Combined T2w volumetry, DW-MRI and DCE-MRI for response assessment after neo-adjuvant chemoradiation in locally advanced rectal cancer

Figures & data

Figure 1. cT3N2 rectal adenocarcinoma in a 60-year-old male patient. T2-weighted imaging (a), Ktrans map (b), b800 image (c) and apparent diffusion coefficient map (d).

Table I. Comparison of response groups.

	pCR (n = 6)	non-pCR (n = 49)	p-value	GR (n = 10)	non-GR (n = 45)	p-value
Relative ∆volume, % (range)	−76.0 (−83.2 – −63.3)	−53.2 (−88.7 – −14.6)	0.012	−76.0 (−88.7 – −59.3)	−52.4 (−86.8 – −14.6)	0.001
Relative ∆ADC, % (range)	48.0 (36.5–63.1)	26.2 (−1–71.8)	< 0.001	48.0 (36.5–71.8)	25.6 (−1–42.7)	< 0.001
Relative ∆Ktrans, % (range)	−34.1 (−43.0–1.5)	0.5 (−64.9–64.5)	0.015	−34.1 (−43.7–1.5)	0.8 (−64.9–64.5)	< 0.001

Median therapy induced change in volume, apparent diffusion coefficient (ADC) and Ktrans for the pathological complete response (pCR) versus the non-pCR group and for the pathological good response (GR) versus the non-GR group. Values reported are median values with range. ADC, apparent diffusion coefficient; CRT, chemoradiation.

Table II. Classification tables for the diagnosis of pathological complete response (pCR).

Model	Observed		AUC	Accuracy	PPV	NPV
	PCR (n = 6)	non-PCR (n = 49)
Predicted
∆volume
PCR	5	11	0.81	78%	31%	97%
non-PCR	1	38
∆ADC
PCR	5	2	0.93	95%	71%	98%
non-PCR	1	47
∆Ktrans
PCR	4	5	0.80	87%	44%	96%
non-PCR	2	44
Predicted
∆volume ∆ADC ∆Ktrans
PCR	4	2	0.95	93%	67%	96%
non-PCR	2	47
∆volume ∆ADC
PCR	4	2	0.95	93%	67%	96%
non-PCR	2	47
∆volume ∆Ktrans
PCR	3	3	0.90	89%	50%	94%
non-PCR	3	46
∆ADC ∆Ktrans
PCR	3	1	0.93	93%	75%	94%
non-PCR	3	48

Classification tables for the diagnosis of pathological complete response (pCR) for the different models. In the individual parameter models the optimal cut-off values for ∆volume, ∆ADC and ∆Ktrans were −74%, + 43% and -32%, respectively. ADC, apparent diffusion coefficient; AUC, area under the receiver operating curve; NPV, negative predictive value; pCR, pathological complete response; PPV, positive predictive value.

Table III. Classification tables for the diagnosis of pathological good response (GR).

Model	Observed		AUC	Accuracy	PPV	NPV
	GR (n = 10)	non-GR (n = 45)
Predicted
∆volume
GR	8	8	0.84	82%	50%	95%
non-GR	2	37
∆ADC
GR	8	1	0.97	95%	89%	96%
non-GR	2	44
∆Ktrans
GR	6	3	0.87	91%	67%	91%
non-GR	4	42
Predicted
∆volume ∆ADC ∆Ktrans
GR	9	1	0.99	98%	90%	98%
non-GR	1	44
∆volume ∆ADC
GR	9	1	0.99	98%	90%	98%
non-GR	1	44
∆volume ∆Ktrans
GR	5	1	0.93	89%	83%	90%
non-GR	5	44
∆ADC ∆Ktrans
GR	9	2	0.98	95%	82%	98%
non-GR	1	43

Classification tables for the diagnosis of pathological good response (GR) for the different models. In the individual parameter models the optimal cut-off value for ∆volume, ∆ADC and ∆Ktrans were −74%, + 43% and −32%, respectively. ADC, apparent diffusion coefficient; AUC, area under the receiver operating curve; NPV, negative predictive value; PPV, positive predictive value.

Figure 2. Calibration plots for the best discriminative models for pathological complete response (pCR, a) and pathological good response (GR, b). On the x-axis the true probability of favorable response is shown and on the y-axis the predicted probability. The patient group was divided in three subgroups based on the observed probability. The dashed line shows a perfect calibration. ADC, apparent diffusion coefficient.

Figure 3. ROC curves for the individual response parameters (a and c) and the multiparameter models (b and d) for pathological complete response (pCR, a and b) and pathological good response (GR, c and d).