多任务学习实现垂体大腺瘤自动分割及海绵窦侵犯识别
本研究纳入926例垂体大腺瘤患者,机构1的816例用于训练,机构2的110例用于验证,基于冠状位增强T1加权成像构建多任务多轴注意力UNet(MTMAU-Net),同时完成肿瘤分割与海绵窦侵犯分类;模型分割Dice系数90.84±0.55%、HD95为2.13±0.37 mm,分类准确率88.05±2.84%、AUC 0.89±0.08,验证队列分割Dice系数83.71±5.93%、分类准确率84.55%、AUC 0.87。
要点
- MTMAU-Net在分割任务中Dice系数90.84±0.55%、HD95为2.13±0.37 mm,优于多个单任务模型。
- MTMAU-Net分类准确率88.05±2.84%、AUC 0.89±0.08,与Knosp分级总体表现一致(88.05% vs 87.68%)。
- 在Knosp 3级,模型准确率83.33%,优于人工分级评估的57.62%。
核心结论
- MTMAU-Net在垂体大腺瘤分割和海绵窦侵犯分类上优于单任务模型,在识别海绵窦侵犯方面较Knosp分级系统有优势,可能指导临床医生制定合适的手术策略。
研究设计与样本
- 多任务深度学习模型研究,共纳入926例垂体大腺瘤患者,其中机构1的816例用于模型训练、机构2的110例用于模型验证;在冠状位增强T1加权成像图像上用ITK-SNAP手动分割肿瘤和双侧颈内动脉,由两名神经外科医生在术中评估海绵窦侵犯,提出结合分割与海绵窦侵犯分类任务的MTMAU-Net框架。
诊断标准与关键数值
- 分割性能用Dice相似系数和95% Hausdorff距离(HD95)评估,分类性能用准确率和曲线下面积(AUC)评估;MTMAU-Net分割Dice系数90.84±0.55%、HD95为2.13±0.37 mm,分类准确率88.05±2.84%、AUC 0.89±0.08;与Knosp分级系统相比总体表现一致(准确率88.05% vs 87.68%),在Knosp 3级优于人工分级评估(83.33% vs 57.62%);验证队列分割Dice系数83.71±5.93%、分类准确率84.55%、AUC 0.87。
临床应用与阅片要点
- 作者认为该模型在识别海绵窦侵犯方面较Knosp分级系统有优势,可能指导临床医生制定合适的手术策略。
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
AIM: This study aimed to develop a multitask deep learning model for pituitary macroadenoma (PMA) segmentation and identification of cavernous sinus (CS) invasion.
MATERIALS AND METHODS: A total of 926 patients with PMAs (n=816 from Institution 1 for model training and n=110 from Institution 2 for model validation) were included. Manual segmentation for tumor and bilateral internal carotid arteries was conducted on coronal contrast-enhanced T1-weighted imaging images using ITK-SNAP. Two performing neurosurgeons evaluated the CS invasion during the operation. A Multi-Task Multiaxis-Attention UNet (MTMAU-Net) framework that combined segmentation and CS invasion classification tasks was proposed. Dice similarity coefficient and 95% Hausdorff distance (HD95) were used to evaluate the segmentation performance. Accuracy and area under the curve (AUC) were used to assess the classification performance.
RESULTS: Compared with several single-task models, MTMAU-Net model achieved higher Dice coefficient (90.84±0.55%) and lower HD95 values (2.13±0.37 mm) in segmentation tasks and performed better in the classification task (accuracy: 88.05±2.84%, AUC: 0.89±0.08). Compared with the Knosp grading system, this model showed overall consistent performance (accuracy: 88.05% vs 87.68%) and outperformed human grading evaluation at Knosp grade 3 (83.33% vs 57.62%). MTMAU-Net showed great capability of generalization in both segmentation (Dice coefficient: 83.71±5.93%) and classification tasks (accuracy: 84.55%, AUC: 0.87) in the validation cohort.
CONCLUSION: MTMAU-Net outperformed single-task models in PMA segmentation and CS invasion classification and showed advantages over the Knosp grading system in identifying CS invasion, which might guide clinicians to make appropriate surgical strategies.
CLINICAL TRIAL REGISTRATION: This study is registered at the Chinese Clinical Trial Registration Center with registration number ChiCTR210047614.
引用
- Rui W, Gao W, Qiao N, et al. Automatic pituitary adenoma segmentation and identification of cavernous sinus invasion via multitask learning. Clin Radiol. 2024. PMID: 39689622. doi:10.1016/j.crad.2024.106756
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。