Predicting the probability of converting from mild cognitive impairment (MCI) to Alzheimer's disease (AD) is still a challenging task. This study aims at providing a personalized MCI-to-AD conversion estimation by using a multipredictor nomogram that integrates neuroimaging features, cerebrospinal fluid (CSF) biomarker, and clinical assessments. To do so, 290 MCI patients were collected from the Alzheimer's Disease Neuroimaging Initiative (ADNI), of whom 76 has converted to AD and 214 remained with MCI. All subjects were randomly divided into a primary and validation cohort. Radiomics signature (Rad-sig) was obtained based on 17 cerebral cortex features selected by using Least Absolute Shrinkage and Selection Operator (LASSO) algorithm. Clinical factors and amyloid-beta peptide (Aβ) concentration were selected by using Spearman correlation between the converted and not-converted patients. Then, a nomogram that combines image features, clinical factor, and Aβ concentration was constructed and validated. Furthermore, we explored the associations between various predictors from the macro- to the microperspective by assessing gene expression patterns. Our results showed that the multipredictor nomogram (C-index 0.978 and 0.956 in both cohorts, respectively) outperformed the nomogram using either Rad-sig or Aβ concentration as individual predictors. Significant associations were found between neuropsychological scores, cerebral cortex features, Aβ levels, and underlying gene pathways. Our study may have a clinical impact as a powerful predictive tool for predicting the conversion probability of MCI and providing associations between cognitive impairment, structural changes, Aβ levels, and underlying biological patterns from the macro- to the microperspective.

中文翻译：

---

使用预测列线图预测轻度认知障碍转化为阿尔茨海默病的多重预测模型。

预测从轻度认知障碍 (MCI) 转换为阿尔茨海默病 (AD) 的概率仍然是一项具有挑战性的任务。本研究旨在通过使用集成神经影像特征、脑脊液 (CSF) 生物标志物和临床评估的多预测列线图来提供个性化的 MCI 到 AD 转换估计。为此，从阿尔茨海默病神经影像学倡议 (ADNI) 收集了 290 名 MCI 患者，其中 76 名已转变为 AD，214 名仍留在 MCI。所有受试者被随机分为主要和验证队列。基于使用最小绝对收缩和选择算子 (LASSO) 算法选择的 17 个大脑皮层特征获得放射组学特征 (Rad-sig)。通过使用转换和未转换患者之间的 Spearman 相关性选择临床因素和淀粉样蛋白 β 肽 (Aβ) 浓度。然后，构建并验证了结合图像特征、临床因素和 Aβ 浓度的列线图。此外，我们通过评估基因表达模式探索了从宏观到微观的各种预测因子之间的关联。我们的结果表明，多重预测器列线图（两个队列中的 C 指数分别为 0.978 和 0.956）优于使用 Rad-sig 或 Aβ 浓度作为个体预测因子的列线图。在神经心理学评分、大脑皮层特征、Aβ 水平和潜在基因通路之间发现了显着关联。