卷积神经网络提取超声影像特征预测腹腔镜卵巢囊肿剥除术手术时间延长
本研究为单中心回顾性队列,纳入247例因疑似良性卵巢肿瘤行腹腔镜卵巢囊肿剥除术的患者,构建基于术前临床变量的逻辑回归模型及联合卷积神经网络(CNN)超声特征的模型预测手术时间延长(队列手术时长上四分位数)。多变量分析显示机器人手术、双侧囊肿、CA-125水平及两个CNN影像特征与手术时间延长独立相关;联合CNN特征使AUC从0.889升至0.920,但差异无统计学意义。
要点
- 机器人手术、双侧卵巢囊肿、CA-125水平及两个CNN超声特征与手术时间延长独立相关。
- 联合CNN超声特征使预测模型AUC由0.889升至0.920,但差异未达统计学显著。
- 联合模型展示术前风险分层可行性,尚需外部验证支持手术排程应用。
核心结论
- 联合临床变量与CNN提取的超声影像特征可预测腹腔镜卵巢囊肿剥除术的手术时间延长,但增量预测价值未达统计学显著。
研究设计与样本
- 单中心回顾性队列研究,纳入247例18岁及以上、因疑似良性卵巢肿瘤行腹腔镜卵巢囊肿剥除术且具备术前超声图像和完整手术记录的患者。
主要发现
- 多变量分析显示机器人手术、卵巢囊肿双侧性、CA-125水平及两个CNN衍生影像特征与手术时间延长独立相关。
- 联合CNN超声特征使预测模型AUC从0.889提高至0.920,但差异未达统计学显著。
关键数值与推荐
- 手术时间延长定义为队列中手术时长的上四分位数(25%)。
临床怎么用
- 该联合模型可能辅助术前风险分层,并为手术排程和资源规划研究提供基础,但需外部验证。
局限性
- 单中心回顾性设计,样本量247例,CNN特征的增量预测价值未达统计学显著,需外部验证。
以上解读依据论文摘要(全文未开放获取或许可不允许翻译转载)。
原文
OBJECTIVE: The aim of this study was to identify clinical and imaging factors associated with prolonged operative time and to explore whether convolutional neural network (CNN)-derived ultrasound features provide incremental predictive value beyond conventional clinical variables. DESIGN: Retrospective cohort study. SETTING: Single academic medical center. PARTICIPANTS: A total of 247 patients who underwent laparoscopic ovarian cystectomy for presumed benign ovarian tumors were included in the study. Eligible participants were 18 years or older, had preoperative ultrasound images and complete operative records, and were initially planned for ovarian cystectomy. Patients were excluded if they were scheduled for oophorectomy or if other major procedures, such as myomectomy or hysterectomy, were planned concurrently. INTERVENTIONS: We used two predictive models: a logistic regression model based on preoperative clinical variables (e.g., CA-125 levels, bilaterality of ovarian cysts, and robotic surgery) to predict prolonged operative time, defined as the upper quartile (25%) of operative duration in our cohort, and a second logistic regression model combining these variables with CNN-derived features from preoperative ultrasound images to enhance predictive accuracy. RESULTS: Multivariable analysis revealed that robotic surgery, bilaterality of ovarian cysts, CA-125 levels, and two CNN-derived imaging features were independently associated with prolonged operative time. Incorporation of CNN-derived ultrasound features increased the area under the curve from 0.889 to 0.920, although the difference did not reach statistical significance. CONCLUSION: The combined model, incorporating clinical and CNN-extracted imaging features, demonstrates the feasibility of predicting prolonged operative time in laparoscopic ovarian cystectomy. This approach may support preoperative risk stratification and serve as a foundation for future studies exploring its role in surgical scheduling and resource planning, pending external validation.
引用
- Kim J, Kim Y, Bak H, et al. Preoperative Prediction of Prolonged Operative Time in Laparoscopic Ovarian Cystectomy Using Convolutional Neural Network-Extracted Ultrasound Image Features. J Minim Invasive Gynecol. 2026. PMID: 41621701. doi:10.1016/j.jmig.2026.01.055
供执业医师工作参考,不替代临床判断;数值与分级以指南原文为准。