Feasibility of multi-parametric magnetic resonance imaging combined with machine learning in the assessment of necrosis of osteosarcoma after neoadjuvant chemotherapy: a preliminary study

Table 4 Classification results using the random forest classifier

Classification Task	Features	Sen (%) [95% CI]	Spe (%) [95% CI]	Acc (%) [95% CI]	AUC [95% CI]	p-Value
TS vs. TN	rADC	82 [71 93]	69 [56 82]	75 [67 83]	0.83 [0.66 0.85]	0.0473
TS vs. TN	rADC, rT2WI, rST1WI	94 [87101]	78 [67 89]	85 [78 92]	0.90 [0.78 0.93]	0.0473
CTS vs. TN	rADC	68 [55 81]	57 [31 83]	66 [54 78]	0.61 [0.46 0.80]	0.0153
CTS vs. TN	rADC, rT2WI, rST1WI	66 [53 79]	92 [78106]	71 [60 z 82]	0.81 [0.68 0.91]	0.0153
NCTS vs. TN	rADC	88 [79 97]	89 [79 99]	89 [82 96]	0.93 [0.81 0.96]	0.0933
NCTS vs. TN	rADC, rT2WI, rST1WI	96 [91101]	92 [83101]	94 [89 99]	0.97 [0.88 1.00]	0.0933

TS tumour survival, CTS cartilaginous tumour survival, NCTS non-cartilaginous tumour survival, TN tumour nonviable, Sen sensitivity, Spe specificity, Acc accuracy; the P value represents the significance of the statistical comparisons of the AUCs of the different RF models (constructed with rADC vs. constructed with rADC, rT2WI, and rST1WI)

ISSN: 1471-2407