基于多通道与注意力机制的端到端深度学习模型用于晚期胃癌CT-T分期多分类诊断
本研究为回顾性研究,纳入2011至2024年间460例晚期胃癌术前CT患者,构建基于3D-Conv-UNet自动分割肿瘤与SmallFocusNet三元分类模型的端到端深度学习模型用于CT-T分期。分割模型在测试集DSC为0.860±0.065、独立验证集为0.870±0.164;分类模型测试集宏平均AUC为0.882(95%CI 0.812-0.926),独立验证集为0.862(95%CI 0.723-0.942)。端到端DL模型准确率91.9%优于放射科医师的82.1%(P=0.007)。
要点
- 端到端DL模型在晚期胃癌CT-T分期中准确率91.9%,优于放射科医师的82.1%(P=0.007)。
- 分割模型测试集DSC为0.860±0.065,独立验证集DSC为0.870±0.164。
- 分类模型测试集宏平均AUC为0.882(95%CI 0.812-0.926),独立验证集为0.862(95%CI 0.723-0.942)。
核心结论
- 用于CT-T分期的端到端深度学习模型在晚期胃癌治疗前分期中具有高准确性和一致性。
研究设计与样本
- 回顾性研究纳入2011至2024年间460例晚期胃癌术前CT患者,数据分为Dataset 1(423例,训练与测试集,平均年龄65.0岁±9.46)和Dataset 2(37例,独立验证集,平均年龄68.8岁±9.28);采用3D-Conv-UNet自动分割模型与SmallFocusNet三元分类模型整合为端到端DL模型。
诊断标准与关键数值
- 分割模型测试集DSC为0.860±0.065、IoU为0.760±0.096,Dataset 2的DSC为0.870±0.164、IoU为0.793±0.168;分类模型测试集宏平均AUC为0.882(95%CI 0.812-0.926)、平均敏感度为76.9%(95%CI 67.6%-85.3%),Dataset 2宏平均AUC为0.862(95%CI 0.723-0.942)、平均敏感度为76.3%(95%CI 59.8%-90.0%)。
临床应用与阅片要点
- 端到端DL模型准确率为91.9%,优于放射科医师的82.1%(P=0.007)。
局限性
- 原文未报告局限性。
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)。
英文原文(摘要)· 点击展开
BACKGROUND: Homogeneous AI assessment is required for CT-T staging of gastric cancer.
PURPOSE: To construct an End-to-End CT-based Deep Learning (DL) model for tumor T-staging in advanced gastric cancer.
MATERIALS AND METHODS: A retrospective study was conducted on 460 cases of presurgical CT patients with advanced gastric cancer between 2011 and 2024. A Three-dimensional (3D)-Convolution (Conv)-UNet based automatic segmentation model was employed to segment tumors, and a SmallFocusNet-based ternary classification model was built for CT-T staging. Finally, these models were integrated to create an end-to-end DL model. The segmentation model's performance was assessed using the Dice similarity coefficient (DSC), Intersection over Union (IoU) and 95 % Hausdorff Distance (HD_95), while the classification model's performance was measured with thearea under the Receiver Operating Characteristic curve (AUC), sensitivity, specificity, and F1-score.Eventually, the end-to-end DL model was compared with the radiologist using the McNemar test.
RESULTS: The data were divided into Dataset 1(423 cases for training and test set, mean age, 65.0 years ± 9.46 [SD]) and Dataset 2(37 cases for independent validation set, mean age, 68.8 years ± 9.28 [SD]). For segmentation task, the model achieved a DSC of 0.860 ± 0.065, an IoU of 0.760 ± 0.096 in test set of Dataset 1, and a DSC of 0.870 ± 0.164, an IoU of 0.793 ± 0.168 in Dataset 2. For classification task,the model demonstrated a macro-average AUC of 0.882(95 % CI 0.812-0.926), an average sensitivity of 76.9 % (95 % CI 67.6 %-85.3 %) in test set of Dataset 1 and a macro-average AUC of 0.862(95 % CI 0.723-0.942), an average sensitivity of 76.3 % (95 % CI 59.8 %-90.0 %) in Dataset 2. Meanwhile, the DL model's performance was better than that of radiologist (Accuracy was 91.9 %vs82.1 %, P = 0.007).
CONCLUSION: The end-to-end DL model for CT-T staging is highly accurate and consistent in pre-treatment staging of advanced gastric cancer.
引用
- Liu B, Jiang P, Wang Z, et al. End-to-end deep learning model with multi-channel and attention mechanisms for multi-class diagnosis in CT-T staging of advanced gastric cancer. Eur J Radiol. 2025. PMID: 40913943. doi:10.1016/j.ejrad.2025.112408
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。