Project Details
Projekt
DEvelopment and dEployment of a Pipeline for automated LymphoNodal profiling and staging: DEEP-LN
Subject Area
Nuclear Medicine, Radiotherapy, Radiobiology
Term
since 2019
Project identifier
Deutsche Forschungsgemeinschaft (DFG) - Project number 428219815
In oncology, the correct determination of nodal involvement is essential for patient management and outcome. Traditionally, the determination of total tumor burden and nodal involvement at baseline and in follow-up examinations for treatment response evaluation is based on morphological size criteria only. This, however, has several limitations, such as the limited specificity of N-staging in many solid cancer types or the sheer number of affected lymph nodes (LN) in oncologic diseases of the lymphatic system, making it highly unfeasible to manually segment and analyze all possibly affected LN in a routine clinical setting. Thus, solutions based on artificial intelligence (AI) including radiomics are highly warranted to enable automated LN detection, segmentation, and characterization. In phase I of the SPP 2177, we thus focused on the optimization of AI-based algorithms for fully-automated detection and segmentation of LN for large-scale, single-center datasets of the neck, chest, and abdomen, and applied radiomics for advanced LN characterization. We trained and optimized various segmentation algorithms and radiomics models by generating a dataset of 49.711 3D-segmented LN in total till now. In addition, we developed the open-source framework “AutoRadiomics” with an interactive user interface to enable a reproducible radiomics workflow even for non-programmers. For the next SPP phase II, we plan to further advance our methodology to close the gap toward clinical translation. We will do so by 1) distributing our trained models within the collaborating centers of the SPP for model extension and validation in a large-scale multi-center setting, 2) developing novel methodological approaches to radiomics modeling to address the current limitations of many radiomics models, 3) developing and deploying a full pipeline for automated lymphonodal profiling and staging (DEEP-LN) using the joint imaging platform (JIP) of the RACOON consortium, and 4) building a central AI and radiomics imaging group within the SPP for investigation of any types of LN diseases and extending this network internationally by setting up a data challenge for automatic detection and segmentation of LN.
DFG Programme
Priority Programmes
