Our work also shows that it is possible to train a good perform- ing deep neural network on small medical datasets using augmentation. The best-known AUC (0.87) on this data was obtained from a deep neural network trained on cohort 1. Using our designed CNN network trained on the NLST data as a pretrained CNN for feature extraction was also helpful, as it was solely trained on augmented nodule images. Merging these newly obtained deep features with classical radiomics features generates more powerful feature vectors, which even- tually improved performance. 解释

e) {
e.printStackTrace();
}
}

private void updateCourse(int id, String name, String teacher, int credit)这段话讲述了作者的研究成果，即在医疗数据集较小的情况下，通过数据增 {
try (Connection conn = getConnection(); PreparedStatement stmt = conn.prepareStatement(UPDATE)) {
stmt.setString(1, name);
stmt强技术训练深度神经网络，取得了不错的表现。在这些数据中，之前最.setString(2, teacher);
stmt.setInt(3, credit);
stmt.setInt(4, id);
stmt.executeUpdate();
} catch (SQLException e) {
e.printStackTrace();
}
}

private void deleteCourse(int id) {
try (Connection conn =好的AUC（0.87）是由在cohort 1数据集上训练的深度神经网络获得 getConnection(); PreparedStatement stmt = conn.prepareStatement(DELETE)) {
stmt.setInt(1, id);
stmt.executeUpdate();
} catch (SQLException的。作者还设计了一个卷积神经网络（CNN），并使用NLST数据集上训练的该CNN作为 e) {
e.printStackTrace();
}
}
}

最后，我们还需要创建两个 JSP 页面来显示课特征提取的预训练模型，这对于提高性能非常有帮助，因为该CNN仅程列表和单个课程的详细信息。以下是一个简单的实现：

courses.jsp：

<%@在经过数据增强处理的结节图像上进行了训练。最终，作者将这些新获得的 page language="java" contentType="text/html; charset=UTF-8"
pageEncoding="UTF-8"%>
<%@ taglib prefix深度特征与经典的放射学特征进行合并，生成更强大的特征向量，从而进一步提高了性能。