Table 5 Accuracy metrics for combinational models in distinguishing malignant from benign tumors.

From: A Diagnostic Algorithm using Multi-parametric MRI to Differentiate Benign from Malignant Myometrial Tumors: Machine-Learning Method

Model No.	Model Category	No. of features	AUC	Sen (%)	Spe (%)	NPV (%)	PPV (%)	Accuracy (%)
	Simple Models
A: _Original [1]	Restriction + CN + T2 + T1	4	0.96	100	95	100	84	96.2
2	Original - T2	3	0.96	100	95	100	84	96.2 → ^¤
3	Original - T1	3	0.96	100	95	100	84	96.2→
4	Original - CN	3	0.95	100	94	100	81	95.2↓
5	Original - Restriction	3	0.89	48	99	88	91	88.6↓
*6	Original - (T2 + T1)	2	0.96	100	95	100	84	96.2 →
→The best Model = [Restriction + CN]
	Complex Models
B: _Original [7]	Qualitative Features:Restriction+ CN + Lipid peak +Choline peak+T2 + T1	6	0.96	95	98	99	91	97.1
8	Original - (T2 signal + T1 signal)	4	0.96	95	98	99	91	97.1→
9	Original - (Lipid peak + Choline peak)	4	0.96	100	95	100	84	96.2↓
*10	Original - (Choline peak + T2 signal + T1 signal)	3	0.97	95	99	99	95	98.1↑
11	Original - (Lipid peak + T2 signal + T1 signal)	3	0.96	95	98	99	91	97.1→
12	Original - (Lipid peak + Choline peak + T2 signal + T1 signal)	2	0.96	100	95	100	84	96.2↓
13	Original - (CN + Restriction + T2 signal + T1 signal)	2	0.69	46	92	87	60	82.5↓
14	Original - (CN + T2 signal + T1 signal)	3	0.96	100	95	100	84	96.2↓
15	Original - (CN + Choline peak + T2 signal + T1 signal)	2	0.96	100	95	100	84	96.2↓
16	Original - (Restriction + Choline peak + T2 signal + T1 signal)	2	0.84	62	93	91	68	86.7↓
→The best Model = [Restriction + CN + Lipid peak]
C: _Original [17]	Quantitative Features:T2 map + Mean ADC + T2 scaled + 4 TM/TP ratios	7	0.85	76	94	94	76	90.5
18	Original - (T2 map + mean ADC + T2 scaled)	4	0.86	81	87	95	61	85.7↓
19	Original - (TMCE Ratio + TP Ratio + TM Ratio + TPCE Ratio)	3	0.92	86	94	96	78	92.4↑
20	Original - (TP Ratio + TM Ratio + TPCE Ratio)	4	0.92	86	95	96	82	93.3↑
21	Original - (TMCE Ratio + TM Ratio + TPCE Ratio)	4	0.89	86	93	96	75	91.4↑
22	Original - (TMCE Ratio + TP Ratio + TPCE Ratio)	4	0.92	81	94	95	77	91.4↑
23	Original - (TMCE Ratio + TM Ratio + TP Ratio)	4	0.88	76	95	94	80	91.4↑
*24	Original - (T2 scaled + TP Ratio + TM Ratio + TPCE Ratio)	3	0.89	81	98	95	89	94.3↑
25	Original - (Mean ADC + TP Ratio + TM Ratio + TPCE Ratio)	3	0.89	81	95	95	81	92.4↑
26	Original - (T2 map + TP Ratio + TM Ratio + TPCE Ratio)	3	0.94	90	94	98	79	93.3↑
27	Original - (TMCE Ratio + T2 scaled + TP Ratio + TM Ratio + TPCE Ratio)	2	0.91	76	95	94	80	91.4↑
28	Original - (Mean ADC + T2 scaled + TP Ratio + TM Ratio + TPCE Ratio)	2	0.85	76	96	94	84	92.4↑
29	Original - (T2 map + T2 scaled + TP Ratio + TM Ratio + TPCE Ratio)	2	0.86	86	95	96	82	93.3↑
→The best Model = [T2 map + Mean ADC + TMCE Ratio]
D: _Original [30]	Total: Restriction+ T2map+ mean ADC + TMCE Ratio+ CN + Lipid peak	6	0.98	95	100	99	100	99.0
31	Original - (Lipid peak)	5	0.98	95	100	99	100	99.0→
32	Original - (CN + Lipid peak)	4	0.98	95	100	99	100	99.0→
33	Original - (TMCE Ratio + Lipid peak)	4	0.98	95	100	99	100	99.0→
*34	Original - (Mean ADC + Lipid peak)	4	1	100	100	100	100	100↑
35	Original - (T2map + Lipid peak)	4	0.96	95	96	99	87	96.2↓
36	Original - (Restriction + Lipid peak)	4	0.87	81	96	95	85	93.3↓
37	Mean ADC + Lipid peak	2	0.93	62	100	91	100	92.1↓
38	Mean ADC + (TM Ratio or TMCE Ratio)	2	0.86	86	95	96	82	93.3↓
39	Restriction + Mean ADC + T2 signal	2	0.95	95	96	99	87	96.2↓
→The best Model = [Restriction + T2 map + TMCE Ratio + CN]

¤Arrows indicate the direction of changes. *indicates the best models in each category. No.:number; AUC: Area Under receiver operating Characteristics; Sen: sensitivity; Spe: specificity; NPV: Negative Predictive Value; PPV: Positive Predictive Value; CN: Central Necrosis; TP Ratio: Tumor/Psoas Ratio on T2; TM Ratio: Tumor/Myometrium Ratio on T2; TPCE Ratio: Tumor/Psoas Ratio on Contrast Enhanced images; TMCE Ratio: Tumor/Myometrium Ratio on Contrast Enhanced images.

Back to article page

Table 5 Accuracy metrics for combinational models in distinguishing malignant from benign tumors.

Search

Quick links