Hypovascular Cellular Tumor in Primary Central Nervous System Lymphoma is Associated with Treatment Resistance: Tumor Habitat Analysis Using Physiologic MRI

S.Y. Jeong; J.E. Park; N. Kim; H.S. Kim

doi:10.3174/ajnr.A7351

Article Figures & Data

Figures

Tables

Download figure
Open in new tab
Download powerpoint
FIG 1.
Flow chart of patient recruitment.
Download figure
Open in new tab
Download powerpoint
FIG 2.
The overall process of the deep learning segmentation and tumor habitat analysis. A, Image acquisition, registration, and deep learning segmentation for a contrast-enhancing lesion (CEL). B, Extraction of ADC and nCBV values from the CEL and voxel classifications based on ADC and nCBV values. The individual voxels in each cluster are grouped according to their similarities and differences using a K-means clustering algorithm. C, The cluster number is set to 3 to depict 3 different habitats according to the combinations of ADC and nCBV parameters: hypervascular cellular, hypovascular cellular, and hypovascular hypocellular. D, Voxels are shown as spatial habitats in the original image space. Associations of pretreatment tumor habitats with TTP were analyzed.
Download figure
Open in new tab
Download powerpoint
FIG 3.
A, Demonstration of the 3 spatial habitats defined by clustering of voxels using normalized ADC and nCBV maps in a 53-year-old male patient. The hypervascular cellular habitat (red) shows high nCBV and low ADC, the hypovascular cellular habitat (green) shows low nCBV and low ADC, and the hypovascular hypocellular habitat (blue) shows low nCBV and high ADC. The tumor exhibits a large hypovascular cellular habitat (green), and a persistent enhancing mass was associated with a short TTP after initial chemotherapy. B, Demonstration of the 3 spatial habitats defined by clustering of voxels using normalized ADC and nCBV maps in a 57-year-old male patient. The hypervascular cellular habitat (red) shows high nCBV and low ADC, the hypovascular cellular habitat (green) shows low nCBV and low ADC, and the hypovascular hypocellular habitat (blue) shows low nCBV and high ADC. The tumor has a small hypovascular cellular habitat (green) and showed a complete response at 53 days after initial chemotherapy.
Download figure
Open in new tab
Download powerpoint
FIG 4.
Kaplan-Meier analysis of TTP in patients with PCNSL stratified by hypovascular cellular habitat (log-rank, P = .01).

Tables

Figures

View popup

Table 1:

Patients and imaging characteristics of the PCNSLs included in this study

Clinical Characteristics (n = 81)
Mean age (yr)	61.6 (SD, 11.8)
Sex
Male/female	38/43 (46.9%/53.1%)
ECOG performance status at diagnosis (case No) (%)
1	71 (87.7%)
2	2 (2.5%)
3	6 (7.4%)
4	2 (2.5%)
Mean serum LDH level	225.9 (SD, 59.9)
Mean CSF-total protein	107.0 (SD, 103.0)
Initial treatment response (case No) (%)
Treatment response (CR and PR)	64 (79.0%)
Treatment failure (SD and PD)	17 (21.0%)
Imaging characteristics (case No) (%)
Location
Deep^{^a}	69 (85.2%)
Hemisphere	12 (14.8%)
Atypical findings^{^b} (case No) (%)
Positive	14 (17.3%)
Negative	67 (82.7%)

Note:—CR, complete response; PR, partial response; PD, progressive disease; SD, stable disease.
↵a Deep location: thalamus, basal ganglia, corpus callosum, periventricular area, cerebellum.
↵b Atypical findings: presence of hemorrhage or necrosis.

View popup

Table 2:

Clinical and imaging predictors associated with TTP in PCNSL

Variable	Hazard Ratio	95% CI	P Value
Age	0.98	0.94–1.03	.55
Atypical finding^{^a}	4.41	1.32–14.71	.016
CSF-total protein	1.00	0.99–1.00	.70
ECOG
2 (reference)
3 or 4	5.82	1.07–31.66	.04
Serum LDH level	1.01	1.00–1.02	.013
Deep location^{^b}	1.75	0.33–9.08	.50

↵a Atypical findings: presence of hemorrhage or necrosis.
↵b Deep location: thalamus, basal ganglia, corpus callosum, periventricular area, cerebellum.

View popup

Table 3:

Exploratory analysis of spatial habitats for predicting TTP in patients with PCNSL

Spatial Tumor Habitats	TTP
Spatial Tumor Habitats	Hazard Ratio^{^a}	95% CI	P Value
No. of voxels (20,000 voxels)
Hypervascular cellular habitat	1.39	0.29–6.44	.29
Hypovascular cellular habitat	2.83	1.20–6.65	.017
Hypovascular hypocelluar habitat	2.03	0.58–7.15	.27
Voxel fraction (%)
Hypervascular cellular habitat	0.88	0.04–20.97	.83
Hypovascular cellular habitat	2.07	0.29–14.83	.46
Hypovascular hypocelluar habitat	0.46	0.05–4.01	.48
ADC	0.98	0.95–1.01	.16
CBV	0.80	0.45–1.32	.38

↵a Hazard ratios reported here indicate the relative change in hazard that a 10-unit (20,000 voxels) increase in each imaging parameter incurs.

View popup

Table 4:

Prediction of TTP in PCNSLs according to tumor habitat, ADC, and CBV^{^a}

	Combination of Clinical, Conventional Imaging Predictors, and Tumor Habitats	Combined Clinical and Conventional Predictors	Tumor Habitats	Clinical Predictors Only	Conventional Imaging Predictors Only
C-index	0.73	0.71	0.65	0.68	0.63
95% CI	0.67–0.80	0.54–0.78	0.52–0.78	0.54–0.82	0.50–0.76
P value	Reference	.01	.012	.81	.62

↵a Combined clinical predictors were age, ECOG score, and mean serum LDH level; the conventional imaging predictor was the presence of atypical image findings. P value refers to the significance in the difference of the C-indices between the combined model and the single model assessed using “CompareC” (https://cran.r-project.org/web/packages/compareC/index.html) in the R statistical and computing software.