[en] Complete text of publication follows: Purpose: PET tumour segmentation is currently an active research topic in the field of radiotherapy planning and multi-parametric data quantification. Although being efficient on homogeneous spheroid-shaped tumours, classical threshold-based approaches are of limited value for heterogeneous or complex-shaped tumours, still making expert-based manual delineation the reference standard for tumour imaging, despite several limitations. In this context, the aim of this study was to assess the performances of an active contour-based approach for the PET segmentation of complex-shaped lung tumours, in comparison to an optimized expert-based manual reference standard. Subjects and Methods: Seventy-five thoracic tumours were segmented using the same graphical user interface (GUI) ITK snap software. For each tumour, an optimized expert-based reference standard was generated from the set of six independent expert-based manual segmentation results using the Simultaneous Truth And Performance Level Estimate (STAPLE) algorithm. In addition, four raters semi-automatically segmented the 75 PET tumours twice using the active contour based-procedure of the GUI software, with a delay time of one week between two segmentation sessions. For the 75 tumours, accuracy of the semi-automatic segmentations against the optimized expert-based reference standard was assessed using the DICE similarity coefficient (DSC). Inter-rater and intra-rater reliability analyses were performed using the intra class correlation coefficients (ICC) estimates of the output volumes, along with their 95% confidence intervals (two-way mixed-model, individual-rating, absolute-agreement). For all tumours segmentation procedures, average time per procedure was also estimated. Results: Overall accuracy of the semi-automatic procedure was excellent with a DSC of 0.835 (95%CI = 0.775-0.895). Inter-rater reliabilities provided the following results: ICC = 0.941 (95%CI = 0.913-0.961) for the first session and ICC = 0.935 (95%CI = 0.906- 0.956) for the second session. Intra-rater reliabilities provided the following results: ICC = 0.993 (95%CI = 0.990-0.996) for the rater 1; ICC = 0.987 (95%CI = 0.976-0.993) for the rater 2; ICC = 0.972 (95%CI = 0.956-0.982) for the rater 3; and ICC = 0.977 (95%CI = 0.964-0.985) for the rater 4. Average time was 631 seconds for manual segmentation procedure and 130 seconds for active contour-based. Conclusions: Compared to the state of the art expert-based manual segmentation, the GUI-based active contour procedure provided excellent accuracy and reliability, with a mean procedure duration almost five-times faster than the manual reference procedure. ITK snap software is robust, fast, and easy enough to be routinely applied as a powerful alternative to the manual reference standard in this setting

[en] Complete text of publication follows: Aim/Introduction: To evaluate the impact of geometric distortions inherent to diffusion weighted imaging (DWI) in the field of PET-MRI lung oncology. Materials and Methods: 10 patients were prospectively recruited and underwent an ¹⁸F-FDG PET-MRI for lung oncology purpose. For all patients, the imaging protocol included several thoracic PET-MRI acquisitions: a PET acquisition performed one hour after the intra veinous injection of ¹⁸F-FDG; a set of DWI acquisitions with anteroposterior phase encoding (b values of 0,500, 800 s/mm²) and reverse-phase encoding polarity (b value = 0 s/mm²); a high resolution post contrast-enhanced 3DT1-weighted FSPGR sequence. DWI data were corrected for geometric distortions using the reverse phase encoding method. All the DWI data (non corrected and corrected from the distortion) were warped to the same T1 weighted PET-MRI isotropic reference space before analyses. Quality of the co-registrations to the reference T1 PET-MRI was quantitatively assessed using mutual information metric. The percentage gain compared to non-warped DWI data was also computed. ADC feature maps of each lung lesion were computed from the DWI data non-corrected and corrected from the distortion, and voxel-wise percent differences together with paired t-test were computed. Finally, regional ADC-SUV monotonic correlations were explored from optimal realigned DWI-PET data. Results: Quality of the co-registration between DWI and T1 PET-MRI data was significantly improved by the reverse phase encoding method (relative gain on mutual information compared to non-warped DWI data : 4-46%, vs 0.4-27% for warped data without distortion correction, p inferior to 0.05). The between-ADC feature maps regional differences ranged from -100% to more than +150%. Regional correlations between ADC and SUV computed from optimal realigned DWIPET data revealed only weak monotonic relationships between the two features at the voxel level (Spearman coefficients inferior to 0.5 in all the cases). Conclusion: DWI-related distortions are significant in thoracic PET-MRI, and should be corrected for accurate DWI-PET multimodal analyses. ADC and SUV showed weak monotonic relationship at the voxel level, emphasizing their complementarity