ESTRO 2024 - Abstract Book

S4962

Physics - Radiomics, functional and biological imaging and outcome prediction

ESTRO 2024

of 0.59). When inputting CT and PET (early fusion) without the GTV, DensnetNet81 obtained a much higher C-index in the external test set of 0.63 and a comparable C-index of 0.69 in the internal test set, which is comparable with 0.68 achieved by the winner team [7] in HECKTOR 2022. Apparently adding the GTV contour resulted in a decreased prediction performance in the external test set. This may be caused by differences in the tumor delineation between the two datasets. Additionally, we found that PET-only DenseNet81 obtained a high C-index of 0.67 and 0.65 in both internal and external test sets. DenseNet81 using late fusion of CT and PET achieved a similar C-index of 0.68 as PET only DenseNet81 in the internal test and the best performance of all tested models in the external test set with a C index of 0.66.

Conclusion:

The basic DenseNet architecture with 81 layers could achieve comparable prediction performance as state-of-art models with more complex architectures in the internal test set, and better performance in the external test. Late fusion of CT and PET imaging data resulted in better performance in the external test and adding the GTV contour resulted in comparable performance in the internal test and even decreased performance in the external test set.

Keywords: DenseNet; outcome prediction; oropharngeal cancer

References:

[1] Tumor Segmentation and Outcome Prediction in PET/CT. Head Neck Tumor Segmentation Outcome Predict 2023. [2] Huang G, Liu Z, Van Der Maaten L, Weinberger KQ. Densely connected convolutional networks. Proc. IEEE Conf. Comput. Vis. pattern Recognit., 2017, p. 4700–8. [3] Ma B, Guo J, Van Dijk L, van Ooijen PMA, Both S, Sijtsema NM. TransRP: Transformer-based PET/CT feature extraction incorporating clinical data for recurrence-free survival prediction in oropharyngeal cancer. Med. Imaging with Deep Learn., 2023. [4] Meng M, Bi L, Fulham M, Feng D, Kim J. Merging-Diverging Hybrid Transformer Networks for Survival Prediction in Head and Neck Cancer. ArXiv Prepr ArXiv230703427 2023. [5] Liu Z, Lin Y, Cao Y, Hu H, Wei Y, Zhang Z, et al. Swin transformer: Hierarchical vision transformer using shifted windows. Proc. IEEE/CVF Int. Conf. Comput. Vis., 2021, p. 10012–22. [6] Meng M, Gu B, Fulham M, Song S, Feng D, Bi L, et al. DeepMSS: Deep Multi-Modality Segmentation-to-Survival Learning for Survival Outcome Prediction from PET/CT Images. ArXiv Prepr ArXiv230509946 2023. [7] Rebaud L, Escobar T, Khalid F, Girum K, Buvat I. Simplicity is all you need: out-of-the-box nnUNet followed by binary-weighted radiomic model for segmentation and outcome prediction in head and neck PET/CT. 3D Head Neck Tumor Segmentation PET/CT Chall., Springer; 2022, p. 121–34.

284

Digital Poster

Time-dependent diffusion MRI for tumor characterization in biologically adapted radiotherapy