Convolutional neural networks for PET functional volume fully automatic segmentation: development and validation in a multi-center setting
Por:
Iantsen, A, Ferreira, M, Lucia, F, Jaouen, V, Reinhold, C, Bonaffini, P, Alfieri, J, Rovira, R, Masson, I, Robin, P, Mervoyer, A, Rousseau, C, Kridelka, F, Decuypere, M, Lovinfosse, P, Pradier, O, Hustinx, R, Schick, U, Visvikis, D, Hatt, M
Publicada:
1 oct 2021
Ahead of Print:
1 mar 2021
Resumen:
Purpose In this work, we addressed fully automatic determination of tumor functional uptake from positron emission tomography (PET) images without relying on other image modalities or additional prior constraints, in the context of multicenter images with heterogeneous characteristics. Methods In cervical cancer, an additional challenge is the location of the tumor uptake near or even stuck to the bladder. PET datasets of 232 patients from five institutions were exploited. To avoid unreliable manual delineations, the ground truth was generated with a semi-automated approach: a volume containing the tumor and excluding the bladder was first manually determined, then a well-validated, semi-automated approach relying on the Fuzzy locally Adaptive Bayesian (FLAB) algorithm was applied to generate the ground truth. Our model built on the U-Net architecture incorporates residual blocks with concurrent spatial squeeze and excitation modules, as well as learnable non-linear downsampling and upsampling blocks. Experiments relied on cross-validation (four institutions for training and validation, and the fifth for testing). Results The model achieved good Dice similarity coefficient (DSC) with little variability across institutions (0.80 +/- 0.03), with higher recall (0.90 +/- 0.05) than precision (0.75 +/- 0.05) and improved results over the standard U-Net (DSC 0.77 +/- 0.05, recall 0.87 +/- 0.02, precision 0.74 +/- 0.08). Both vastly outperformed a fixed threshold at 40% of SUVmax (DSC 0.33 +/- 0.15, recall 0.52 +/- 0.17, precision 0.30 +/- 0.16). In all cases, the model could determine the tumor uptake without including the bladder. Neither shape priors nor anatomical information was required to achieve efficient training. Conclusion The proposed method could facilitate the deployment of a fully automated radiomics pipeline in such a challenging multicenter context.
Filiaciones:
Iantsen, A:
Univ Brest, LaTIM, INSERM, UMR 1101, Brest, France
Ferreira, M:
Univ Liege, GIGA CRC Vivo Imaging, Liege, Belgium
Lucia, F:
Univ Brest, LaTIM, INSERM, UMR 1101, Brest, France
Jaouen, V:
Univ Brest, LaTIM, INSERM, UMR 1101, Brest, France
Reinhold, C:
McGill Univ, Hlth Ctr MUHC, Dept Radiol, Montreal, PQ, Canada
Bonaffini, P:
McGill Univ, Hlth Ctr MUHC, Dept Radiol, Montreal, PQ, Canada
Alfieri, J:
McGill Univ, Hlth Ctr MUHC, Dept Radiat Oncol, Montreal, PQ, Canada
Rovira, R:
Hosp Stanta Creu & St Pau, Gynecol Oncol & Laparoscopy Dept, Barcelona, Spain
Masson, I:
Inst Cancerol Ouest ICO, Dept Radiat Oncol, Nantes, France
Robin, P:
Univ Hosp, Nucl Med Dept, Brest, France
Mervoyer, A:
Inst Cancerol Ouest ICO, Dept Radiat Oncol, Nantes, France
Rousseau, C:
Inst Cancerol Ouest ICO, Nucl Med Dept, Nantes, France
Kridelka, F:
Univ Hosp Liege, Div Oncol Gynecol, Liege, Belgium
Decuypere, M:
Univ Hosp Liege, Div Oncol Gynecol, Liege, Belgium
Lovinfosse, P:
Univ Hosp Liege, Div Nucl Med & Oncol Imaging, Liege, Belgium
Pradier, O:
Univ Brest, LaTIM, INSERM, UMR 1101, Brest, France
Hustinx, R:
Univ Liege, GIGA CRC Vivo Imaging, Liege, Belgium
Schick, U:
Univ Brest, LaTIM, INSERM, UMR 1101, Brest, France
Visvikis, D:
Univ Brest, LaTIM, INSERM, UMR 1101, Brest, France
Hatt, M:
Univ Brest, LaTIM, INSERM, UMR 1101, Brest, France
Green Published, hybrid
|