Dana Jérémy, McNabb Evan, Castro Juan, Al-Qanoobi Ibtisam, Omiya Yoshie, Ah-Lan Kenny, Fortier Véronique, Artho Giovanni, Reinhold Caroline, Gauvin Simon
McGill University, Department of Diagnostic Radiology, Montréal, Canada.
Research Institute of the McGill University Health Centre, Augmented Intelligence & Precision Health Laboratory (AIPHL), Montréal, Canada.
Res Diagn Interv Imaging. 2025 May 22;14:100059. doi: 10.1016/j.redii.2025.100059. eCollection 2025 Jun.
To assess the impact of a commercially available deep learning reconstruction (DLR) algorithm on qualitative and radiomics analyses in prostate MRI.
This single-centre retrospective cohort included 25 consecutive patients who underwent a prostate MRI (1.5 T) in 2022. T2-weighted (T2WI), diffusion-weighted (DWI; b = 50, 1000, extrapolated 2000 s/mm) and apparent diffusion coefficient (ADC) images were reconstructed using DLR and standard (non-DLR) techniques. The two sets were mixed and blind-reviewed independently by six radiologists. Images were qualitatively scored according to PI-QUAL score, overall image quality, diagnostic confidence, anatomical conspicuity, artifact, and noise. Transition and peripheral zones were segmented and radiomics features extracted from region-of-interests using Pyradiomics package. Qualitative criteria and radiomics were compared using a pairwise Wilcoxon signed-rank test.
PI-QUAL score was not significantly different ( = 0.32). Overall image quality was not significantly different ( = 0.21 on T2WI and 0.56 on DWI/ADC). Noise was lower on DLR images for T2WI ( < 0.01) and DWI/ADC ( = 0.04). Diagnostic confidence in excluding clinically significant cancer (PI-RADS ≥ 3) in the transition zone was lower with DLR images (p = 0.02). In the transition zone, 89/93 (96 %) of the radiomics features were significantly different between non-DLR and DLR images on T2WI, 68/93 (73 %) on DWI b-2000 s/mm, and 55/93 (59 %) on ADC images. In the peripheral zone, 91/93 (98 %) were significantly different on T2WI, 50/93 (54 %) on DWI b-2000 s/mm, and 70/93 (75 %) on ADC images.
Radiomics features were significantly different on DLR images which should encourage caution for clinical and research purposes. DLR algorithm decreases noise while preserving overall image quality.
评估一种商用深度学习重建（DLR）算法对前列腺MRI定性分析和影像组学分析的影响。
本单中心回顾性队列研究纳入了2022年连续接受前列腺MRI（1.5T）检查的25例患者。使用DLR和标准（非DLR）技术重建T2加权（T2WI）、扩散加权（DWI；b=50、1000、外推2000 s/mm²）和表观扩散系数（ADC）图像。将这两组图像混合后由6名放射科医生独立进行盲法评估。根据PI-QUAL评分、整体图像质量、诊断置信度、解剖结构清晰度、伪影和噪声对图像进行定性评分。对移行区和外周区进行分割，并使用Pyradiomics软件包从感兴趣区域提取影像组学特征。使用配对Wilcoxon符号秩检验比较定性标准和影像组学特征。
PI-QUAL评分无显著差异（P=0.32）。整体图像质量无显著差异（T2WI上P=0.21，DWI/ADC上P=0.56）。T2WI（P<0.01）和DWI/ADC（P=0.04）的DLR图像噪声更低。DLR图像对移行区排除临床显著性癌症（PI-RADS≥3）的诊断置信度较低（P=0.02）。在移行区，T2WI上非DLR和DLR图像之间93个影像组学特征中的89个（96%）有显著差异，DWI b值为2000 s/mm²时为68/93（73%），ADC图像上为55/93（59%）。在外周区，T2WI上91/93（98%）有显著差异，DWI b值为2000 s/mm²时为50/93（54%），ADC图像上为70/93（75%）。
DLR图像上的影像组学特征有显著差异，这在临床和研究中应谨慎对待。DLR算法在保留整体图像质量的同时降低了噪声。
