Table 2. Frequency of variable inclusion in LR predictor.
Feature | Frequency at t = 2 years | Frequency at t = 3 years | Frequency at t = 4 years | Frequency at t = 6 years | Frequency at t = 8 years | Avg. |
---|---|---|---|---|---|---|
Maximum diameter (mm) | 1 | 1 | 1 | 1 | 1 | 1 |
Prior embolization (yes/no) | 0.88 | 1 | 1 | 1 | 0.99 | 0.974 |
Marginal dose (Gy) | 0.92 | 0.97 | 0.94 | 0.98 | 0.92 | 0.946 |
Number of isocenters (no.) | 0.84 | 0.81 | 0.78 | 0.96 | 0.95 | 0.868 |
Location 8 – brain stem (yes/no) | 0.63 | 0.71 | 0.9 | 0.96 | 0.93 | 0.826 |
Associated aneurysm (yes/no) | 0.68 | 0.88 | 0.84 | 0.86 | 0.83 | 0.818 |
Location 5 – thalamic (yes/no) | 0.77 | 0.78 | 0.81 | 0.88 | 0.75 | 0.798 |
Location 2 – temporal (yes/no) | 0.68 | 0.73 | 0.91 | 0.82 | 0.76 | 0.78 |
Surgery (yes/no) | 0.81 | 0.49 | 0.76 | 0.86 | 0.74 | 0.732 |
Deep venous drainage (yes/no) | 0.69 | 0.64 | 0.64 | 0.85 | 0.82 | 0.728 |
History of Hemorrhage (yes/no) | 0.69 | 0.73 | 0.62 | 0.75 | 0.73 | 0.704 |
Sex (male/female) | 0.68 | 0.61 | 0.69 | 0.79 | 0.7 | 0.694 |
Age (years) | 0.6 | 0.64 | 0.6 | 0.77 | 0.67 | 0.656 |
3D surface dose (Gy × mm2)* | 0.87 | 0.71 | 0.6 | 0.51 | 0.59 | 0.656 |
Location 9 – cerebellum (yes/no) | 0.64 | 0.76 | 0.62 | 0.62 | 0.59 | 0.646 |
Location 1 – frontal (yes/no) | 0.5 | 0.56 | 0.63 | 0.71 | 0.75 | 0.63 |
Location 7 – callosal (yes/no) | 0.65 | 0.51 | 0.61 | 0.72 | 0.62 | 0.622 |
Location 4 – occipital (yes/no) | 0.55 | 0.42 | 0.6 | 0.73 | 0.8 | 0.62 |
Volume (mm3) | 0.59 | 0.4 | 0.57 | 0.75 | 0.72 | 0.606 |
Location 3 – parietal (yes/no) | 0.82 | 0.58 | 0.53 | 0.56 | 0.47 | 0.592 |
Location 6 – BGߤ (yes/no) | 0.8 | 0.51 | 0.5 | 0.57 | 0.57 | 0.59 |
Isodose (%) | 0.81 | 0.51 | 0.49 | 0.56 | 0.55 | 0.584 |
Max dose (Gy) | 0.48 | 0.4 | 0.7 | 0.52 | 0.59 | 0.538 |
The set of utilized features in the logistic regression predictor (LR predictor) and their frequency of model inclusion at different time points. Bolded text indicates the top five utilized features at each time point, with a final column (Avg.) denoting the average rate of inclusion in the predictors across all time points. The majority of features were derived from original patient data, while one feature, marginal dose (Gy) × surface area (mm2), was consistently selected for inclusion in the learning model and of particular use at predicting early obliteration rates.
*The engineered feature selected for inclusion in the model.
ߤbasal ganglia.