Gholampour Seifollah, Rosen Jacob Benjamin, Pagan Michelangelo, Chen Sonja, Gomaa Ibrahim, Dehghan Arshia, Waterstraat Mark Graham
Department of Neurological Surgery, University of Chicago, Chicago , Illinois , USA.
Neurosurgery. 2025 Jun 1;96(6):1386-1396. doi: 10.1227/neu.0000000000003248. Epub 2024 Nov 7.
Hydrocephalus involves abnormal cerebrospinal fluid accumulation in brain ventricles. Early and accurate diagnosis is crucial for timely intervention and preventing progressive neurological deterioration. The aim of this study was to identify key neuroimaging biomarkers for the diagnosis of hydrocephalus using artificial intelligence to develop practical and accurate diagnostic tools for neurosurgeons.
Fifteen 1-dimensional (1-D) neuroimaging parameters and ventricular volume of adult patients with non-normal pressure hydrocephalus and healthy subjects were measured using manual image processing, and 10 morphometric indices were also calculated. The data set was analyzed using 8 machine, ensemble, and deep learning classifiers to predict hydrocephalus. SHapley Additive exPlanations (SHAP) feature importance analysis identified key neuroimaging diagnostic biomarkers.
Gradient Boosting achieved the highest performance, with an accuracy of 0.94 and an area under the curve of 0.97. SHAP analysis identified ventricular volume as the most important parameter. Given the challenges of measuring volume for clinicians, we identified key 1-D morphometric biomarkers that are easily measurable yet provide similar classifier performance. The results showed that the frontal-temporal horn ratio, modified Evan index, modified cella media index, sagittal maximum lateral ventricle height, and coronal posterior callosal angle are key 1-D diagnostic biomarkers. Notably, higher modified Evan index, modified cella media index, and sagittal maximum lateral ventricle height, and lower frontal-temporal horn ratio and coronal posterior callosal angle values were associated with hydrocephalus prediction. The results also elucidated the relationships between these key 1-D morphometric parameters and ventricular volume, providing potential diagnostic insights.
This study highlights the importance of a multifaceted diagnostic approach incorporating 5 easily measurable 1-D neuroimaging biomarkers for neurosurgeons to differentiate non-normal pressure hydrocephalus from healthy subjects. Incorporating our artificial intelligence model, interpreted through SHAP analysis, into routine clinical workflows may transform the diagnostic landscape for hydrocephalus by standardizing diagnosis and overcoming the limitations of visual evaluations, particularly in early stages and challenging cases.
脑积水涉及脑室中脑脊液异常积聚。早期准确诊断对于及时干预和预防进行性神经功能恶化至关重要。本研究的目的是利用人工智能识别用于诊断脑积水的关键神经影像学生物标志物，为神经外科医生开发实用且准确的诊断工具。
使用人工图像处理测量了成人常压性脑积水患者和健康受试者的15个一维（1-D）神经影像学参数及脑室体积，并计算了10个形态学指标。使用8种机器学习、集成学习和深度学习分类器对数据集进行分析以预测脑积水。SHapley加法解释（SHAP）特征重要性分析确定了关键的神经影像学诊断生物标志物。
梯度提升算法表现最佳，准确率为0.94，曲线下面积为0.97。SHAP分析确定脑室体积是最重要的参数。鉴于临床医生测量体积存在挑战，我们确定了易于测量但能提供相似分类器性能的关键一维形态学生物标志物。结果表明，额颞角比值、改良埃文指数、改良中脑室指数、矢状位最大侧脑室高度和冠状位胼胝体后角是关键的一维诊断生物标志物。值得注意的是，较高的改良埃文指数、改良中脑室指数和矢状位最大侧脑室高度，以及较低的额颞角比值和冠状位胼胝体后角值与脑积水预测相关。结果还阐明了这些关键的一维形态学参数与脑室体积之间的关系，提供了潜在的诊断见解。
本研究强调了多方面诊断方法的重要性，该方法纳入了5种易于测量的一维神经影像学生物标志物，有助于神经外科医生区分常压性脑积水与健康受试者。将通过SHAP分析解释的人工智能模型纳入常规临床工作流程，可能会通过标准化诊断并克服视觉评估的局限性，特别是在早期阶段和具有挑战性的病例中，改变脑积水的诊断格局。
