Diagnostic value of retrospective PET-MRI fusion in head-and-neck cancer

Background: To assess the diagnostic value of retrospective PET-MRI fusion and to compare the results with side-by-side analysis and single modality use of PET and of MRI alone for locoregional tumour and nodal staging of head-and-neck cancer. Methods: Thirty-three patients with head-and-neck cancer underwent preoperative contrast-enhanced MRI and PET/CT for staging. The diagnostic data of MRI, PET, side-by-side analysis of MRI and PET images and retrospective PET-MRI fusion were systematically analysed for tumour and lymph node staging using receiver operating characteristic (ROC) analysis. The results were correlated to the histopathological evaluation. Results: The overall sensitivity/specificity for tumour staging for MRI, PET, side-by-side analysis and retrospective PET-MRI fusion was 79%/66%, 82%/100%, 86%/100% and 89%/100%, respectively. The overall sensitivity/specificity for nodal staging on a patient basis for MRI, PET, side-by-side analysis and PET-MRI fusion was 94%/64%, 94%/91%, 94%/82% and 94%/82%, respectively. MRI, PET, side-by-side analysis and retrospective image fusion were associated with correct diagnosis/over-staging/under-staging of N-staging in 70.4%/18.5%/11.1%, 81.5%/7.4%/11.1%, 81.5%/11.1%/7.4% and 81.5%/11.1%/7.4%, respectively. ROC analysis showed no significant differences in tumor detection between the investigated methods. The Area Under the Curve (AUC) for MRI, PET, side-by-side analysis and retrospective PET-MRI fusion were 0.667/0.667/0.702/0.708 (p > 0.05). The most reliable technique in detection of cervical lymph node metastases was PET imaging (AUC: 0.95), followed by side-by-side analysis and retrospective image fusion technique (AUC: 0.941), which however, was not significantly better then the MRI (AUC 0.935; p > 0.05). Conclusions: We found a beneficial use of multimodal imaging, compared with MRI or PET imaging alone, particular in individual cases of recurrent tumour disease. Side-by-side analysis and retrospective image fusion analysis did not perform significantly differently.