In a groundbreaking achievement, Mehdi Astaraki, a postdoctoral researcher in medical radiation physics at Stockholm University, secured the top position in a highly competitive international competition organized by the International Conference on Medical Image Computing and Computer Assisted Intervention (MICCAI). The accolade was awarded for his deep learning-based method aimed at precise segmentation of tumors, affected lymph nodes, and healthy tissues in computed tomography (CT) images, specifically focusing on head and neck (H&N) cancer.
Unprecedented success in the SegRap 2023 challenge
Astaraki’s victory was realized in the Segmentation of Organs-at-Risk and Gross Tumor Volume of Nasopharyngeal Carcinoma for Radiotherapy Planning (SegRap2023) challenge, a MICCAI-sponsored event. This challenge addressed the critical need for an automated tool to effectively delineate 54 healthy structures (Task 1) and 2 tumoral regions (Task 2) in CT images of the H&N region.
Out of the 395 global teams participating, Astaraki’s contribution emerged as a frontrunner, clinching the 1st place in the final evaluation phase for Task 2 and securing 2nd place in the second evaluation phase for both tasks. This success underscores the method’s robustness and efficiency in segmenting vital structures, crucial for the accurate planning of radiation therapy.
Global recognition in the medical imaging community
MICCAI, known as the most significant and competitive community for medical image analysis tasks, attracted participation from 395 teams worldwide, with China and the United States leading in registrations. Astaraki’s accomplishment signifies not only individual excellence but also positions Stockholm University among the top-tier institutions in the rapidly evolving field of medical imaging.
In the second phase of the competition, where algorithms were submitted for evaluation on 20 subjects, Astaraki’s team stood out with over 150 models submitted for Task 1 and more than 70 for Task 2. The final evaluation phase saw more than 30 models assessed for each task, where the team’s algorithm consistently demonstrated exceptional performance.
Future integration with research PACS for clinical application
Looking ahead, Astaraki and his collaborators plan to integrate the developed model with the Picture Archiving and Communication System (PACS), a crucial component in healthcare for securely storing and communicating medical images. This strategic move aims to validate the model’s performance against clinical datasets, paving the way for its application in radiation treatment planning.
Astaraki emphasized the significance of active contribution to open source and open science communities, emphasizing that such collaboration is essential in refining and advancing solutions for complex challenges like medical image analysis.
Presentation and recognition
The groundbreaking methodology developed by Astaraki and his team was presented on the 12th of October at the MICCAI conference in Vancouver, Canada. The recognition received at this esteemed platform not only highlights the excellence of the Stockholm University team but also underscores the global impact of their innovative approach.
Mehdi Astaraki is scheduled to present his work in a News and Views session during the upcoming spring, providing a detailed overview of the developed method and its implications for the field of medical image analysis.
Astaraki’s triumph in the SegRap2023 challenge marks a significant stride in the realm of medical imaging, where the fusion of deep learning and clinical application showcases the potential for transformative breakthroughs in cancer treatment planning. As the methodology undergoes further integration and validation, the prospects for enhanced precision and efficiency in radiotherapy planning for Head and Neck cancer patients appear promising.
