Chartrand Gabriel, Emiliani Ramón D, Pawlowski Sophie A, Markel Daniel A, Bahig Houda, Cengarle-Samak Alexandre, Rajakesari Selvan, Lavoie Jeremi, Ducharme Simon, Roberge David
AFX Medical Inc., Montréal, Canada.
Department of Radiation Oncology, Centre Hospitalier de l'Université de Montréal, Montréal, Québec, Canada.
J Magn Reson Imaging. 2022 Dec;56(6):1885-1898. doi: 10.1002/jmri.28274. Epub 2022 May 27.
Detection of brain metastases (BM) and segmentation for treatment planning could be optimized with machine learning methods. Convolutional neural networks (CNNs) are promising, but their trade-offs between sensitivity and precision frequently lead to missing small lesions.
Combining volume aware (VA) loss function and sampling strategy could improve BM detection sensitivity.
Retrospective.
A total of 530 radiation oncology patients (55% women) were split into a training/validation set (433 patients/1460 BM) and an independent test set (97 patients/296 BM).
FIELD STRENGTH/SEQUENCE: 1.5 T and 3 T, contrast-enhanced three-dimensional (3D) T1-weighted fast gradient echo sequences.
Ground truth masks were based on radiotherapy treatment planning contours reviewed by experts. A U-Net inspired model was trained. Three loss functions (Dice, Dice + boundary, and VA) and two sampling methods (label and VA) were compared. Results were reported with Dice scores, volumetric error, lesion detection sensitivity, and precision. A detected voxel within the ground truth constituted a true positive.
McNemar's exact test to compare detected lesions between models. Pearson's correlation coefficient and Bland-Altman analysis to compare volume agreement between predicted and ground truth volumes. Statistical significance was set at P ≤ 0.05.
Combining VA loss and VA sampling performed best with an overall sensitivity of 91% and precision of 81%. For BM in the 2.5-6 mm estimated sphere diameter range, VA loss reduced false negatives by 58% and VA sampling reduced it further by 30%. In the same range, the boundary loss achieved the highest precision at 81%, but a low sensitivity (24%) and a 31% Dice loss.
Considering BM size in the loss and sampling function of CNN may increase the detection sensitivity regarding small BM. Our pipeline relying on a single contrast-enhanced T1-weighted MRI sequence could reach a detection sensitivity of 91%, with an average of only 0.66 false positives per scan.
3 TECHNICAL EFFICACY: Stage 2.
利用机器学习方法可优化脑转移瘤（BM）的检测及用于治疗规划的分割。卷积神经网络（CNN）很有前景，但其在敏感性和精确性之间的权衡常常导致小病灶漏检。
结合体积感知（VA）损失函数和采样策略可提高BM检测的敏感性。
回顾性研究。
共530例放射肿瘤学患者（55%为女性）被分为训练/验证集（433例患者/1460个BM）和独立测试集（97例患者/296个BM）。
场强/序列：1.5T和3T，对比增强三维（3D）T1加权快速梯度回波序列。
真实掩膜基于专家审核的放射治疗规划轮廓。训练了一个受U-Net启发的模型。比较了三种损失函数（骰子损失、骰子损失+边界损失和VA损失）和两种采样方法（标签采样和VA采样）。结果以骰子分数、体积误差、病灶检测敏感性和精确性报告。真实掩膜内检测到的体素构成真阳性。
采用McNemar精确检验比较模型间检测到的病灶。采用Pearson相关系数和Bland-Altman分析比较预测体积与真实体积之间的体积一致性。统计学显著性设定为P≤0.05。
结合VA损失和VA采样表现最佳，总体敏感性为91%，精确性为81%。对于估计球体直径在2.5 - 6mm范围内的BM，VA损失使假阴性减少58%，VA采样进一步减少30%。在同一范围内，边界损失达到最高精确性81%，但敏感性较低（24%），骰子损失为31%。
在CNN的损失和采样函数中考虑BM大小可能会提高对小BM的检测敏感性。我们基于单一对比增强T1加权MRI序列的流程可达到91%的检测敏感性，每次扫描平均仅有0.66个假阳性。
3 技术效能：2级
