Abstract
Purpose: To develop and evaluate a framework for aligning diffusion‐weighted MR images (DWI) to CT and PET/CT in locally‐advanced lung cancer, and to evaluate the correspondence of FDG‐PET and DWI prior to radiation therapy. Methods: Pretreatment FDG‐PET/CT, 4D fan beam planning CT, and MR were acquired in five locally‐advanced lung cancer patients. Deformable registration was used to align the b=0 s/m2 DWI to a high‐resolution morphological MRI to reduce spatial distortion and respiratory effects. The morphological MRI was then registered to planning CT to map DWI to the planning CT reference frame. Apparent diffusion coefficient (ADC) was computed from DWI. PET/CT was also registered to planning CT. 40% of maximum lesion SUV (SUVmax) threshold was used to delineate primary tumor on PET, and a physician delineated the lesion on DWI. Dice coefficient between PET and DWI primary tumor volumes was computed to assess whole lesion location. To establish correspondence of tumor subvolumes, the region of 70% SUVmax on PET (high SUV) was delineated, and the ADC distribution from the spatially‐registered DWI extracted in this region. This process was repeated for the 40%–50% SUVmax (low SUV) region. Results: All registrations were judged visually acceptable. The PET and DWI‐defined primary tumor Dice coefficient was 0.6 +/‐0.1 after registration. The mean ADC corresponding to the high SUV region was lower than the mean ADC in the low SUV region in 4/5 patients (mean difference of −130 m2/s). One patient had higher ADC in the high SUV region, with a difference of 94 m2/s between regions. This patient had a highly heterogeneous lesion with small hyperintense pockets on ADC images indicating necrosis. Conclusion: Deformable registration of thoracic MR appears feasible to reduce the effects of distortion and motion artifacts. High SUV generally corresponded with the most restricted diffusion signal from DWI.
