Table 3 Results and performances of the multivariable modeling involving the deep learning risk scores and the FNCLCC grade.

From: Prognostic prediction in soft-tissue sarcomas using deep learning and digital pathology of tumor and margin areas

	Predictors		HR	P value	c-index in Training	c-index in Validation-1	c-index in Validation-2
DL-C model
	Size (mm)		1.01 (1.01–1.02)	< 0.0001***	0.793 (0.732–0.854)	0.759 (0.643–0.876)	0.741 (0.662–0.821)
	Histologic type (ref: M/RC-LPS)
		Leiomyosarcoma	1.66 (0.42–6.59)	0.4679
		Liposarcoma	0.26 (0.06–1.15)	0.0760
		Myxofibrosarcoma	0.56 (0.13–2.35)	0.4298
		Other	1.41 (0.39–5.11)	0.5987
		Undifferentiated sarcoma	0.54 (0.15–1.98)	0.3547
	DL-C High risk		3.28 (1.56–6.89)	0.0018**
	Age (years)		1.03 (1–1.05)	0.0416*
	Chemotherapy (yes)		0.63 (0.24–1.67)	0.3583
	Radiotherapy (yes)		0.57 (0.28–1.16)	0.1201
DL-P model
	Size (mm)		1.02 (1.01–1.02)	< 0.0001***	0.851 (0.803–0.900)	0.74 (0.635–0.845)	0.667 (0.573–0.760)
	Histologic type (ref: M/RC-LPS)
		Leiomyosarcoma	0.43 (0.09–2.05)	0.2926
		Liposarcoma	0.1 (0.02–0.5)	0.0051*
		Myxofibrosarcoma	0.16 (0.03–0.8)	0.0253*
		Other	0.52 (0.13–2.2)	0.3769
		Undifferentiated sarcoma	0.12 (0.03–0.54)	0.0054*
	DL-P High risk		23.41 (6.98–78.5)	< 0.0001***
	Age (years)		1.01 (0.99–1.04)	0.2782
	Chemotherapy (yes)		0.62 (0.23–1.67)	0.3451
	Radiotherapy (yes)		0.52 (0.26–1.04)	0.0658
DL-R model
	Size (mm)		1.01 (1.01–1.02)	< 0.0001***	0.790 (0.733–0.847)	0.690 (0.564–0.817)	0.664 (0.567–0.760)
	Histologic type (ref: M/RC-LPS)
		Leiomyosarcoma	2.6 (0.67–10.03)	0.1656
		Liposarcoma	0.46 (0.12–1.83)	0.2695
		Myxofibrosarcoma	1.19 (0.31–4.53)	0.8036
		Other	1.95 (0.56–6.81)	0.2958
		Undifferentiated sarcoma	0.92 (0.27–3.12)	0.8904
	DL-R (binary)		3.74 (1.76–7.96)	0.0006***
	Age (years)		1.02 (1–1.04)	0.1094
	Chemotherapy (yes)		1.01 (0.39–2.63)	0.9865
	Radiotherapy (yes)		0.46 (0.22–0.95)	0.0347*
DL-CP model
	Size (mm)		1.01 (1.01–1.02)	< 0.0001***	0.819 (0.761–0.877)	0.74 (0.623–0.856)	0.698 (0.604–0.791)
	Histologic type (ref: M/RC-LPS)
		Leiomyosarcoma	0.57 (0.13–2.51)	0.4605
		Liposarcoma	0.22 (0.05–0.97)	0.0458*
		Myxofibrosarcoma	0.26 (0.06–1.17)	0.0793
		Other	0.87 (0.23–3.26)	0.8379
		Undifferentiated sarcoma	0.22 (0.06–0.89)	0.0342*
	DL-CP High risk		8.82 (3.39–22.96)	< 0.0001***
	Age (years)		1.03 (1.01–1.06)	0.0068*
	Chemotherapy (yes)		0.68 (0.25–1.8)	0.4324
	Radiotherapy (yes)		0.79 (0.38–1.64)	0.5224
DL-CPR model
	Size (mm)		1.01 (1.01–1.02)	< 0.0001***	0.804 (0.738–0.870)	0.786 (0.689–0.883)	0.698 (0.620–0.775)
	Histologic type (ref: M/RC-LPS)
		Leiomyosarcoma	1.76 (0.46–6.67)	0.4082
		Liposarcoma	0.55 (0.14–2.16)	0.3927
		Myxofibrosarcoma	0.69 (0.18–2.61)	0.5808
		Other	1.99 (0.56–7.06)	0.2860
		Undifferentiated sarcoma	0.58 (0.17–1.96)	0.3809
	DL-CPR High risk		6.86 (2.91–16.18)	< 0.0001***
	Age (years)		1.02 (1–1.05)	0.0693
	Chemotherapy (yes)		0.79 (0.3–2.06)	0.6271
	Radiotherapy (yes)		0.54 (0.27–1.07)	0.0758
Grade-based model
	Size (mm)		1.01 (1.01–1.02)	< 0.0001***	0.764 (0.699–0.828)	0.739 (0.621–0.856)	0.723 (0.652–0.794)
	Histologic type (ref: M/RC-LPS)
		Leiomyosarcoma	2.61 (0.71–9.65)	0.1493
		Liposarcoma	0.4 (0.1–1.69)	0.2136
		Myxofibrosarcoma	1.09 (0.28–4.28)	0.8981
		Other	2.21 (0.66–7.44)	0.2002
		Undifferentiated sarcoma	0.86 (0.25–2.96)	0.8093
	FNCLCC Grade (ref: grade I)		1.87 (0.59–5.96)	0.2908
	Age (years)		1.02 (1–1.05)	0.0444*
	Chemotherapy (yes)		0.71 (0.27–1.81)	0.4697
	Radiotherapy (yes)		0.61 (0.3–1.23)	0.1689

c-index concordance index, DL deep learning risk score (based on C: tumor centrum, P: tumor periphery and R: tumor surrounding tissues, and their combination), HR hazard ratio, M/RC-LPS myxoid/round cells liposarcoma, ref reference.
Hazard ratio and c-index are given with 95% confidence intervals.
*: P < 0.05, **: P < 0.005, ***: P < 0.001. Significant results are in bold.

Back to article page

Table 3 Results and performances of the multivariable modeling involving the deep learning risk scores and the FNCLCC grade.

Search

Quick links