arXivDaily arXiv每日学术速递 周一至周五更新
​通知:联邦节假日调休:哥伦布日 / 原住民日(Columbus Day / Indigenous Peoples' Day),今日(13日)无更新。
arXiv 2606.15019cs.CV

迈向全球人工智能驱动的宫颈癌筛查

Towards Global AI-Driven Cervical Cancer Screening

German Cancer Research Center (DKFZ)(德国癌症研究中心);Heidelberg University(海德堡大学);National Center for Tumor Diseases (NCT) Heidelberg(海德堡国家肿瘤疾病中心);Helmholtz Association(亥姆霍兹联合会);University of Heidelberg(海德堡大学);Medical Faculty Heidelberg(海德堡医学院);University Hospital Heidelberg(海德堡大学医院);German Consortium for Translational Cancer Research (DKTK)(德国转化癌症研究联盟);National Center for Tumor Diseases (NCT) Dresden(德累斯顿国家肿瘤疾病中心);University Hospital Carl Gustav Carus Dresden(德累斯顿卡尔·古斯塔夫·卡鲁斯大学医院);Technische Universität Dresden(德累斯顿工业大学);Helmholtz-Zentrum Dresden-Rossendorf (HZDR)(亥姆霍兹德累斯顿罗森多夫研究中心);University of Bonn(波恩大学);University Hospital Bonn(波恩大学医院);University of Cologne(科隆大学);University Hospital Cologne(科隆大学医院);University of Duisburg-Essen(杜伊斯堡-埃森大学);University Hospital Essen(埃森大学医院);University of Freiburg(弗莱堡大学);University Hospital Freiburg(弗莱堡大学医院);University of Göttingen(哥廷根大学);University Hospital Göttingen(哥廷根大学医院);University of Hamburg(汉堡大学);University Hospital Hamburg-Eppendorf(汉堡-埃彭多夫大学医院);University of Jena(耶拿大学);University Hospital Jena(耶拿大学医院);University of Kiel(基尔大学);University Hospital Schleswig-Holstein(石勒苏益格-荷尔斯泰因大学医院);University of Leipzig(莱比锡大学);University Hospital Leipzig(莱比锡大学医院);University of Lübeck(吕贝克大学);University Hospital Lübeck(吕贝克大学医院);University of Magdeburg(马格德堡大学);University Hospital Magdeburg(马格德堡大学医院);University of Mainz(美因茨大学);University Hospital Mainz(美因茨大学医院);University of Marburg(马尔堡大学);University Hospital Marburg(马尔堡大学医院);University of Munich (LMU)(慕尼黑大学);University Hospital Munich (LMU)(慕尼黑大学医院);Technical University of Munich (TUM)(慕尼黑工业大学);University Hospital rechts der Isar (TUM)(慕尼黑工业大学伊萨尔河右岸医院);University of Münster(明斯特大学);University Hospital Münster(明斯特大学医院);University of Regensburg(雷根斯堡大学);University Hospital Regensburg(雷根斯堡大学医院);University of Rostock(罗斯托克大学);University Hospital Rostock(罗斯托克大学医院);University of Tübingen(蒂宾根大学);University Hospital Tübingen(蒂宾根大学医院);University of Ulm(乌尔姆大学);University Hospital Ulm(乌尔姆大学医院);University of Würzburg(维尔茨堡大学);University Hospital Würzburg(维尔茨堡大学医院)
Thuy Nuong Tran, Ömer Sümer, Evangelia Christodoulou, Lennart Nauschütte, Simon Kalteis, Martin Paulikat, Esmira Pashayeva, Klara Steinheuer, Isabella Borges, P… 展开作者

Thuy Nuong Tran, Ömer Sümer, Evangelia Christodoulou, Lennart Nauschütte, Simon Kalteis, Martin Paulikat, Esmira Pashayeva, Klara Steinheuer, Isabella Borges, Piotr Kalinowski, Hermann Bussmann, Sieng Sokmney, Poeung Kuong, Sathiarany Vong, Achim Schneider, Magnus von Knebel-Doeberitz, Patrick Godau, Lena Maier-Hein

AI总结:

提出首个基于深度学习、在多国数据上验证的宫颈癌筛查方法,通过多任务学习同时进行图像分类和病变分割,在内部验证中优于医生,但外部验证显示性能因国家而异。

摘要:

全球消除宫颈癌是世界卫生组织设定的关键公共卫生目标,筛查项目可将死亡率降低高达80%。然而,中低收入国家在专家和活检服务方面资源有限。基于深度学习的算法为筛查提供了有前景的支持,但现有方法大多在单一国家的私有数据集上开发和验证。我们提出了首个基于深度学习的宫颈癌筛查方法,并在多国数据上进行了验证。技术上,我们将阴道镜图像中病变的检测和分类问题表述为多任务学习问题,同时进行图像级分类和病变分割。我们的模型在带有手动病变分割掩膜和相应组织病理结果的醋酸染色阴道镜图像私有数据集上训练,采用大量数据增强以应对图像变异性。在以内部分布验证中,以病理结果作为金标准,我们的算法在CIN1-(宫颈上皮内瘤变1级或更低)与CIN2+(2级或更高)分类中优于医学专家(平衡准确率:0.68 vs 0.64)。在来自四个国家的四个阴道镜数据集上进行外部验证,这些数据集在患病率和患者特征上存在显著差异,我们的方法相比基线方法表现出更优性能。不同国家间的性能差异较大,AUC值范围从0.54到0.80。总体而言,算法性能随年龄、转化区(最易发生病变的宫颈区域)、合并症和特征性体征的存在而变化,其中合并症的负面影响最大。未来工作应侧重于提高模型的鲁棒性和泛化能力。

英文摘要:

The global elimination of cervical cancer is a key public health goal set by the World Health Organization (WHO), with screening programs reducing mortality by up to 80%. However, access to experts and biopsy services is limited in low- to middle-income countries (LMICs). Deep learning (DL)-based algorithms offer promising support for screening, but most existing approaches have been developed and validated on private datasets from single countries. We present the first DL-based approach to cervical cancer screening validated on data from multiple countries. Technically, we phrase the problem of detecting and classifying lesions in colposcopy images as a multi-task learning problem, in which we simultaneously perform image-level classification and lesion segmentation. Our model was trained on a private data set of acid stain colposcopy images with manually generated lesion segmentation masks and corresponding histopathological results, employing extensive data augmentation to address image variability. In an in-distribution validation with pathology results serving as ground truth, our algorithm outperformed medical experts (Balanced Accuracy: 0.68 vs 0.64) in CIN1- (Cervical intraepithelial neoplasia grade 1 or lower) versus CIN2+ (grade 2 or higher) classification. External validation on four colposcopy data sets from four countries featuring radical differences in prevalence and patient characteristics yielded superior performance of our method compared to baseline methods. Performance variability across countries was high with AUC values ranging from 0.54 - 0.80. Overall, algorithm performance varied with age, transformation zone (cervical area most prone to lesion development), presence of comorbidities and pathognomonic signs, with comorbidities having by far the largest negative effect. Future work should focus on improving model robustness and generalizability.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑