Department of Medical Physics and Biomedical Engineering, Tehran University of Medical Sciences, Tehran, Iran.
Department of Neuroscience and Addiction Studies, School of Advanced Technologies in Medicine, Tehran University of Medical Sciences, Tehran, Iran.
Eur J Radiol. 2024 Mar;172:111349. doi: 10.1016/j.ejrad.2024.111349. Epub 2024 Feb 1.
Radiomics analysis of oncologic positron emission tomography (PET) images is an area of significant activity and potential. The reproducibility of radiomics features is an important consideration for routine clinical use. This preliminary study investigates the robustness of radiomics features in PSMA-PET images across penalized-likelihood (Q.Clear) and standard ordered subset expectation maximization (OSEM) reconstruction algorithms and their setting parameters in phantom and prostate cancer (PCa) patients.
A NEMA image quality (IQ) phantom and 8 PCa patients were selected for phantom and patient analyses, respectively. PET images were reconstructed using Q.Clear (reconstruction β-value: 100-700, at intervals of 100 for both NEMA IQ phantom and patients) and OSEM (duration: 15sec, 30sec, 1 min, 2 min, 3 min, 4 min and 5 min for NEMA phantom and duration: 30 s, 1 min and 2 min for patients) reconstruction methods. Subsequently, 129 radiomic features were extracted from the reconstructed images. The coefficient of variation (COV) of each feature across reconstruction methods and their parameters was calculated to determine feature robustness.
The extracted radiomics features showed a different range of variability, depending on the reconstruction algorithms and setting parameters. Specifically, 23.0 % and 53.5 % of features were found as robust against β-value variations in Q.Clear and different durations in OSEM reconstruction algorithms, respectively. Taking into account the two algorithms and their parameters, eleven features (8.5 %) showed COV ≤ 5 % and eighteen (14 %) showed 5 %<COV ≤ 10 %, independent of changes in the methods and parameters. Forty-two (32.5 %) were found with 10 %<COV ≤ 20 %. The remaining fifty-eight features (45 %) showed high variability with COV > 20 %. The mean COVs of the extracted radiomics features were significantly different between the two reconstruction methods (p < 0.05) except for the phantom morphological features.
All radiomics features were affected by reconstruction methods and parameters, but features with small or very small variations are considered better candidates for reproducible quantification of either tumor or metastatic tissues in clinical trials. There is a need for standardization before the implementation of PET radiomics in clinical practice.
肿瘤正电子发射断层扫描（PET）图像的放射组学分析是一个非常活跃且具有潜力的领域。放射组学特征的可重复性是常规临床应用的一个重要考虑因素。本初步研究旨在探讨在体模和前列腺癌（PCa）患者中，基于 penalized-likelihood（Q.Clear）和标准有序子集期望最大化（OSEM）重建算法及其设置参数，对放射组学特征的稳健性。
选择 NEMA 图像质量（IQ）体模和 8 名 PCa 患者分别进行体模和患者分析。使用 Q.Clear（重建β值：100-700，NEMA IQ 体模和患者均为 100 间隔）和 OSEM（持续时间：15sec，30sec，1min，2min，3min，4min 和 5min 用于 NEMA 体模和持续时间：30s，1min 和 2min 用于患者）重建方法重建 PET 图像。随后，从重建图像中提取 129 个放射组学特征。计算每个特征在重建方法及其参数之间的变异系数（COV），以确定特征的稳健性。
提取的放射组学特征表现出不同的变化范围，这取决于重建算法和设置参数。具体而言，在 Q.Clear 中，有 23.0%和 53.5%的特征在不同β值变化下以及 OSEM 重建算法的不同持续时间下具有稳健性。考虑到两种算法及其参数，有 11 个特征（8.5%）的 COV≤5%，18 个特征（14%）的 5%<COV≤10%，不受方法和参数变化的影响。有 42 个特征（32.5%）的 COV 在 10%<COV≤20%之间。其余 58 个特征（45%）的 COV 较高，COV>20%。提取的放射组学特征的平均 COV 在两种重建方法之间有显著差异（p<0.05），除了体模形态特征外。
所有放射组学特征都受到重建方法和参数的影响，但变化小或非常小的特征被认为是在临床试验中对肿瘤或转移组织进行可重复定量的更好候选者。在将 PET 放射组学应用于临床实践之前，需要进行标准化。
