摘要
目的:探究基于深度学习的智慧医疗辅助诊断系统在早期宫颈癌筛查中的价值。方法:在本院电子病历系统中检索2022年4月—2024年4月确诊的早期宫颈癌患者206例、宫颈上皮内瘤变患者798例。按7∶3的比例划分训练集与验证集并进行模型训练,确定损失函数与优化算法,运用数据增强提升模型性能。比较各模型性能指标、ResNeXt模型与传统诊断方法及ResNeXt模型与不同年资医生的诊断性能。结果:共纳入训练集702例患者,验证集302例患者。在训练集和验证集中,早期宫颈癌组和宫颈上皮内瘤变组在年龄、BMI方面的差异无统计学意义(P>0.05)。在各模型中,ResNeXt模型的准确率、召回率、特异度及F1值指标表现最优。在验证集中,ResNeXt模型的诊断效率显著高于TCT和宫颈活检(P<0.05),且漏诊率和误诊率均低于液基薄层细胞学检查(TCT)和宫颈活检(P<0.05);其准确率、召回率和特异度均高于不同年资医生。结论:基于深度学习的智慧医疗辅助诊断系统在早期宫颈癌筛查中展现出显著优势,其中ResNeXt模型表现尤为突出。该系统不仅提高了诊断准确性和效率,而且降低漏诊和误诊率,可成为临床医生在早期宫颈癌筛查中的重要辅助工具。
Objective:To explore the value of deep learning-based intelligent medical auxiliary diagnosis system in early cervical cancer screening.Methods:206 patients diagnosed with early-stage cervical cancer and 798 patients with cervical intraepithelial neoplasia were retrieved from our hospital’s electronic medical record system between April 2022 and April 2024.The dataset was divided into the training set and the validation set at a ratio of 7∶3 for model training,with the loss function and optimization algorithm being specified.Data augmentation techniques were utilized to improve model performance.Performance metrics of different models,ResNeXt model versus traditional diagnostic methods(TCT/cervical biopsy),and ResNeXt model versus physicians of varying seniority were compared.Results:A total of 702 patients were included in the training set and 302 patients in the validation set.No significant differences were observed in age or BMI between patients with early-stage cervical cancer and those with cervical intraepithelial neoplasia in the two sets(P>0.05).Among all models,ResNeXt achieved optimal accuracy,recall,specificity,and F1-scores.In the validation set,ResNeXt’s diagnostic efficiency surpassed TCT and biopsies(P<0.05),with lower missed and misdiagnosis rates(P<0.05);its accuracy,recall,and specificity also outperformed physicians across seniority levels.Conclusion:The intelligent medical auxiliary diagnosis system based on deep learning has significant advantages in early cervical cancer screening,with ResNeXt excelling in improving diagnostic accuracy,efficiency,and reducing errors,which can be a vital tool for clinicians in cervical cancer screening.
作者
王红
王佩
梁月娟
惠莉
郭晓云
WANG Hong;WANG Pei;LIANG Yuejuan;HUI Li;GUO Xiaoyun(Department of Gynecology,Northwest Women’s and Children’s Hospital,Xi’an 710061,China)
基金
2024年华中科技大学同济医学院附属同济医院肿瘤侵袭转移教育部重点实验室课题(2024KFKT010)。
关键词
深度学习
智慧医疗
早期宫颈癌
筛查
Deep Learning
Intelligent Healthcare
Early Cervical Cancer
Screening
