CT图像机器学习在预测良性、侵袭前和侵袭性肺结节的价值_J Thorac Cardiovasc Surg. 2 ...

方法：收集多个来源的病理诊断为肺结节的胸部ct扫描数据。数据集被随机分为训练组（70%）、内部验证组（15%）和独立测试组（15%）。开发、训练和验证了两种机器学习算法。第一种算法采用支持向量机模型，第二种算法采用深度学习技术：卷积神经网络。在测试数据集上，用ROC分析来评价分类的性能。

结果：基于支持向量机/卷积神经网络的模型将结节分为6类，不典型腺瘤性增生与原位腺癌的鉴别曲线下面积为0.59/0.65，微浸润腺癌与浸润性腺癌的鉴别ROC曲线下面积为0.87/0.86，0.76/0.72非典型腺瘤增生 原位腺癌与微浸润腺癌，0.89/0.87非典型腺瘤增生 原位腺癌与微浸润腺癌 浸润性腺癌，非典型腺瘤样增生 原位腺癌 微浸润腺癌与浸润性腺癌的比值为0.93/0.92。对良性和非典型腺瘤性增生 原位腺癌 微浸润腺癌和浸润性腺癌进行分类，支持向量机/卷积神经网络模型的曲线下微平均面积分别为0.93/0.94。基于卷积神经网络的方法比基于支持向量机的方法具有更高的灵敏度，但其特异性和准确性较低。

结论：机器学习算法在区分良性、侵袭前和侵袭性腺癌与单纯CT图像方面表现出了合理的性能。然而，预测精度因其子类型而异。这就有可能以低创伤性手段提高诊断能力。

关键词：分类；计算机断层扫描；肺腺癌；病理亚型。

Objective: The study objective was to investigate if machine learning algorithms can predict whether a lung nodule is benign, adenocarcinoma, or its preinvasive subtype from computed tomography images alone.

Methods: A dataset of chest computed tomography scans containing lung nodules was collected with their pathologic diagnosis from several sources. The dataset was split randomly into training (70%), internal validation (15%), and independent test sets (15%) at the patient level. Two machine learning algorithms were developed, trained, and validated. The first algorithm used the support vector machine model, and the second used deep learning technology: a convolutional neural network. Receiver operating characteristic analysis was used to evaluate the performance of the classification on the test dataset.

Results: The support vector machine/convolutional neural network-based models classified nodules into 6 categories resulting in an area under the curve of 0.59/0.65 when differentiating atypical adenomatous hyperplasia versus adenocarcinoma in situ, 0.87/0.86 with minimally invasive adenocarcinoma versus invasive adenocarcinoma, 0.76/0.72 atypical adenomatous hyperplasia adenocarcinoma in situ versus minimally invasive adenocarcinoma, 0.89/0.87 atypical adenomatous hyperplasia adenocarcinoma in situ versus minimally invasive adenocarcinoma invasive adenocarcinoma, and 0.93/0.92 atypical adenomatous hyperplasia adenocarcinoma in situ minimally invasive adenocarcinoma versus invasive adenocarcinoma. Classifying benign versus atypical adenomatous hyperplasia adenocarcinoma in situ minimally invasive adenocarcinoma versus invasive adenocarcinoma resulted in a micro-average area under the curve of 0.93/0.94 for the support vector machine/convolutional neural network models, respectively. The convolutional neural network-based methods had higher sensitivities than the support vector machine-based methods but lower specificities and accuracies.

Conclusions: The machine learning algorithms demonstrated reasonable performance in differentiating benign versus preinvasive versus invasive adenocarcinoma from computed tomography images alone. However, the prediction accuracy varies across its subtypes. This holds the potential for improved diagnostic capabilities with less-invasive means.

Keywords: classification; computed tomography; lung adenocarcinoma; pathological subtype.