Article Figures & Data
Figures
- FIG 1.
Schematic illustration of the preprocessing, data augmentation, and model training pipeline. A, 3D contrast-enhanced T1 input volume. B, Preprocessed and augmented CET1 to mitigate overfitting. C, 3D-DenseNet model. D, Prediction of absence or presence of CSF leak.
- FIG 2.
Receiver operating characteristic (ROC) curves for 5-fold cross-validation. Each curve represents the performance of the model on a distinct validation fold. The curves demonstrate the model’s ability to distinguish between the absence and presence of CSF leaks from brain MR imaging scans.
- FIG 3.
Three-part representation of the regions that are crucial to the model’s decision-making process in detecting CSF leaks. A, Sagittal view of contrast-enhanced T1 brain MR imaging. B, Occlusion mask overlaid on the original contrast-enhanced T1 image, highlighting the regions that significantly influence the model’s predictions. C, Occlusion mask generated to identify regions of interest.
Tables
Subject Characteristics All Subjects (n = 129) Median age in years (interquartile range) 54 (20) Age range in years 23–87 Female 84 (65.12%) Male 45 (34.48%) Bern score Low risk (Bern score 0–2) 47 (36.43%) Intermediate risk (Bern score 3–4) 26 (20.15%) High risk (Bern score >4) 56 (43.42%) Fold Number AUROC 1 0.8988 2 0.8284 3 0.8580 4 0.8580 5 0.8910
