Table 5 Multivariable regression of case-level DLS score using clustering-derived features as input.

From: Interpretable survival prediction for colorectal cancer using deep learning

Feature #

Description

Validation Set 1

Validation Set 2

Coefficient

p

R2

Coefficient

p

R2

72

Small clusters of moderate to high-grade tumor cells intermixed with substantial adipose and a minor component of desmoplastic stroma

0.2269

<0.001

0.57

0.2913

<0.001

0.61

139

Low-intermediate grade tumor with predominant stroma of mature and intermediate desmoplasia

0.1977

<0.001

0.1650

<0.001

23

Small clusters of high-grade tumor cells with predominant, mature desmoplasia and moderate TILs

0.1096

<0.001

0.1931

<0.001

96

Small clusters of high-grade tumor cells, including single tumor cells, and a moderate amount of mature and intermediate desmoplasia

0.1031

<0.001

0.1996

<0.001

146

Low-grade tumor with moderate differentiation and desmoplastic stroma with mature desmoplasia and occasional TILs

−0.1248

<0.001

−0.2133

<0.001

122

Out of focus regions; predominantly low-grade tumor with tubule formation.

−0.1323

<0.001

−0.2867

0.187

104

Low and Intermediate grade tumor with tubule formation and small, solid regions; Stroma with mature desmoplasia

−0.1461

<0.001

−0.0505

<0.001

44

Intermediate grade tumor with irregular tubule formation; mature desmoplasia and focal areas of TILs

−0.1510

<0.001

−0.1081

<0.001

101

Predominantly intermediate grade tumor with irregular tubule formation; minor component of mature, desmoplasia

−0.2312

<0.001

−0.0420

0.313

144

Low-grade tumor with tubule formation and minor component of mixed stroma containing mature and intermediate desmoplasia with occasional, moderate TILs

−0.3476

<0.001

−0.3794

<0.001

Intercept

N/A

0.1256

0.002

0.1996

<0.001

Adding remaining 190 features

N/A

N/A

0.73

N/A

N/A

0.80

  1. For the overall model, p < 0.001 (t test). Each coefficient represents the relative increase of the DLS score associated with that variable. Bold indicates statistically significant input variables (p < 0.05).