AI Challenge in JRC2027
１．Purpose of the Challenge
The Special Committee for Next-Generation Data Science Promotion of the Japanese Society of Radiological Technology (JSRT) promotes activities aimed at enhancing members’ understanding and application of artificial intelligence (AI) and data science.
As part of these activities, we are pleased to announce the AI Challenge in JRC2027, following the success of the AI Challenge held in conjunction with JRC2026.
Last year’s challenge was conducted using Neural Network Console (Sony), a development environment that does not require programming. For this year’s challenge, participants will have greater flexibility and may freely choose from a variety of development environments according to their experience and technical skills. These include programming-based approaches using Python, MATLAB, and other programming languages, as well as console-based tools.
Prizes will be awarded at JRC2027 to the top three teams, as well as to three additional teams whose work demonstrates outstanding technical originality and innovation.
２．Eligibility and Participation Requirements
Participants must apply as a team, and each team must include at least one member of one of the following academic societies: Japanese Society of Radiological Technology (JSRT), Japan Radiological Society (JRS), or Japanese Society of Medical Physics (JSMP). There is no limit on the number of members per team; individual participation (a team of one) is also permitted. However, an individual may not participate in more than one team.
A representative (one person per team) from each team selected for an award, as described in Section 4, will be required to give an oral presentation at the Award Ceremony and Presentation Session at JRC2027, scheduled for April 17, 2027, from 10:20 to 11:50 (tentative).
Please note that registration fees for JRC2027, travel expenses, and other related costs must be covered by the participants themselves. Further details will be provided separately to the selected award recipients.
３．Challenge Task and Evaluation
＜Challenge Task＞
Localization of anatomical landmarks using CT scout images
＜Training Data＞
- The training images will consist of virtual Digital Reconstruction Radiogrpahs (vDRRs) generated by our committee from axial CT images in the publicly available TotalSegmentator dataset provided by University Hospital Basel [1,2]. Each vDRR is generated from either the frontal direction (coronal projection) or the lateral direction (lateral projection), with the projection direction varying from case to case.
- The relative cranio-caudal positions of the lung apex and lung base will be used as the ground-truth labels. The relative position is represented by a numerical value ranging from 0 to 1, with 0 corresponding to the cranial direction and 1 corresponding to the caudal direction. Based on the segmentation labels included in the TotalSegmentator dataset, our committee determines the positions of the lung apex and lung base and calculates their relative coordinates.
- In addition to the training data, including the vDRRs and their ground-truth labels, the Python code used to generate the vDRRs will be provided in advance.
＜Evaluation Data＞
- The evaluation images will consist of vDRRs generated by our committee from axial CT images obtained from multiple publicly available databases, including the TotalSegmentator dataset. The vDRRs will be generated using the Python code provided to participants, and the projection direction will vary from case to case.
- The ground-truth labels will be determined according to the same rules used for the training data. For cases obtained from databases other than the TotalSegmentator dataset, our committee will independently determine the positions of the lung apex and lung base and calculate the corresponding ground-truth labels.
- The evaluation vDRRs are scheduled to be released approximately two weeks before the submission deadline (around January 15, 2027). To ensure fairness, the sources and details of the evaluation data will not be disclosed until the results are announced.
＜Evaluation Method＞
- Participants will submit their predicted relative coordinates for the evaluation data in CSV format. The committee will calculate the errors between the predicted values and the ground-truth labels, including mean absolute error (MAE) and root mean square error (RMSE), as well as the Pearson correlation coefficient. The final ranking will be determined based on these results.
- If two or more teams receive the same score, the committee may take into consideration the methods and technical innovations described by the participants in a separate Word document when determining the final ranking.
４．Prizes and Awards
The top three teams will receive prizes as follows:
- 1st price：JPY 100,000
- 2nd price：JPY 50,000
- 3rd price：JPY 40,000
In addition to the top three teams, additional three teams whose work demonstrates outstanding technical originality and ingenuity will receive a Special Award from the Committee, with a prize of JPY 20,000 per team.
One representative from each of the top three teams and the three Special Award-winning teams will be asked to give a presentation on their methods and technical innovations at the Award Ceremony and Presentation Session at JRC2027.
５．Submission Materials
Participants are required to submit the following materials to the cloud storage designated by the Society:
- A CSV file listing the relative coordinate positions of the lung apex and lung base for each evaluation DRR image (specified format).
A Word document describing an overview of the method and the technical innovations or approaches adopted (specified format).
６．Entry Procedure
Please complete the required information on the entry form and submit your application to participate in the challenge.
https://forms.gle/mkSiLtFzpce9QKos5
Registered participants will be provided with the training data, relevant documents, and other materials (scheduled for early November 2026).
７．Entry Deadline
December 18, 2026 (Friday), 17:00 JST
８．Submission Deadline for Evaluation Results
January 29, 2027 (Friday), 17:00 JST
For insuiries：jsrt@info.email.ne.jp
References
[1] Dataset with segmentations of 117 important anatomical structures in 1228 CT images: https://zenodo.org/records/10047292
[2] Wasserthal J, Breit HC, Meyer MT, Pradella M, Hinck D, Sauter AW, Heye T, Boll DT, Cyriac J, Yang S, Bach M, Segeroth M. TotalSegmentator: Robust Segmentation of 104 Anatomic Structures in CT Images. Radiol Artif Intell. 2023 Jul 5;5(5):e230024. doi: 10.1148/ryai.230024.
