Research Article

Diagnostic Accuracy of Machine Learning-Based Radiomics in Grading Gliomas: Systematic Review and Meta-Analysis

Table 5

Result of multiple subgroup analysis of machine learning-based radiomics for grading gliomas.

SubgroupStudy numberPatient numberSensitivitySpecificityPLRNLRDiagnostic odds ratio

All combined56290.96 (0.93–0.98)0.90 (0.85–0.93)9.53 (3.55–25.57)0.07 (0.02–0.20)153.85 (32.36–731.44)

Populations
>10025070.98 (0.95–0.99)0.90 (0.85–0.94)12.099 (1.37–107.12)0.03 (0.02–0.06)393.81 (80.89–1917.3)_
<10031220.88 (0.78–0.95)0.90 (0.77–0.96)7.89 (2.21–28.15)0.14 (0.05–0.39)65.13 (7.84–540.95)

Sequence
Single (CS or advanced)22620.96 (0.93–0.98)0.81 (0.72–0.88)4.61 (3.14–6.77)0.09 (0.02–0.44)66.75 (10.33–431.19)
Multiple (CS and advanced)33670.96 (0.91–0.99)0.97 (0.92–0.99)30.391 (11.585–79.726)0.04 (0.017–0.09)774.25 (202.54–2959.77)

Feature number
≥Sample size2820.85 (0.72–0.94)0.85 (0.69–0.95)5.71 (1.39–23.46)0.18 (0.07–0.52)33.76 (3.36–339.14)
<Sample size35470.97 (0.95–0.99)0.90 (0.85–0.94)13.48 (2.56–71.12)0.03 (0.02–0.06)369.98 (19.68–6956.0)

Training and testing set
Training set33310.94 (0.88–0.97)0.95 (0.89–0.98)12.91 (2.02–82.22)0.09 (0.02–0.47)154.56 (7.30–3276.9)
Training + testing set22980.97 (0.94–0.99)0.81 (0.72–0.89)5.32 (2.55–11.09)0.05 (0.1–0.22)176.99 (63.76–491.30)