深度学习算法在深度学习重建加速前列腺双参数MRI中的表现
本研究回顾性纳入141例男性,评估基于常规双参数MRI训练的前列腺癌深度学习算法对深度学习重建加速MRI输入的稳健性。四种输入配置的诊断性能总体接近,但C-T2WI+DLR-DWI配置检出病灶数减少且假阳性降低。算法与重建的厂商特异性限制外推性。
要点
- 深度学习重建加速的前列腺双参数MRI可基本保持深度学习算法的诊断性能。
- C-T2WI+DLR-DWI组合减少检出病灶数和假阳性,提示重建可能影响算法敏感度。
- 阅片时需注意不同重建算法的图像纹理变化可能影响AI辅助检测结果。
核心结论
- 深度学习算法在常规与深度学习重建加速的前列腺双参数MRI输入间表现总体接近,但厂商特异性限制外推性,需进一步评估。
研究设计与样本
- 回顾性研究,纳入141例男性,中位年龄70.5岁,中位PSA 6.6 ng/mL。
- 同一检查内采集四种输入配置:C-T2WI+C-DWI(参考)、C-T2WI+DLR-DWI、DLR-T2WI+C-DWI、DLR-T2WI+DLR-DWI。
- 115例经组织病理或PSA随访≥2年确诊,临床显著前列腺癌定义为Gleason分级组≥2。
关键影像表现/征象
- C-T2WI+DLR-DWI配置检出病灶数较参考配置减少(99 vs. 131)。
诊断标准与关键数值
- C-T2WI+DLR-DWI配置FP/case低于参考(0.478 vs. 0.722;P=0.003),IoU较高(0.194 vs. 0.153;P=0.014)。
- 各配置间病灶级敏感度和阳性预测值差异相对较小。
临床应用与阅片要点
- 深度学习重建DWI可能导致AI检出病灶数量减少,临床报告中需结合原始图像复核。
- 不同输入配置间FP/case和IoU存在差异,提示重建算法可能改变AI的定位精度。
局限性
- 深度学习算法和深度学习重建均为厂商特定,结果可推广性受限。
- 研究为初步评估,需进一步验证。
以上解读依据论文摘要(全文未开放获取或许可不允许翻译转载)。
英文原文(摘要)· 点击展开
PURPOSE: To evaluate the robustness of a deep learning algorithm (DLA) for prostate lesion detection and classification trained on conventional biparametric MRI (bpMRI) to changes in imaging inputs introduced by deep learning reconstruction (DLR)-accelerated MRI.
MATERIALS AND METHODS: This retrospective study included 141 men (median age 70.5 years; median prostate-specific antigen [PSA] level 6.6 ng/mL). A DLA for prostate cancer (PCa) detection and classification was evaluated using four bpMRI input configurations acquired within the same examination: conventional T2-weighted imaging (C-T2WI) plus conventional diffusion-weighted imaging (C-DWI; reference), C-T2WI + DLR-DWI, DLR-T2WI + C-DWI, and DLR-T2WI + DLR-DWI. A definitive diagnosis was established in 115 patients using histopathology or ≥2 years of PSA follow-up; clinically significant PCa was defined as a Gleason grade group ≥2. Performance was assessed at the patient and lesion levels using diagnostic metrics, DLA level of suspicion (LoS), false positives per case (FP/case), and intersection over union (IoU).
RESULTS: Per-patient diagnostic performance estimates were generally close across the input configurations, although the magnitude of the differences varied across metrics. With C-T2WI + DLR-DWI, the number of detected lesions decreased compared with the reference configuration (99 vs. 131), accompanied by lowerFP/case (0.478 vs. 0.722; P = 0.003) and greater IoU (0.194 vs. 0.153; P = 0.014). Differences in lesion-level sensitivity and positive predictive value were relatively small across configurations.
CONCLUSION: The evaluated DLA showed generally close performance estimates across conventional and DLR-accelerated bpMRI inputs. However, the vendor-specific nature of both the DLA and DLR limits generalizability and warrants further evaluation.
引用
- Lee, Choi, Grimm, et al. Performance of a deep learning algorithm on deep learning reconstruction-accelerated prostate biparametric MRI: A preliminary study. Eur J Radiol. 2026. PMID: 42736178. doi:10.1016/j.ejrad.2026.113227
供执业医师工作参考,不替代临床判断;数值与分级以指南原文为准。