Grand Challenges on the development of AI algorithms for lung cancer screening.
Deep learning (DL) has emerged as a helpful tool to contribute to reducing radiologist’ workload and assist radiologists in their work. Many DL systems have already been proposed for a variety of classification, detection and segmentation tasks in medical imaging. These DL systems can be designed for specific tasks (narrow AI) or designed as foundational models capable of performing multiple tasks across different domains.
Despite the advancements, many DL models lack external validation and comparison with the performance of human experts. The platform www.grand-challenge.org offers a solution by enabling organizers to host challenges in medical image analysis. These challenges provide a method for participants to test, benchmark, and compare their algorithms against other submissions and the performance of human readers.
Our project aims to organize community-wide, publicly accessible challenges to advance the development of innovative DL methods. We propose to host at least two challenges:
1. UNICORN: This challenge will provide a unified set of benchmarks to assess the performance of foundation models in pathology and radiology.
2. LUNA24: This challenge will focus on providing a benchmark for DL models for pulmonary nodules malignancy risk estimation in low-dose chest CT.
Participants can utilize the NLST data to develop and submit their algorithms, allowing for a direct comparison of different DL architectures, training methodologies and human performance. These challenges help build a collaborative environment where DL solutions can be tested and refined, ultimately contributing to the improvement of diagnostic tools in medical image analysis and lung cancer screening specifically.
1. UNICORN: to be used in a potential AI challenge for the MICCAI 2025 conference, which is the Unified beNchmarks for Imaging in Computational pathology, Radiology and Natural language (UNICORN) challenge. This data will be used by participants to train their submitted AI models to perform a variety of classification, detection, segmentation, regression tasks in medical imaging, including the tasks of nodule detection, malignancy estimation, and incidental finding classification and detection.
2. LUNA24: we aim to provide a large-scale evaluation by benchmarking state-of-the-art DL algorithms for pulmonary nodule malignancy risk estimation. This allows researchers to publicly compare their algorithms with other submissions.
Dré Peeters, Co-Investigator, Department of Medical Imaging, Radboud University Medical Center Nijmegen, Netherlands
Bogdan Obreja, Co-Investigator, Department of Medical Imaging, Radboud University Medical Center Nijmegen, Netherlands
Fennie van der Graaf, Co-Investigator, Department of Medical Imaging, Radboud University Medical Center Nijmegen, Netherlands
Lena Philipp, Co-Investigator, Department of Medical Imaging, Radboud University Medical Center Nijmegen, Netherlands
Michel Vitale, Co-Investigator, Department of Medical Imaging, Radboud University Medical Center Nijmegen, Netherlands
Colin Jacobs, PI, Department of Medical Imaging, Radboud University Medical Center Nijmegen, Netherlands
Mathias Prokop, Department of Medical Imaging, Radboud University Medical Center Nijmegen, Netherlands