文献速递肿瘤 PET/CT

深度学习软件与资深核医学医师判读前列腺癌首次生化复发患者[18F]-匹氟索司他PSMA-PET/CT的检出率比较研究

Deep-learning-based software versus experienced nuclear medicine physician to read [18F]-piflufolastat used for PSMA-PET/CT imaging in prostate cancer patients presenting with first biochemical recurrence: A comparative study of detection rate.
PubMed 文献 · Eur J Nucl Med Mol Imaging · 2025-06-17

本研究基于III期PYTHON研究数据回顾性分析192例可评估扫描,比较CE认证深度学习网络平台(PSMA-DLwp)与资深核医学医师(ENMP)判读[18F]-匹氟索司他PET/CT的检出率。结果显示每患者检出率一致率为75%,两者无统计学显著差异(p=0.11),SUVmax与肿瘤负荷强相关,8%的不一致病例中PSMA-DLwp改善了ENMP的初始评估。

要点

  • 每患者检出率一致率75%,ENMP与PSMA-DLwp无统计学显著差异(p=0.11)
  • T与N分期区域分析检出率无显著差异(p=0.72和p=0.30)
  • SUVmax(⍴s=0.85;p<0.0001)与肿瘤负荷(⍴s=0.89;p<0.0001)强正相关

核心结论

  • PSMA-DLwp达到与专家判读者相当的检出率,并可靠量化肿瘤负荷与SUVmax,支持其作为[18F]-匹氟索司他PET/CT判读的辅助工具。
  • 将其整合入临床工作流程可提升诊断一致性与效率,尤其在高通量或复杂病例中,从而支持更明智的临床决策。

研究设计与样本

  • 回顾性分析,使用III期PYTHON研究数据,该研究比较[18F]-匹氟索司他PET/CT与[18F]-氟甲基胆碱PET/CT在前列腺癌首次生化复发患者中的应用。
  • ENMP与PSMA-DLwp独立判读[18F]-匹氟索司他PET/CT扫描,评估检出率一致性、区域分析及半定量参数(SUVmax、肿瘤负荷)相关性;不一致病例由ENMP重新评估以判断PSMA-DLwp的潜在改进。

主要发现

  • 192例可评估扫描中,每患者检出率一致率为75%,ENMP与PSMA-DLwp之间无统计学显著差异(p=0.11)。
  • T与N分期区域分析显示检出率无显著差异(p=0.72和p=0.30)。
  • PSMA-DLwp在远隔区域(M1a、M1b)识别出更多病灶,但无统计学显著差异。
  • ENMP与PSMA-DLwp之间SUVmax(⍴s=0.85;p<0.0001)与肿瘤负荷(⍴s=0.89;p<0.0001)呈强正相关。
  • 8%的不一致病例中,PSMA-DLwp改善了ENMP的初始评估。

临床怎么用

  • 作者认为PSMA-DLwp可作为[18F]-匹氟索司他PET/CT判读的辅助工具,支持其整合入临床工作流程以提升诊断一致性与效率。

局限性

  • 原文未报告局限性。

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)。

原文

This study evaluated whether a CE-marked deep-learning-based web platform (PSMA-DLwp) maintains a per-patient detection rate (DR) comparable to that of an experienced nuclear medicine physician (ENMP) for identifying prostate cancer (PCa) foci using [18F]-piflufolastat PET/CT in patients with first biochemical recurrence. Secondary objectives included per-region DR and the correlation of SUVmax and tumor burden assessments between the ENMP and PSMA-DLwp. Methods: A retrospective analysis was conducted using data from the phase III PYTHON study, which compared [18F]-piflufolastat PET/CT to [18F]-fluoromethylcholine PET/CT in patients with first biochemical recurrence of PCa. Both ENMP and PSMA-DLwp independently reviewed [18F]-piflufolastat PET/CT scans. Concordance in DRs, regional analysis, and correlation of semi-quantitative parameters (SUVmax, tumor burden) were assessed. Discordant cases were re-evaluated by the ENMP to assess potential improvements provided by PSMA-DLwp. Results: Out of 192 evaluable scans, per-patient DRs were concordant in 75% of cases, with no statistically significant difference between ENMP and PSMA-DLwp (p = 0.11). Per-region analysis for T and N staging showed no significant DR differences (p = 0.72 and p = 0.30, respectively). PSMA-DLwp identified more foci in distant regions (M1a, M1b), but these were not statistically significant. Strong positive correlations were observed between ENMP and PSMA-DLwp for SUVmax (⍴s=0.85; p < 0.0001) and tumor burden (⍴s=0.89; p < 0.0001). In 8% of discordant cases, PSMA-DLwp improved the ENMP’s initial assessment. Conclusion: The PSMA-DLwp achieves DRs comparable to an expert reader and reliably quantifies tumor burden and SUVmax, supporting its use as an adjunct in the interpretation of [18F]-piflufolastat PET/CT for PCa. Its integration into clinical workflows can enhance diagnostic consistency and efficiency, particularly in high-volume or complex cases, thereby supporting more informed clinical decision-making.

来源:Eur J Nucl Med Mol Imaging,查看原文。版权归原作者所有。

引用

  1. Gauthé M, Balogova S. Deep-learning-based software versus experienced nuclear medicine physician to read [18F]-piflufolastat used for PSMA-PET/CT imaging in prostate cancer patients presenting with first biochemical recurrence: A comparative study of detection rate. Eur J Nucl Med Mol Imaging. 2025. PMID: 40526126. doi:10.1007/s00259-025-07404-x

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。