Table 2.

Prediction accuracy of optimised kernel SVM based on different combinations of input features; including 95% confidence intervals and corresponding p-values.

* Summaries include total, mean, median, minimum, maximum and standard deviation (SD) of lesion volume, surface area and mean breadth, respectively. Measures are based on T1-Gd MRI data. ** Total, mean and SD of volume, surface area and mean breadth (excluding median, minimum, maximum measures), and EP.

Abbreviations: DC -demographic and clinical covariates, EDSS - expanded disability status scale, EP - Euler-Poincare characteristic, GEO-brain - whole-brain summaries of geometric measures, GM - grey matter, ROI - region of interest.

model	input features [resulting number of features]	placebo group (N=61, conv+: 39, conv-: 22)	IFN-b group (N=99, conv+: 50, conv-: 49)
M1	demographic and clinical covariates (DC): age, sex, EDSS [3 features]	56.7% [48.5–64.9%] (p=0.053) [77:63] *	55.5% [49.0-62.0%] (p=0.051) [98:126] *
M2	DC, lesion count, lesion load (i.e. total lesion volume) [5 f.]	50.1% [37.6-62.6%] (p=0.537)	57.6% [47.9-67.3%] (p=0.072)
M3	DC, GM volume ratio (GM) [4 f.]	54.3% [41.8-66.8%] (p=0.261)	58.6% [48.9-68.3%] (p=0.048)
M4	DC, GM, lesion count, lesion load [6 f.]	60.9% [48.7-73.1%] (p=0.046)	60.6 [51.0-70.2%] (p=0.020)
M5	DC, GM, count, load, Euler-Poincare characteristic (EP) [7 f.]	56.6% [44.2-69.0%] (p=0.145)	54.5% [44.7-64.3%] (p=0.193)
M6	whole-brain summaries* of geometric measures (GEO-brain) [21 f.]	57.9% [45.5-70.3%] (p=0.126)	58.5% [48.8-68.2%] (p=0.046)
M7	DC, GM, GEO-brain [25 f.]	67.6% [55.9-79.3%] (p=0.0022)	64.6% [55.2-74.0%] (p=0.0017)
M8	DC, GM by ROI, geometric measures by ROI (GEO-ROI) [276 f.]	51.2% [38.7-63.7%] (p=0.468)	59.6% [49.9-69.3%] (p=0.035)
M9	DC, GM, lesion count, subset** of GEO- brain [15 f.]	57.8% [45.4-70.2%] (p=0.133)	70.4% [61.4-79.4%] (p=8e-5)
M10	DC, GM by ROI, lesion count, subset** of GEO-ROI [159 f.]	50.5% [38.0-63.0%] (p=0.491)	66.6% [57.3-75.9%] (p=0.0004)