Intestinal obstruction is a common surgical emergency in children. However, it is challenging to seek appropriate treatment for childhood ileus since many diagnostic measures suitable for adults are not applicable to children. The rapid development of machine learning has spurred much interest in its application to medical imaging problems but little in medical text mining. In this paper, a two-layer model based on text data such as routine blood count and urine tests is proposed to provide guidance on the diagnosis and assist in clinical decision-making. The samples of this study were 526 children with intestinal obstruction. Firstly, the samples were divided into two groups according to whether they had intestinal obstruction surgery, and then, the surgery group was divided into two groups according to whether the intestinal tube was necrotic. Specifically, we combined 63 physiological indexes of each child with their corresponding label and fed them into a deep learning neural network which contains multiple fully connected layers. Subsequently, the corresponding value was obtained by activation function. The 5-fold cross-validation was performed in the first layer and demonstrated a mean accuracy (Acc) of 80.04%, and the corresponding sensitivity (Se), specificity (Sp), and MCC were 67.48%, 87.46%, and 0.57, respectively. Additionally, the second layer can also reach an accuracy of 70.4%. This study shows that the proposed algorithm has direct meaning to processing of clinical text data of childhood ileus.

中文翻译：

---

用人工智能分析儿童肠梗阻的手术治疗

肠梗阻是儿童常见的外科急症。然而，寻求儿童肠梗阻的适当治疗具有挑战性，因为许多适合成人的诊断措施不适用于儿童。机器学习的快速发展激发了人们对其在医学成像问题中的应用的极大兴趣，但对医学文本挖掘的兴趣却很少。本文提出基于血常规、尿检等文本数据的两层模型，为诊断提供指导，辅助临床决策。本研究的样本为526名患有肠梗阻的儿童。首先根据是否进行过肠梗阻手术将样本分为两组，然后根据肠管是否坏死将手术组分为两组。具体来说，我们将每个孩子的 63 个生理指标与其相应的标签相结合，并将它们输入到包含多个全连接层的深度学习神经网络中。随后通过激活函数得到相应的值。在第一层进行5倍交叉验证，平均准确度（Acc）为80.04%，相应的敏感性（Se）、特异性（Sp）和MCC分别为67.48%、87.46%和0.57，分别。此外，第二层的准确率也可以达到70.4%。本研究表明，所提出的算法对儿童肠梗阻临床文本数据的处理具有直接意义。