Guan Bo, Huang Cong, Wang Yalei, Zhang Jialong, Li Xiaowei, Hao Zongyao
Department of Urology, Fuyang People's Hospital of Anhui Medical University, Fuyang, China.
Department of Urology, the First Affiliated Hospital of Anhui Medical University, Hefei, China.
Front Oncol. 2025 Apr 17;15:1541413. doi: 10.3389/fonc.2025.1541413. eCollection 2025.
This study aims to detect vascular and neural invasion in prostate cancer through MRI, utilize habitat analysis of the tumor microenvironment, construct a radiomic feature model, thereby enhancing diagnostic accuracy and prognostic assessment for prostate cancer, ultimately improving patients' quality of life.
We retrospectively collected records of 400 patients with pathologically verified prostate cancer from January to December 2023. We developed a radiomic features model within the tumor habitat using MRI data and identified independent risk factors through multivariate analysis to construct a clinical model. Finally, we assessed the performance of these features using the DeLong test (through the area under the receiver operating characteristic curve, AUC), evaluated the calibration curve with the Hosmer-Lemeshow test, and performed decision curve analysis.
In the training set, the optimal algorithm based on the intratumoral heterogeneity score had an AUC value of 0.882 (CI: 0.843-0.921); in the test set, the AUC value was 0.860 (CI: 0.792-0.928). The traditional radiomics model (considering the entire tumor) had an AUC value of 0.761 (CI: 0.695-0.827) in the training set and 0.732 (CI: 0.630-0.834) in the test set. The combined model that integrates habitat scores and Gleason scores had an AUC value of 0.889 (CI: 0.8509-0.9276) in the training set and 0.886 (CI: 0.8183-0.9533) in the test set, outperforming the single models.
By deeply analyzing the tumor microenvironment and combining radiomics models, the diagnostic precision and predictive accuracy of vascular and nerve invasion in prostate cancer can be significantly improved. This approach provides a valuable tool for optimizing treatment plans, improving patient prognosis, and reducing unnecessary medical interventions.
本研究旨在通过磁共振成像（MRI）检测前列腺癌中的血管和神经侵犯，利用肿瘤微环境的栖息地分析，构建放射组学特征模型，从而提高前列腺癌的诊断准确性和预后评估，最终改善患者的生活质量。
我们回顾性收集了2023年1月至12月400例经病理证实的前列腺癌患者的记录。我们使用MRI数据在肿瘤栖息地内开发了一个放射组学特征模型，并通过多变量分析确定独立危险因素以构建临床模型。最后，我们使用德龙检验（通过受试者操作特征曲线下面积，AUC）评估这些特征的性能，用霍斯默 - 莱梅肖检验评估校准曲线，并进行决策曲线分析。
在训练集中，基于肿瘤内异质性评分的最优算法的AUC值为0.882（CI：0.843 - 0.921）；在测试集中，AUC值为0.860（CI：0.792 - 0.928）。传统放射组学模型（考虑整个肿瘤）在训练集中的AUC值为0.761（CI：0.695 - 0.827），在测试集中为0.732（CI：0.630 - 0.834）。整合栖息地评分和 Gleason评分的联合模型在训练集中的AUC值为0.889（CI：0.
