Dataset Information

Robustness and Generalizability of Deep Learning Synthetic Computed Tomography for Positron Emission Tomography/Magnetic Resonance Imaging-Based Radiation Therapy Planning of Patients With Head and Neck Cancer.

ABSTRACT:

Purpose

Radiotherapy planning based only on positron emission tomography/magnetic resonance imaging (PET/MRI) lacks computed tomography (CT) information required for dose calculations. In this study, a previously developed deep learning model for creating synthetic CT (sCT) from MRI in patients with head and neck cancer was evaluated in 2 scenarios: (1) using an independent external dataset, and (2) using a local dataset after an update of the model related to scanner software-induced changes to the input MRI.

Methods and materials

Six patients from an external site and 17 patients from a local cohort were analyzed separately. Each patient underwent a CT and a PET/MRI with a Dixon MRI sequence over either one (external) or 2 (local) bed positions. For the external cohort, a previously developed deep learning model for deriving sCT from Dixon MRI was directly applied. For the local cohort, we adapted the model for an upgraded MRI acquisition using transfer learning and evaluated it in a leave-one-out process. The sCT mean absolute error for each patient was assessed. Radiotherapy dose plans based on sCT and CT were compared by assessing relevant absorbed dose differences in target volumes and organs at risk.

Results

The MAEs were 78 ± 13 HU and 76 ± 12 HU for the external and local cohort, respectively. For the external cohort, absorbed dose differences in target volumes were within ± 2.3% and within ± 1% in 95% of the cases. Differences in organs at risk were <2%. Similar results were obtained for the local cohort.

Conclusions

We have demonstrated a robust performance of a deep learning model for deriving sCT from MRI when applied to an independent external dataset. We updated the model to accommodate a larger axial field of view and software-induced changes to the input MRI. In both scenarios dose calculations based on sCT were similar to those of CT suggesting a robust and reliable method.

SUBMITTER: Olin AB

PROVIDER: S-EPMC8452789 | biostudies-literature |

REPOSITORIES: biostudies-literature

ACCESS DATA

Similar Datasets

Project description:PurposeTo investigate interobserver delineation variability for gross tumor volumes of primary lung tumors and associated pathologic lymph nodes using magnetic resonance imaging (MRI), and to compare the results with computed tomography (CT) alone- and positron emission tomography (PET)-CT-based delineations.Methods and materialsSeven physicians delineated the tumor volumes of 10 patients for the following scenarios: (1) CT only, (2) PET-CT fusion images registered to CT ("clinical standard"), and (3) postcontrast T1-weighted MRI registered with diffusion-weighted MRI. To compute interobserver variability, the median surface was generated from all observers' contours and used as the reference surface. A physician labeled the interface types (tumor to lung, atelectasis (collapsed lung), hilum, mediastinum, or chest wall) on the median surface. Contoured volumes and bidirectional local distances between individual observers' contours and the reference contour were analyzed.ResultsComputed tomography- and MRI-based tumor volumes normalized relative to PET-CT-based volumes were 1.62 ± 0.76 (mean ± standard deviation) and 1.38 ± 0.44, respectively. Volume differences between the imaging modalities were not significant. Between observers, the mean normalized volumes per patient averaged over all patients varied significantly by a factor of 1.6 (MRI) and 2.0 (CT and PET-CT) (P=4.10 × 10-5 to 3.82 × 10-9). The tumor-atelectasis interface had a significantly higher variability than other interfaces for all modalities combined (P=.0006). The interfaces with the smallest uncertainties were tumor-lung (on CT) and tumor-mediastinum (on PET-CT and MRI).ConclusionsAlthough MRI-based contouring showed overall larger variability than PET-CT, contouring variability depended on the interface type and was not significantly different between modalities, despite the limited observer experience with MRI. Multimodality imaging and combining different imaging characteristics might be the best approach to define the tumor volume most accurately.

Project description:Functional imaging modalities enable practitioners to identify functional lung regions. This analysis evaluated the feasibility of nuclear medicine imaging to avoid doses to the functional lung in radiotherapy (RT) planning for patients with lung cancer. This systematic review and meta-analysis was carried out according to PRISMA-P guidelines. A search of EMBASE and PubMed for studies published throughout the last 20 years was performed using the following search criteria: (a) 'lung cancer' or 'lung malignancy' and (b) 'radiotherapy' or 'radiation therapy' or 'RT planning' and (c) 'SPECT' or 'single positron emission computed tomography' or 'functional image.' The analyzed planning parameters were the volumes of the normal lung that have received ≥ 10 Gy and ≥ 20 Gy of radiation (V10 and V20, respectively) and the mean lung dose (MLD). We compared the planning parameters obtained from anatomical RT planning and functional RT planning using perfusion or ventilation imaging ('V10, V20 or MLD' in anatomical plan vs. 'fV10, fV20 or fMLD' in functional plan). A total of 309 patients with 344 RT plan sets from 15 publications (11 perfusion SPECT, 2 ventilation SPECT, and 1 SPECT and 1 PET with both perfusion and ventilation) were included in the meta-analysis. The standard mean differences in planning parameters in functional plans using nuclear imaging were significantly reduced compared to those of anatomical plans (P < 0.01 for all): - 0.42 (95% confidence interval (CI) - 0.78 to - 0.07) for 'V10 vs. fV10', - 0.41 (95% CI - 0.64 to - 0.17) for 'V20 vs. fV20', and - 0.24 (95% CI - 0.45 to - 0.03) for 'MLD vs. fMLD'. In subgroup analysis, the functional plan using perfusion was significantly lower than the anatomical plan in all planning parameters, but there was no significant difference for ventilation. RT planning with nuclear functional lung imaging has potential to reduce radiation-induced lung injury. Perfusion imaging seems to be more promising than ventilation imaging for all planning parameters. There were not enough studies using ventilation imaging to determine what the effect is on the lung plan parameters.