Unit of Nuclear Medicine, Fondazione Istituto G. Giglio, Cefalù, Palermo, Italy -
Unit of Nuclear Medicine, Fondazione Istituto G. Giglio, Cefalù, Palermo, Italy.
Q J Nucl Med Mol Imaging. 2022 Dec;66(4):352-360. doi: 10.23736/S1824-4785.20.03227-6. Epub 2020 Jun 15.
Radiomic features are increasingly utilized to evaluate tumor heterogeneity in PET imaging but to date its role has not been investigated for Cho-PET in prostate cancer. The potential application of radiomics features analysis using a machine-learning radiomics algorithm was evaluated to select F-Cho PET/CT imaging features to predict disease progression in PCa.
We retrospectively analyzed high-risk PCa patients who underwent restaging F-Cho PET/CT from November 2013 to May 2018. F-Cho PET/CT studies and related structures containing volumetric segmentations were imported in the "CGITA" toolbox to extract imaging features from each lesion. A Machine-learning model has been adapted using NCA for feature selection, while DA was used as a method for feature classification and performance analysis.
One hundred and six imaging features were extracted for 46 lesions for a total of 4876 features analyzed. No significant differences between the training and validating sets in terms of age, sex, PSA values, lesion location and size (P>0.05) were demonstrated by the machine-learning model. Thirteen features were able to discriminate FU disease status after NCA selection. Best performance in DA classification was obtained using the combination of the 13 selected features (sensitivity 74%, specificity 58% and accuracy 66%) compared to the use of all features (sensitivity 40%, specificity 52%, and accuracy 51%). Per-site performance of the 13 selected features in DA classification were as follows: T = sensitivity 63%, specificity 83%, accuracy 71%; N = sensitivity 87%, specificity 91% of and accuracy 90%; bone-M = sensitivity 33%, specificity 77% and accuracy 66%.
An artificial intelligence model demonstrated to be feasible and able to select a panel of F-Cho PET/CT features with valuable association with PCa patients' outcome.
放射组学特征越来越多地用于评估 PET 成像中的肿瘤异质性，但迄今为止，其在前列腺癌的 Cho-PET 中的作用尚未得到研究。本研究评估了使用机器学习放射组学算法的放射组学特征分析在选择 F-Cho PET/CT 成像特征以预测前列腺癌疾病进展中的潜在应用。
我们回顾性分析了 2013 年 11 月至 2018 年 5 月接受 F-Cho PET/CT 分期的高危前列腺癌患者。F-Cho PET/CT 研究和包含容积分割的相关结构被导入“CGITA”工具包中，以从每个病变中提取成像特征。使用 NCA 进行特征选择，而使用 DA 进行特征分类和性能分析，适应了一种机器学习模型。
共对 46 个病变提取了 106 个成像特征，共分析了 4876 个特征。通过机器学习模型，在年龄、性别、PSA 值、病变位置和大小方面，训练集和验证集之间没有显著差异（P>0.05）。在 NCA 选择后，13 个特征能够区分 FU 疾病状态。在 DA 分类中，使用 13 个选定特征的组合获得了最佳性能（敏感性 74%，特异性 58%和准确性 66%），而使用所有特征的敏感性为 40%，特异性为 52%，准确性为 51%。在 DA 分类中，13 个选定特征的每个站点性能如下：T = 敏感性 63%，特异性 83%，准确性 71%；N = 敏感性 87%，特异性 91%和准确性 90%；骨-M = 敏感性 33%，特异性 77%和准确性 66%。
人工智能模型被证明是可行的，并能够选择一组与前列腺癌患者预后有价值关联的 F-Cho PET/CT 特征。
