Magnetic resonance imaging-based deep learning model for prediction of the neoadjuvant chemotherapy response and survival prognosis in adolescents with osteosarcoma

Table 1 Demographic characteristics of adolescents and young adults with osteosarcoma in the training and testing cohort, n (%)/median (interquartile range)

Characteristic	All subjects (n = 134)	Training cohort (n = 94)	Testing cohort (n = 40)	P value
Response to neoadjuvant chemotherapy				0.883
pGR	59 (44.0)	41 (43.6)	18 (45.0)
Non-pGR	75 (56.0)	53 (56.4)	22 (55.0)
Age, years	16 (3)	16 (3)	17 (4)	0.478
≤ 14	97 (72.4)	70 (74.5)	27 (67.5)	0.409
> 14	37 (27.6)	24 (25.5)	13 (32.5)
Gender				0.184
Female	62 (46.3)	47 (50.0)	15 (37.5)
Male	72 (53.7)	47 (50.0)	25 (62.5)
Tumor size, cm	7.3 (5.2)	7.3 (4.7)	8.1 (5.9)	0.662
≤ 10	98 (73.1)	71 (75.5)	27 (67.5)	0.337
> 10	36 (26.9)	23 (24.5)	13 (32.5)
Location				0.623
Extremity	78 (58.2)	56 (59.6)	22 (55.0)
Trunk	56 (41.8)	38 (40.4)	18 (45.0)
HGB1, g/L	133 (26)	133 (22)	130 (28)	0.399
Abnormal	41 (30.6)	26 (27.7)	15 (37.5)	0.258
Normal	93 (69.4)	68 (72.3)	25 (62.5)
Platelet, 109/L	211 (103)	210 (102)	219 (98)	0.846
Abnormal (> 300)	26 (19.4)	19 (20.2)	7 (17.5)	0.716
Normal (≤ 300)	108 (80.6)	75 (79.8)	33 (82.5)
WBC, 109/L	5.84 (2.34)	6.00 (2.30)	5.68 (2.69)	0.307
Abnormal (≤ 4)	20 (14.9)	13 (13.8)	7 (17.5)	0.585
Normal (> 4)	114 (85.1)	81 (86.2)	33 (82.5)
ALB, g/L	42.8 (6.2)	42.8 (5.8)	42.8 (7.0)	0.967
Abnormal (≤ 40)	40 (29.9)	26 (27.7)	14 (35.0)	0.395
Normal (> 40)	94 (70.1)	68 (72.3)	26 (65.0)
ALP, U/L	99 (104)	98 (101)	114 (114)	0.874
Abnormal (> 140)	42 (31.3)	29 (30.9)	13 (32.5)	0.851
Normal (≤ 140)	92 (68.7)	65 (69.1)	27 (67.5)
LDH, U/L	170 (77)	169 (77)	170 (77)	0.948
Abnormal (> 220)	35 (26.1)	25 (26.6)	10 (25.0)	0.847
Normal (≤ 220)	99 (73.9)	69 (73.4)	30 (75.0)
Duration of hospitalization, day	15 (12)	15 (12)	15 (13)	0.608
≤ 14	66 (49.3)	46 (48.9)	20 (50.0)	0.910
> 14	68 (50.7)	48 (51.1)	20 (50.0)
OS, month	28 (24)	30 (30)	26 (17)	0.192

¹For male, hemoglobin < 130 is defined abnormal; for female, hemoglobin < 115 is defined abnormal.

pGR: Pathological good response; HGB: Hemoglobin; WBC: White blood cell; ALB: Serum albumin; ALP: Alkaline phosphatase; LDH: Lactate dehydrogenase; OS: Overall survival.

Table 2 Predictive performance of radiological, clinical, and integrated nomogram models in prediction of response to neoadjuvant chemotherapy for adolescents and young adults with osteosarcoma in the training and testing cohorts

Model		MRI sequence	Training cohort						Testing cohort
			AUC	Accuracy	Sensitivity	Specificity	PPV	NPV	AUC	Accuracy	Sensitivity	Specificity	PPV	NPV
DL-based model	Xception	T1	0.983	94.68	96.23	92.68	94.44	95.00	0.604	62.50	72.73	50.00	64.00	60.00
		T2	0.944	88.30	83.02	95.12	95.65	81.25	0.508	52.50	22.73	88.89	71.43	48.48
		T1 + T2	0.981	94.68	92.45	97.56	98.00	90.91	0.770	77.50	77.27	77.78	80.95	73.68
	VGG16	T1	0.804	75.53	73.58	78.05	81.25	69.57	0.568	52.50	54.55	50.00	57.14	47.37
		T2	0.889	79.79	73.58	87.80	88.64	72.00	0.649	62.50	45.45	83.33	76.92	55.56
		T1 + T2	0.867	79.79	66.04	97.56	97.22	68.97	0.540	52.50	36.36	72.22	61.54	48.15
	VGG19	T1	0.887	78.72	69.81	90.24	90.24	69.81	0.626	57.50	45.45	72.22	66.67	52.00
		T2	0.862	78.72	84.91	70.73	78.95	78.38	0.545	60.00	86.36	27.78	59.38	62.50
		T1 + T2	0.887	78.72	69.81	90.24	90.24	69.81	0.626	57.50	45.45	72.22	66.67	52.00
	ResNet50	T1	0.909	80.85	83.02	78.05	83.02	78.05	0.649	60.00	59.09	61.11	65.00	55.00
		T2	0.912	81.91	75.47	90.24	90.91	74.00	0.639	60.00	63.64	55.56	63.64	55.56
		T1 + T2	0.948	86.17	83.02	90.24	91.67	80.43	0.770	65.00	54.55	77.78	75.00	58.33
	InceptionV3	T1	0.977	90.43	86.79	95.12	95.83	84.78	0.682	62.50	77.27	44.44	62.96	61.54
		T2	0.962	90.43	92.45	87.80	90.74	90.00	0.535	60.00	77.27	38.89	60.71	58.33
		T1 + T2	0.978	89.36	81.13	100.00	100.00	80.39	0.566	60.00	68.18	50.00	62.50	56.25
	InceptionResNetV2	T1	0.850	77.66	83.02	70.73	78.57	76.32	0.747	65.00	86.36	38.89	63.33	70.00
		T2	0.820	77.66	86.79	65.85	76.67	79.41	0.581	67.50	95.45	33.33	63.64	85.71
		T1 + T2	0.830	76.60	81.13	70.73	78.18	74.36	0.710	65.00	86.36	38.89	63.33	70.00
Handcrafted radiomics model		T1	0.897	80.85	77.36	85.37	87.23	74.47	0.639	60.00	45.45	77.78	71.43	53.85
		T2	0.903	80.85	77.36	85.37	87.23	74.47	0.660	60.00	40.91	83.33	75.00	53.57
		T1 + T2	0.891	82.98	81.13	85.37	87.76	77.78	0.588	55.00	50.00	61.11	61.11	50.00
Clinical model1		/	0.647	47.87	28.30	73.17	57.69	44.12	0.729	52.50	36.36	72.22	61.54	48.15
Nomogram model2		/	0.961	90.43	92.45	87.80	90.74	90.00	0.816	70.00	54.55	88.89	85.71	61.54

¹The clinical model was constructed by clinical variables selected by multivariate analysis with P value less than 0.05 in the training cohort.

²The nomogram model was constructed by significant clinical variables and the deep learning signature generated from the best deep learning magnetic resonance imaging (T1 + T2) model considering area under the receiver operating characteristic curve of the testing cohort among all models.

MRI: Magnetic resonance imaging; AUC: Area under the receiver operating characteristic curve; PPV: Positive predictive value; NPV: Negative predictive value; DL: Deep learning.

Table 3 Performance of the deep learning-based signature and integrated nomogram model in prediction of overall survival for adolescents and young adults with osteosarcoma in the training and testing cohorts

Model	Time	Training cohort		Testing cohort
		C-index (95%CI)	Brier score (95%CI)	C-index (95%CI)	Brier score (95%CI)
Nomogram model1	3-year	0.773 (0.638-0.908)	14.9 (9.3-20.5)	0.688 (0.421-0.955)	16.1 (2.9-29.4)
	5-year	0.855 (0.725-0.984)	16.6 (9.3-23.9)	0.831 (0.527-1.000)	20.8 (5.7-35.8)
DL-based signature2	3-year	0.773 (0.638-0.908)	14.9 (9.3-20.5)	0.688 (0.421-0.955)	17.0 (3.2-30.9)
	5-year	0.855 (0.725-0.984)	16.8 (9.4-24.3)	0.831 (0.527-1.000)	21.9 (6.4-37.3)

¹The nomogram model was constructed by prognostic clinical variables selected by multivariate Cox regression analysis on overall survival and the deep learning signature originated from diagnosis prediction.

²The deep learning signature was based on the best deep learning-radiological model considering AUC of the testing cohort from histological diagnosis identification.

C-index: Harrell’s concordance index; CI: Confidence interval; DL: Deep learning.