Deep learning reconstruction accelerated reduced field-of-view DWI in rectal cancer: mucosa-submucosa-muscularis visualization and T staging.
Authors
Affiliations (3)
Affiliations (3)
- Department of Radiology, National Cancer Center/National Clinical Research Center for Cancer/Cancer Hospital, Chinese Academy of Medical Sciences and Peking Union Medical College, Beijing, China.
- GE Healthcare, Beijing, China.
- Department of Radiology, National Cancer Center/National Clinical Research Center for Cancer/Cancer Hospital, Chinese Academy of Medical Sciences and Peking Union Medical College, Beijing, China. [email protected].
Abstract
We compared the image quality and diagnostic performance of deep learning reconstruction (DLR) accelerated reduced field-of-view (rFOV<sub>DL</sub>) diffusion-weighted imaging (DWI) with standard-reconstructed full field-of-view (fFOV<sub>STA</sub>) DWI in rectal cancer. This prospective study enrolled 173 participants with biopsy-confirmed rectal adenocarcinoma from November 2022 to August 2023 undergoing rFOV<sub>DL</sub> and fFOV<sub>STA</sub> DWI scans. Two radiologists evaluated qualitative image quality, objective image quality, and apparent diffusion coefficient (ADC) independently. T and N staging were evaluated in 94 participants undergoing radical surgery. Diagnostic sensitivity, specificity, and accuracy were calculated using histopathologic results as the gold standard. ADC values were analyzed for correlations with histopathologic staging. We observed that rFOV<sub>DL</sub> DWI reduced acquisition time by 30% compared to fFOV<sub>STA</sub> DWI. rFOV<sub>DL</sub> DWI outperformed fFOV<sub>STA</sub> DWI in all qualitative image quality metrics (p ≤ 0.013), especially in mucosa-submucosa-muscularis visualization, spatial resolution, overall image quality, and diagnostic confidence, accompanied by comparable objective image quality (p ≥ 0.054). When applied with T2-weighted imaging, rFOV<sub>DL</sub> DWI significantly enhanced primary T-staging accuracy than fFOV<sub>STA</sub> DWI (p < 0.001), especially for early-stage tumors (T1 or T2). Tumor ADC values of rFOV<sub>DL</sub> DWI were lower than those of fFOV<sub>STA</sub> DWI, yet remained solid inverse correlations with histopathologic T-staging (p < 0.001). Higher inter-reader agreements of locoregional staging and ADC measurements were obtained by rFOV<sub>DL</sub> DWI. rFOV<sub>DL</sub> DWI significantly improved image quality than fFOV<sub>STA</sub> DWI, with a 30% reduced acquisition time. rFOV<sub>DL</sub> DWI facilitated higher primary T-staging accuracy, especially for early-stage rectal cancer (T1-T2). Reduced acquisition time and improved imaging quality highlighted the clinical feasibility of applying DLR to rFOV DWI. rFOV<sub>DL</sub> DWI could significantly enhance primary T-staging accuracy, especially for early-stage rectal cancer (T1-T2), facilitating more precise treatment management. Applying deep learning reconstruction (DLR) to reduced field-of-view (rFOV) diffusion-weighted imaging (DWI) improved mucosa-submucosa-muscularis visualization and reduced acquisition time. DLR-based rFOV DWI significantly enhanced primary T-staging accuracy for rectal cancer, especially for early-stage tumors (T1 or T2). DLR-based rFOV DWI facilitated higher inter-reader agreements for locoregional staging and apparent diffusion coefficient measurement in rectal cancer.