data source(s): A variety of Philips/Siemens MRI scanners (1.5 and 3T) were used to acquire the MRI scans due to the clinical nature of this dataset and the fact that it was taken from multiple studies across many years. The acquisition protocols are different for different scanners, as these represent scans of real routine clinical practice. Specific details (e.g., echo time, repetition time, original acquisition plane) of each scan of each patient will be published as supplementary material together with the challenge meta-analysis manuscript.

Annotation: Each case in the dataset is annotated by two experienced radiologists where one is the primary annotator, and the other is the reviewer. Annotators were selected from a range of experience levels and clinical/academic ranks. Each radiologists have a minimum of 15 years of clinical experience. The annotators were given the freedom to use their preferred annotation tool and choose between a complete manual annotation approach or a hybrid approach involving initial automated annotations followed by manual refinements. Once the annotators finalized the annotations, they were passed to the corresponding reviewer. The reviewer then reviewed the annotations alongside the corresponding MRI scans. If the quality of the annotations was deemed unsatisfactory, they were returned to the annotators for further refinement. This iterative process continued for all cases until the annotations reached a satisfactory quality, as determined by the approver, and were deemed suitable for public release as the final ground truth segmentation labels for these scans.

Participate


The data distribution, registration and automatic evaluation will be handled by GBMS challenge team. The following link let you register into the challenge and await for the administrators confirmation in participating.

Registration

It is highly recommended to use your institutional email address for the registration.

After registration:

You will receive an email with an acceptance or decline in your team participating telling you the reasons. We cordially ask you for your patience while waiting for a response from GBMS team. Later on other mail, you will find an unique link to download the dataset. In case you didn't get to download the data, please send us an email to our email to generate a new link.

Submission


How to submit?

Validation Submission:
Click the following link to submit your predicted outputs for the validation data. You can have two submissions per day. The metrics calculated on your output will be sent to your registered email id.

Validation Prediction Submission

Method Submission

Click the following link to submit your method.

Submit your solution

You will be able to upload your model contained in a zip file following the next steps:

  1. Select the zip file that contains your Dockerfile and model weights (Remember to follow the docker template).
  2. Press the "Submit" button and wait for your docker to be correctly uploaded.

Docker template guidelines:
In order to make a successful submission you can download and follow the comments instructions in the docker template zip which contains a Dockerfile and evaluate.py files with their respective instructions. The following are general rules to submit your solution in the challenges:

Organizers