Preoperative prediction of macrotrabecular-massive hepatocellular carcinoma through dynamic contrast-enhanced magnetic resonance imaging-based radiomics

Table 1 Clinical and pathological features in the training and test sets

Characteristics	Training set (n = 162)			Test set (n = 70)			P value
	nMTM-HCC (n = 118)	MTM-HCC (n = 44)	P value	nMTM-HCC (n = 51)	MTM-HCC (n = 19)	P value
Age, mean ± SD	60.6 ± 10.9	52.4 ± 9.9	< 0.001	60.7 ± 10.4	54.7 ± 9.7	0.034	0.647
Sex (men, %)	102 (86.4)	42 (95.5)	0.179	39 (76.5)	18 (94.7)	0.161	0.125
HBsAg (positive, %)	91 (77.1)	33 (75.0)	0.777	38 (74.5)	14 (73.7)	0.944	0.712
AFP > 400 µg/L, n (%)	18 (15.3)	31 (70.5)	< 0.001	12 (23.5)	9 (47.4)	0.053	0.970
CEA (ug/L)	2.6 (2.1)	2.7 (1.9)	0.719	2.2 (2.0)	2.2 (1.2)	0.543	0.332
CA19-9 (U/mL)	15.0 (19.3)	28.0 (25.4)	0.021	16.7 (17.1)	16.0 (34.4)	0.687	0.770
PLT (× 109/L)	135.0 (83.0)	174.0 (88.5)	0.027	135.0 (57.0)	175.0 (127.0)	0.001	0.970
TBIL (µmol/L)	16.6 (10.8)	20.7 (9.8)	0.067	15.7 (10.3)	14.4 (7.6)	0.916	0.509
DBIL (µmol/L)	3.8 (2.6)	4.8 (2.9)	0.060	3.2 (2.1)	3.3 (2.7)	0.620	0.419
IBIL (µmol/L)	12.5 (7.0)	15.0 (7.4)	0.136	12.2 (7.8)	10.9 (4.5)	0.712	0.061
Albumin, mean ± SD	37.8 ± 5.0	37.0 ± 5.8	0.387	37.8 ± 5.0	38.3 ± 5.1	0.715	0.647
AST (U/L)	32.0 (23.5)	43.0 (68.8)	0.004	33.0 (29.0)	43.0 (61.0)	0.135	0.980
ALT (U/L)	31.5 (34.0)	46.0 (64.0)	0.023	28.0 (35.0)	57.0 (96.0)	0.074	0.748
AST/ALT	1.1 (0.6)	1.1 (0.5)	0.702	1.3 (0.5)	1.1 (0.7)	0.207	0.842
Edmondson-Steiner grade (III-IV, %)	38 (32.2)	18 (40.9)	0.300	14 (27.5)	10 (52.6)	0.048	0.967
Microvascular invasion, n (%)	56 (47.5)	20 (45.5)	0.820	18 (35.3)	10 (52.6)	0.188	0.331
Satellite nodules, n (%)	9 (7.6)	5 (11.4)	0.452	1 (2.0)	1 (5.3)	0.461	0.110
Biliary invasion, n (%)	4 (3.4)	1 (2.3)	0.715	2 (3.9)	1 (5.3)	0.805	0.646

Data in parentheses are presented as median (interquartile range) or n (%). MTM-HCC: Macrotrabecular-massive hepatocellular carcinoma; nMTM-HCC: Non-macrotrabecular-massive hepatocellular carcinoma; HBsAg: Hepatitis B surface antigen; AFP: Alpha-fetoprotein; CEA: Carcinoembryonic antigen; CA19-9: Carbohydrate antigen 19-9; PLT: Platelet count; TBIL: Total bilirubin; DBIL: Direct bilirubin; IBIL: Indirect bilirubin; AST: Aspartate transaminase; ALT: Alanine transferase; SD: Standard deviation.

Table 2 Radiological features of patients in the training and test sets

Radiological features	Training set (n = 162)			Test set (n = 70)			P value
	nMTM-HCC (n = 118)	MTM-HCC (n = 44)	P value	nMTM-HCC (n = 51)	MTM-HCC (n = 19)	P value
Liver cirrhosis (positive, %)	69 (58.5)	29 (65.9)	0.389	34 (66.7)	15 (78.9)	0.319	0.168
Tumour size > 5 cm, n (%)	29 (24.6)	22 (50.0)	0.002	12 (23.5)	10 (52.6)	0.020	0.994
Tumour shape (irregular, %)	25 (21.2)	17 (38.6)	0.024	17 (33.3)	7 (36.8)	0.783	0.195
Intratumor fat, n (%)	24 (20.3)	10 (22.7)	0.740	11 (21.6)	1 (5.3)	0.210	0.500
Intratumor necrosis, n (%)	34 (28.8)	23 (52.3)	0.005	13 (25.5)	8 (42.1)	0.177	0.443
Intratumor hemorrhage, n (%)	24 (20.3)	17 (38.6)	0.017	7 (13.7)	8 (42.1)	0.025	0.526
Enhancing capsule, n (%)	85 (72.0)	35 (79.5)	0.332	39 (76.5)	15 (78.9)	0.826	0.620
Tumour-to-liver ADC ratio	0.9 (0.2)	0.8 (0.3)	0.018	0.9 (0.1)	0.8 (0.2)	0.045	0.312

Data in parentheses are presented as median (interquartile range) or n (%). MTM-HCC: Macrotrabecular-massive hepatocellular carcinoma; nMTM-HCC: Non-macrotrabecular-massive hepatocellular carcinoma; ADC: Apparent diffusion coefficient.

Table 3 Results of univariate and multivariate logistic regression analyses

Variables	Univariate logistic regression		Multivariate logistic regression
	OR (95%CI)	P value	OR (95%CI)	P value
Clinical features
Age	0.930 (0.896, 0.964)	< 0.001	0.956 (0.918, 0.997)	0.034
AFP > 400 µg/L	13.248 (5.839, 30.058)	< 0.001	10.066 (4.304, 23.541)	< 0.001
PLT	1.006 (1.002, 1.011)	0.009	NA	NA
Radiological features
Tumour size > 5 cm	3.069 (1.487, 6.333)	0.002	3.316 (1.579, 6.962)	0.002
Tumour shape	2.342 (1.106, 4.961)	0.026	NA	NA
Intratumor necrosis	2.706 (1.326, 5.521)	0.006	NA	NA
Intratumor hemorrhage	2.466 (1.160, 5.244)	0.019	NA	NA
Tumour-to-liver ADC ratio	0.183 (0.035, 0.972)	0.046	0.156 (0.027, 0.894)	0.037
Radiomics
Rad-score	2.718 (1.809, 4.084)	< 0.001	2.923 (1.740, 4.911)	< 0.001

AFP: Alpha-fetoprotein; PLT: Platelet count; ADC: Apparent diffusion coefficient; OR: Odds ratio; CI: Confidence interval; NA: Not available.

Table 4 Predictive performance of different models in training and test sets

Models	Training set				Test set
	AUC (95%CI)	Accuracy	Sensitivity	Specificity	AUC (95%CI)	Accuracy	Sensitivity	Specificity
Clinical	0.836 (0.773-0.888)	0.802	0.750	0.822	0.701 (0.575-0.814)	0.671	0.579	0.706
Radiological	0.688 (0.604-0.769)	0.685	0.636	0.703	0.723 (0.610-0.829)	0.700	0.632	0.725
Radiomics	0.766 (0.692-0.836)	0.772	0.568	0.847	0.739 (0.634-0.837)	0.743	0.579	0.804
Clinical-radiomics	0.888 (0.835-0.934)	0.802	0.909	0.763	0.793 (0.682-0.893)	0.686	0.737	0.667
Radiological-radiomics	0.796 (0.725-0.858)	0.772	0.636	0.822	0.764 (0.661-0.859)	0.729	0.632	0.765
Combined	0.896 (0.847-0.939)	0.796	0.932	0.746	0.805 (0.704, 0.895)	0.700	0.895	0.628

AUC: Area under the receiver operating characteristic curve; CI: Confidence interval.