基于CT图像的人工智能模型预测卵巢癌新辅助化疗疗效

A Deep Learning Model for Predicting the Efficacy of Neoadjuvant Chemotherapy for Ovarian Cancer Based on CT Images

目的应用治疗前卵巢癌患者CT增强动脉期图像,结合深度学习及机器学习算法预测卵巢癌新辅助化疗疗效。资料与方法回顾性选取2013年3月—2019年8月天津市肿瘤医院卵巢癌患者302例,按照7∶3随机分为训练集、测试集;在python环境中,采用迁移学习方法,利用torch深度学习平台构建VGG13模型,联合机器学习,采用套索算法筛选特征,建立联合深度学习及机器学习综合预测模型对患者治疗前CT图像进行分类预测,计算模型的曲线下面积、准确度、敏感度、特异度及F1-Score。结果本研究所建模型在训练集中的曲线下面积、准确度、敏感度、特异度及F1-Score分别为0.87、0.81、0.80、0.82、0.79;测试集中分别为0.90、0.84、0.93、0.77、0.83。五折交叉验证曲线下面积分别为0.86、0.88、0.88、0.90、0.87。结论采用CT增强动脉期图像联合人工智能方法建立预测模型可为制订卵巢癌化疗方案提供一种新的临床思路。

Purpose To use CT-arterial phase images of pre-treatment ovarian cancer patients,combined with deep learning algorithms and machine learning to build a model to predict the efficacy of neoadjuvant chemotherapy in ovarian cancer.Materials and Methods A total of 302 consecutive patients who underwent surgery and were pathologically diagnosed with ovarian cancer from March 2013 to August 2019 in Tianjin Medical University were retrospectively collected.All patients were partitioned into training and test sets according to the ratio of 7∶3.In the python environment,VGG13 model was integrated via combining deep learning network and machine learning,and features were filtered via least absolute shrinkage and selection operator algorithm to build a prediction model for classification and prediction of CT images.The area under the curve(AUC),accuracy,sensitivity,specificity,and F1-Score were calculated,respectively.Results The AUC,accuracy,sensitivity,specificity,and F1-Score of the model in the training set were 0.87,0.81,0.80,0.82 and 0.79,and 0.90,0.84,0.93,0.77 and 0.83 in the test set,respectively.The AUC of five-fold cross-validation were 0.86,0.88,0.88,0.90 and 0.87,respectively.Conclusion Predictive model based on CT images combined with deep learning and machine learning methods can provide a new clinical perspective for developing chemotherapy regimens for ovarian cancer.