Table 2 Overview of tumor-relevant features extracted from the preoperative radiology report. The chi-square test was used to assess categorical variables with more than two categories, in which case the resulting p-value is denoted in the first row relating to that category. For binary variables, Fisher’s exact test was used, and unpaired t-test with Welsh’s correction for continuous variables.

From: A cost-sensitive multiclass machine learning framework for postoperative neurosurgical triage (Neuro-TACTIC)

Feature

Main dataset n (%)

Evaluation dataset n (%)

p-value

Tumor volume

21.47 ± 31.23

19.82 ± 27.21

0.6068

Hydrocephalus on MRI

61 (5.7%)

3 (3.7%)

0.6162

Tumor location

  

0.002

Intratentorial

251 (23.4%)

33 (40.7%)

 

Supratentorial

789 (73.6%)

48 (59.3%)

 

Midline shift (> 3 mm) on MRI

186 (17.4%)

21 (25.9%)

0.0736

Suspected diagnosis

  

0.0001

Meningeoma

364 (34.0%)

31 (38.3%)

 

Metastasis

188 (17.5%)

18 (22.2%)

 

LGG

74 (6.9%)

2 (2.5%);

 

HGG

39 (3.6%)

3 (3.7%)

 

GBM

164 (15.3%)

11 (13.6%)

 

Neuroma (including vestibular)

75 (7%)

0 (0.0%)

 

Haemangioblastoma

1 (0.09%)

0 (0.0%)

 

Other

167 (15.6%)

14 (17.3%)