Magnetic resonance imaging-based lymph node radiomics for predicting the metastasis of evaluable lymph nodes in rectal cancer

Table 1 Demographics of training and validation set

Characteristic	All patients	Training set	Validation set	P value
	n = 144	n = 100	n = 44
Age in yr	59 ± 10	59 ± 9	60 ± 10	0.514
Sex				0.805
Male	97 (67.4)	68 (68.0)	29 (65.9)
Female	47 (32.6)	32 (32.0)	15 (34.1)
CEA in ng/mL1		4.0 (2.4, 7.8)	3.8 (2.3, 9.3)	0.787
Location				0.165
Upper	31 (21.5)	22 (22.0)	9 (20.5)
Middle	66 (45.8)	41 (41.0)	25 (56.8)
Lower	47 (32.6)	37 (37.0)	10 (22.7)
Surgical approach				0.489
Dixon	120 (83.3)	82 (82.0)	38 (86.4)
Miles	21 (14.6)	15 (15.0)	6 (13.6)
Hartman	3 (2.1)	3 (3.0)	0 (0.0)
Surgical specimen histological type				0.877
Ulcerative	96 (66.7)	68 (68.0)	28 (63.6)
Infiltrative	6 (4.2)	4 (4.0)	2 (4.5)
Nodular	42 (29.2)	28 (28.0)	14 (31.8)
Differentiation				0.689
Well	8 (5.6)	5 (5.0)	3 (6.8)
Moderate	86 (59.7)	62 (62.0)	24 (54.5)
Poor	50 (34.7)	33 (33.0)	17 (38.6)
Pathological T stage				0.579
T1	7 (4.9)	4 (4.0)	3 (6.8)
T2	39 (27.1)	29 (29.0)	10 (22.7)
T3	84 (58.3)	59 (59.0)	25 (56.8)
T4	14 (9.7)	8 (8.0)	6 (13.6)
Pathological N status				0.293
Positive	53 (36.8)	34 (34.0)	19 (43.2)
Negative	91 (63.2)	66 (66.0)	25 (56.8)

¹Data are n (%) or median (IQR).

CEA: Carcinoembryonic antigen.

Table 2 Baseline characteristics of lymph nodes in the training and validation sets

Features	Training set			Validation set
	NLNM, n = 111	LNM, n = 78	P value	NLNM, n = 47	LNM, n = 34	P value
Size
Short axis1 in mm	6.10 (5.40, 7.20)	7.80 (6.60, 9.00)	< 0.001	5.70 (5.10, 7.40)	7.25 (6.57, 8.67)	< 0.001
Long-axis in mm	7.63 ± 1.72	9.13 ± 1.94	< 0.001	7.55 ± 1.79	8.67 ± 1.43	0.004
Shape			0.105			0.185
Non-round	82 (73.88)	29 (37.17)		31 (65.95)	27 (79.41)
Round	29 (26.12)	49 (62.83)		16 (34.04)	7 (20.59)
Margin			< 0.001			0.001
Clear	55 (49.54)	13 (16.67)		24 (51.00)	4 (14.70)
Unclear	56 (50.46)	65 (83.33)		23 (49.00)	30 (85.30)
T2WI heterogeneous signal			< 0.001			0.029
Absent	63 (56.76)	17 (21.79)		25 (53.19)	5 (14.71)
Present	48 (43.24)	61 (78.21)
Patterns of enhancement			< 0.001			< 0.001
Homogeneous	65 (58.55)	8 (10.26)		32 (68.08)	8 (23.52)
Heterogeneous	46 (41.45)	70 (89.74)		15 (31.92)	26 (76.48)
Radiomics score	-1.28 ± 1.28	0.62 ± 1.41	< 0.001	-1.29 ± 1.27	0.60 ± 1.60	< 0.001

¹Data are n (%) or median (IQR).

LNM: Lymph node metastasis; NLNM: Non-lymph node metastasis; T2WI: T2-weighted imaging.

Table 3 Risk factors of lymph node metastasis in lymph nodes

Intercept and variable	Conventional MRI model			Nomogram model
	Coefficient	Odds ratio (95%CI)	P value	Coefficient	Odds ratio (95%CI)	P value
Intercept	-6.429	0.002 (0.001, 0.012)	< 0.001	-4.270	0.014 (0.001, 0.113)	< 0.001
Short axis	0.478	1.612 (0.962, 2.745)	0.072	0.088	1.092 (0.650, 1.845)	0.739
Long-axis	0.009	1.010 (0.679, 1.520)	0.963	0.178	1.194 (0.794, 1.806)	0.394
Margin	1.431	4.184 (1.831, 10.105)	0.001	1.278	3.588 (1.592, 8.443)	0.003
T2WI heterogeneous signal	0.309	1.362 (0.586, 3.120)	0.467	0.556	1.744 (0.793, 3.856)	0.166
Patterns of enhancement	2.042	7.709 (3.200, 20.561)	< 0.001	1.477	4.380 (2.004, 9.983)	< 0.001
Radiomics score	NA	NA	NA	0.839	2.314 (1.795, 3.086)	< 0.001

Data are results of the multivariable regression analysis. The conventional magnetic resonance imaging model was developed based on independent factors of lymph node metastasis without the addition of radiomics signature. CI: Confidence interval; MRI: Magnetic resonance imaging; NA: Not available.

Table 4 Diagnostic performance of the conventional magnetic resonance imaging model, radiomics model, and nomogram

Data set	Model	AUC (95%CI)	Sensitivity, %	Specificity, %	Accuracy, %
Training cohort	Conventional MRI model	0.82 (0.76, 0.87)	75.0	80.0	77.8
	T1WI radiomics model	0.83 (0.79, 0.87)	59.5	82.6	73.5
	T2WI radiomics model	0.85 (0.79, 0.91)	63.1	93.0	79.0
	T1WI & T2WI Radiomics model	0.89 (0.84, 0.93)	74.3	86.1	81.5
	Nomogram model	0.92 (0.84, 0.99)	72.2	91.1	82.8
Validation cohort	Conventional MRI model	0.80 (0.76, 0.83)	71.0	78.8	75.7
	T1WI radiomics model	0.81 (0.75, 0.86)	60.5	83.7	72.8
	T2WI radiomics model	0.84 (0.80, 0.88)	62.2	87.0	77.2
	T1WI & T2WI Radiomics model	0.86 (0.79, 0.92)	65.8	90.7	79.0
	Nomogram model	0.91 (0.81, 0.96)	81.6	86.7	84.7

AUC: Area under the receiver operating characteristic curve; CI: Confidence interval; MRI: Magnetic resonance imaging; T1WI: T1-weighted imaging; T2WI: T2-weighted imaging.