Pre-treatment CT radiomics to predict 3-year overall survival following chemoradiotherapy of esophageal cancer

Figures & data

Table 1. Patient characteristics.

Characteristic	AMC n (%)	Maastro n (%)	Total n (%)
Age, median (range)	62 (37–82)	64.5 (41–77)	63 (37–82)
Gender
Male	130 (79)	62 (84)	192 (80)
Female	35 (21)	12 (16)	47 (20)
Histology
Adenocarcinoma	132 (80)	61 (82)	193 (81)
Squamous cell carcinoma	33 (20)	13 (18)	46 (19)
Tumor location
Proximal - Middle	23 (14)	6 (8)	29 (12)
Distal	114 (69)	39 (53)	153 (64)
GEJ - Cardia	28 (17)	29 (39)	57 (24)
cT-stage
T1	1 (1)	–	1 (1)
T2	29 (18)	17 (23)	46 (19)
T3	129 (78)	53 (72)	182 (76)
T4	1 (1)	2 (3)	3 (1)
Tx	5 (3)	2 (3)	7 (3)
cN-stage
N0	36 (22)	16 (22)	52 (22)
N1	74 (45)	34 (46)	108 (45)
N2	47 (28)	13 (18)	60 (25)
N3	5 (3)	10 (14)	15 (6)
Nx	3 (2)	1 (1)	4 (2)
cM-stage
M0	132 (80)	73 (99)	205 (86)
M1a	22 (13)	1 (1)	23 (10)
Mx	11 (7)	–	11 (3)
Pathologic response
TRG 1	38 (23)	15 (20)	53 (22)
TRG 2	28 (17)	16 (22)	44 (18)
TRG 3	60 (36)	14 (19)	74 (31)
TRG 4	27 (16)	18 (24)	45 (19)
TRG 5	12 (7)	6 (8)	18 (8)
TRG x	–	5 (7)	5 (2)

Figure 1. ROC-curves for the AMC cohort (training) and Maastro cohort (validation) for the random forest models based on (a) 40 radiomic features and (b) 6 clinical variables.

Figure 2. Survival curves of the (a) AMC cohort (training) and (b) Maastro cohort (validation). Division in risk groups was based on the predictions by the random forest model with radiomics features.

Figure 3. Survival curves of the (a) AMC cohort (training) and (b) Maastro cohort (validation). Division in risk groups was based on the predictions by the random forest model with clinical variables.

Figure 4. Survival curves of the (a) AMC cohort (training) and (b) Maastro cohort (validation). Division in risk groups was based on the pathological response in the primary tumor.