Rauschecker Andreas M, Gleason Tyler J, Nedelec Pierre, Duong Michael Tran, Weiss David A, Calabrese Evan, Colby John B, Sugrue Leo P, Rudie Jeffrey D, Hess Christopher P
Department of Radiology & Biomedical Imaging, University of California, San Francisco, 513 Parnassus Ave, Room S-261, Box 0628, San Francisco, CA 94143-0628 (A.M.R., T.J.G., P.N., D.A.W., E.C., J.B.C., L.P.S., J.D.R., C.P.H.); and Department of Radiology, Hospital of the University of Pennsylvania, Philadelphia, Pa (M.T.D., D.W.).
Radiol Artif Intell. 2021 Nov 10;4(1):e200152. doi: 10.1148/ryai.2021200152. eCollection 2022 Jan.
To assess how well a brain MRI lesion segmentation algorithm trained at one institution performed at another institution, and to assess the effect of multi-institutional training datasets for mitigating performance loss.
In this retrospective study, a three-dimensional U-Net for brain MRI abnormality segmentation was trained on data from 293 patients from one institution (IN1) (median age, 54 years; 165 women; patients treated between 2008 and 2018) and tested on data from 51 patients from a second institution (IN2) (median age, 46 years; 27 women; patients treated between 2003 and 2019). The model was then trained on additional data from various sources: 285 multi-institution brain tumor segmentations, 198 IN2 brain tumor segmentations, and 34 IN2 lesion segmentations from various brain pathologic conditions. All trained models were tested on IN1 and external IN2 test datasets, assessing segmentation performance using Dice coefficients.
The U-Net accurately segmented brain MRI lesions across various pathologic conditions. Performance was lower when tested at an external institution (median Dice score, 0.70 [IN2] vs 0.76 [IN1]). Addition of 483 training cases of a single pathologic condition, including from IN2, did not raise performance (median Dice score, 0.72; = .10). Addition of IN2 training data with heterogeneous pathologic features, representing only 10% (34 of 329) of total training data, increased performance to baseline (Dice score, 0.77; < .001). This final model produced total lesion volumes with a high correlation to the reference standard (Spearman = 0.98).
For brain MRI lesion segmentation, adding a modest amount of relevant training data from an external institution to a previously trained model supported successful application of the model to this external institution. Neural Networks, Brain/Brain Stem, Segmentation © RSNA, 2021.
评估在一个机构训练的脑MRI病变分割算法在另一个机构的表现，并评估多机构训练数据集对减轻性能损失的效果。
在这项回顾性研究中，一个用于脑MRI异常分割的三维U-Net在来自一个机构（IN1）的293例患者的数据上进行训练（年龄中位数为54岁；女性165例；2008年至2018年期间接受治疗的患者），并在来自第二个机构（IN2）的51例患者的数据上进行测试（年龄中位数为46岁；女性27例；2003年至2019年期间接受治疗的患者）。然后，该模型在来自各种来源的额外数据上进行训练：285例多机构脑肿瘤分割、198例IN2脑肿瘤分割以及34例来自各种脑部病理状况的IN2病变分割。所有训练后的模型在IN1和外部IN2测试数据集上进行测试，使用Dice系数评估分割性能。
U-Net能够准确分割各种病理状况下的脑MRI病变。在外部机构进行测试时性能较低（Dice评分中位数，0.70 [IN2] 对比 0.76 [IN1]）。添加483例单一病理状况的训练病例，包括来自IN2的病例，并未提高性能（Dice评分中位数，0.72；P = 0.10）。添加具有异质病理特征的IN2训练数据，仅占总训练数据的10%（329例中的34例），可将性能提高到基线水平（Dice评分，0.77；P < 0.001）。这个最终模型生成的总病变体积与参考标准具有高度相关性（Spearman相关系数 = 0.98）。
对于脑MRI病变分割，向先前训练的模型中添加少量来自外部机构的相关训练数据有助于该模型成功应用于这个外部机构。神经网络，脑/脑干，分割 © RSNA，2021。
