Accuracy of apparent diffusion coefficients and enhancement ratios on magnetic resonance imaging in differentiating primary cerebral lymphomas from glioblastoma

Abstract

Background and purpose

This study aimed to determine the accuracy of apparent diffusion coefficient (ADC) and enhancement ratio (ER) in discriminating primary cerebral lymphomas (PCL) and glioblastomas.

Materials and methods

Circular regions of interest were randomly placed centrally within the largest solid-enhancing area of all lymphomas and glioblastomas on both post-contrast T1-weighted images and corresponding ADC maps. Regions of interest were also drawn in the contralateral hemisphere to obtain enhancement and ADC values of normal-appearing white matter. This helped us to calculate the ER and ADC ratio.

Results

Mean enhancement and ADC (mm²/s) values for PCL were 2220.56 ± 2948.30 and 712.00 ± 137.87, respectively. Mean enhancement and ADC values for glioblastoma were 1537.07 ± 1668.33 and 1037.93 ± 280.52, respectively. Differences in ADC values, ratios and ERs were all statistically significant between the two groups (p < 0.05). ADC values correctly predicted 71.4% of the lesions as glioblastoma and 83.3% as PCL (area under the curve (AUC) = 0.86 on receiver operating characteristic curve analysis). ADC ratios correctly predicted 85.7% of the lesions as glioblastoma and 100% as PCL (AUC = 0.93). ERs correctly predicted 71.4% of the lesions as glioblastoma and 88.9% as PCL (AUC = 0.92). The combination of ADC ratio and ER correctly predicted 100% tumour type in both patient subgroups.

Conclusions

ADC values, ADC ratios and ERs may serve as useful variables to distinguish PCL from glioblastoma. The combination of ADC ratio and ER yielded the best results in identification of both patient subgroups.

Introduction

The incidence of primary cerebral lymphoma (PCL) has seen an increase over the last three decades.^1–5 Currently, it occurs at the rate of 7 cases per 1,000,000 people in the USA.⁶ It is an uncommon form of extranodal non-Hodgkin's lymphoma (NHL) which occurs in both immunocompromised as well as immunocompetent patients.^2,3,7 According to the current literature, PCL represents about 4–7% of primary brain tumors.^1,5 Glioblastoma is the most aggressive primary malignant brain tumour in humans and also the most common,^3,4,7,8 accounting for about 80% of cases. It is hence a source of considerable morbidity and mortality.

It is imperative that we distinguish PCL from glioblastoma. This is due to the fact that both neoplasms have significantly differing management protocols.^1,4,8,9 Imaging findings suggestive of PCL warrant a neurosurgical consult for stereotactic biopsy. It is then managed with a combination of chemotherapy and whole-brain radiation therapy (WBRT).^2,10,11 Surgical resection has no role in management and should be reserved for rare occurrences of neurological deterioration secondary to herniation.² Glioblastoma, on the other hand, is treated via maximal surgical resection and adjuvant radiotherapy and/or chemotherapy with temozolomide.^9,10,12

Although there are certain characteristic features, it is often difficult to distinguish the two types of tumours on conventional magnetic resonance imaging (MRI)^1,5 because there is significant overlap between the findings.⁹ PCL in immunocompetent patients appear homogenously enhanced on post-contrast T1-weighted MR images (T1WI). Glioblastoma, however, only shows ring or ring-like or partial enhancement on post-contrast T1WI. This is due to the presence of necrosis in typical cases of glioblastomas. Confusion arises when trying to distinguish between atypical cases of glioblastomas (which do not contain necrosis) and PCL, as well as between glioblastomas and atypical cases of PCL (which contain necrosis).^9,12,13

Over the past two decades, diffusion-weighted imaging (DWI) has developed into an important modality for preoperative differentiation between the two pathologies. Studies have shown significant differences between the apparent diffusion coefficients (ADC) calculated from DWI of PCL and glioblastoma.^5,14–19 The ADC value in DWI has shown to be inversely related to tumour cellularity.¹⁶ The cellularity of lymphomas has been shown to be higher than that of glioblastoma. Therefore, the ADC values of lymphomas are significantly less than those of glioblastoma.

Considering the high cellularity of lymphoma, we suppose that the percentage of enhancement within PCL compared to normal-appearing white matter (NAWM) will be considerably more compared to glioblastoma. Therefore, we hypothesised that by calculating the enhancement ratio (ER; calculated on post-contrast T1WI), we can develop a quick and easy tool to discriminate between the two tumours in the absence of a perfusion study.⁹

We hence conducted a similar experiment within our cohort of patients. We intended to determine the accuracy of ADC and ER in discriminating PCL and glioblastoma in patients with histopathology-confirmed disease. To the best of our knowledge, this is the first-time ERs have been compared between the two tumours.

Materials and methods

This retrospective study was conducted at the Aga Khan University Hospital in Karachi, Pakistan, after undergoing review and acceptance by the institute's ethical review committee. The study duration was around 35 months (from January 2014 to December 2016). Our study sample included all patients with proven PCL and glioblastoma who underwent a MRI scan and biopsy/surgery along with histopathology evaluation from our hospital. Those patients who lacked proper preoperative imaging, had previously taken corticosteroids, had undergone radiation/chemotherapy, had a previous brain biopsy at the time of MRI or had an inconclusive histopathology result were excluded from the final study sample.

Using these criteria, total of 21 patients were included, 11 of whom had glioblastoma and 10 PCL. Six patients were excluded from the study (two patients had received steroids prior to baseline imaging, one had recurrence of glioblastoma after initial treatment, and three patients had an inconclusive biopsy). Eight of our patients had multifocal lesions. For the sake of simplicity, each lesion was considered and analysed separately. This gave us a total of 32 lesions (14 glioblastoma and 18 PCL).

MRI

All MRI scans were performed with a 1.5 T machine (Magnetom Avanto; Siemens) and a 3T machine (Toshiba Vantage Titan) using a head coil. Standardised departmental protocol was observed in all participants. DWI and ADC maps (mm²/s) were obtained (FOV 230; section thickness: 5 mm; TR: 3600; TE: 115; and b-values of 0, 500 and 1000 s/mm²). In addition, sagittal T2W, axial T1W and T2W coronal FLAIR sequences pre and post contrast (0.1 mmol/kg of body weight of gadopentetate dimeglumine; Magnevist) and sagittal and axial T1-weighted post-contrast images were acquired with same section thickness. The contrast injection rate was constant in each patient.

Enhancement and ADC measurement

A neuroradiologist (F.M.) with seven years' experience and a neuroimaging fellow (S.S.M.A.) reviewed images on a diagnostic workstation. Qualitative assessment was performed on conventional sequences, including parameters such as multifocality, enhancement pattern, neo-vascularity on SWI and perifocal oedema. Simultaneous quantitative assessment of contrast images and ADC maps of all patients (lymphomas and glioblastomas) was undertaken by drawing regions of interest (ROIs) on ADC maps and post-contrast T1 images with consensus. In cases of a difference of opinion, the final judgement was made by the neuroradiologist (F.M.) based on experience. Quantitative values were obtained by each reader. The readers were blinded to the biopsy reports.

Post-contrast T1WI and ADC maps were manually linked together for adequate registration. Circular ROIs with diameters ranging from 0.8 to 1.3 cm were randomly placed centrally within the largest solid-enhancing area of all lymphomas and glioblastomas (Figure 1) on both post-contrast T1WI and corresponding ADC maps in order to avoid volume averaging with cystic or necrotic regions that might influence enhancement and ADC values. Cases with multifocal disease were individually quantified based on the same parameters.

Figure 1.

Regions of interest (ROIs) placed on contralateral normal-appearing white matter on post-contrast T1 weighted imaging and corresponding apparent diffusion coefficient (ADC) maps in case of primary cerebral lymphoma (upper row) and glioblastoma (lower row) to calculate enhancement ratios (ERs) and ADC ratios, respectively.

Approximate same-size ROIs were also drawn in matching structures in the contralateral hemisphere in each patient to obtain enhancement value and ADC values of NAWM for the purpose of normalisation (Figure 1). Measurement of normal contralateral white matter helped us to calculate the ER and ADC ratio.

The data acquired were stratified into two groups: glioblastoma and PCL. Mean enhancement and ADC values for each group were calculated. We then proceeded to calculate the ER and ADC ratio for each lesion. This was done by dividing the enhancement and ADC value in the solid enhancing areas of the tumours by the enhancement and ADC values of NAWM on the contralateral side. Means were also calculated for ERs and ADC ratios.

Statistical analysis

All data collected were manually entered and edited using SPSS Statistics for Windows v21 (IBM Corp.) for subsequent analysis. Double entry and validation were done to avoid data-entry errors. p-Values of < 0.05 were considered to indicate a statistically significant difference.

The normality of the data was evaluated using the Shapiro–Wilk test and a visual inspection of histograms, Q-Q plots and box plots. A significant number of our parameters were non-normally distributed. Therefore, non-parametric tests were used for analysis. The Mann–Whitney U-test was used to compare ADC values, ERs and ADC ratios between the two groups of disease. Univariate logistic regression analysis was performed to assess the diagnostic value of each parameter of interest (ADC value, ADC ratio and ER). Each parameter's ability to distinguish between PCL and glioblastoma was quantified by the area under the curve (AUC) on receiver operating characteristic (ROC) curve analysis. The AUC value serves as an index of overall discriminative ability of a given method. Cut-off values for the above variables were also calculated using ROC curve analysis. Next, multivariate logistic regression models using two or three of these imaging parameters were generated.

Results

Table 1 summarises patient demographics and tumour focality. Our sample consisted of a total of 21 patients: 11 (52.4%) males, with an average age of 49.45 ± 10.60 years, and 10 (47.6%) females, with an average age of 59.60 ± 11.03 years. There were 11 (52.4%) patients with glioblastoma and 10 (47.6%) with PCL.

Table 2.

Differences in ADC values, ADC ratios and ERs between glioblastoma and PCL.

	Glioblastoma	PCL	p-Values
ADC value	1037.93 ± 280.52	712.00 ± 137.87	<0.01
ADC ratio	1.30 ± 0.30	0.82 ± 0.12	<0.01
ER	1.29 ± 0.33	1.94 ± 0.40	<0.01

ADC: apparent diffusion coefficient; PCL: primary cerebral lymphoma; ER: enhancement ratio.

Table 1.

Patient sample demographics stratified according to tumour type.

	Type of tumour
	Glioblastoma (n = 11)		PCL (n = 10)		p-Value
Sex	8 (72.7%) males	3 (27.3%) females	3 (30%) males	7 (70%) females	0.05
Age	52.0 ± 8.88		56.8 ± 14.33		>0.05
Focality	10 (90.9) single	1 (9.1) multifocal	3 (30%) single	7 (70%) multifocal	<0.01

Of the patients with glioblastoma, eight (72.7%) were male and three (27.3%) were female. Of the patients with PCL, three (30%) were male and seven (70%) were female. The relationship between the tumour type and the sex of the patient was not proven to be statistically significant (p = 0.05).

Patients with glioblastoma had an average age of 52.0 ± 8.88 years. Patients with PCL had an average age of 56.8 ± 14.33 years. These differences were not statistically significant (p > 0.05).

Patients with PCL had more multifocal lesions (70%) compared to patients with glioblastoma of whom only one (9.1%) patient had a multifocal lesion. This was statistically significant (p < 0.01)

Comparing enhancement of PCL lesions to contralateral NAWM

The mean enhancement of the PCL lesions was 2220.56 ± 2948.30. The mean enhancement of the contralateral NAWM was 994.39 ± 1176.81. This was statistically significant (p < 0.01).

Comparing enhancement of glioblastoma lesions to contralateral NAWM

The mean enhancement of the glioblastoma lesions was 1537.07 ± 1668.33. The mean enhancement of the contralateral NAWM was 1261.50 ± 1412.69. This was not statistically significant (p = 0.49).

Comparing ADC value of PCL lesions to contralateral NAWM

The mean ADC value of PCL lesions was 712.00 ± 137.87 (Table 2). The mean ADC value of contralateral NAWM was 870.83 ± 118.52. This was statistically significant (p < 0.01).

Comparing ADC value of glioblastoma lesions to contralateral NAWM

The mean ADC value of glioblastoma lesions was 1037.93 ± 280.52 (Table 2). The mean ADC value of contralateral NAWM was 800.43 ± 86.78. This was statistically significant (p < 0.01).

Comparisons between PCL and glioblastoma

The differences in mean enhancement between PCL and glioblastoma were not statistically significant (p = 0.12), whereas differences in ADC values were statistically significant (p < 0.01). The mean ER of PCL lesions was 1.94 ± 0.40, whereas that of glioblastoma lesions was 1.29 ± 0.33. This was statistically significant (p < 0.01). The mean ADC ratio of PCL lesions was 0.82 ± 0.12, whereas that of glioblastoma lesions was 1.30 ± 0.30. This was also statistically significant (p < 0.01).

ADC values correctly predicted 71.4% (10/14) of the glioblastoma lesions and 83.3% (15/18) of the PCL lesions. This corresponded to an AUC of 0.86 on ROC curve analysis (Figure 2). Optimal ADC value to differentiate PCL from glioblastoma should be <820.50. This gives us a sensitivity of 83.3% and a specificity of 85.70%.

Figure 2.

Receiver operating characteristic (ROC) curve for ADC values. Area under the curve (AUC) is 0.857.

ADC ratios correctly predicted 85.7% (12/14) of the glioblastoma lesions and 100% (18/18) of the PCL lesions. This corresponded to an AUC of 0.93 on ROC curve analysis (Figure 3). The optimal ADC ratio to differentiate PCL from glioblastoma should be <1.05. This gives us a sensitivity of 100% and a specificity of 85.70%.

Figure 3.

ROC curve for ADC ratios. AUC is 0.931.

The ER correctly predicted 71.4% (10/14) of the glioblastoma lesions and 88.9% (16/18) of the PCL lesions. This corresponded to an AUC of 0.92 on ROC curve analysis (Figure 4). The optimal ER to differentiate PCL from glioblastoma should be >1.66. This gives us a sensitivity of 83.3% and a specificity of 85.70%.

Figure 4.

ROC curve for ERs. AUC is 0.917.

We then performed a multivariate logistic regression to assess the influence of combining different parameters. ADC value, ADC ratio and ER were the variables selected for multivariate assessment. Evaluation using a two-variable model increased the probability of differentiating between the two tumour types. The combination of ADC value and ADC ratio correctly identified 92.9% (13/14) of the glioblastoma lesions and 94.4% (17/18) of the PCL lesions (AUC = 0.95; see Figure 5). The combination of ADC value and ER correctly identified 85.7% (12/14) of the glioblastoma lesions and 100% (18/18) of the PCL lesions (AUC = 0.96; see Figure 6). The combination of ADC ratio and ER yielded the best results, with 100% correct identification of both tumour subgroups (AUC = 1.0). These results are summarised in Table 3.

Figure 5.

ROC curve for ADC value + ADC ratio. AUC is 0.95.

Figure 6.

ROC curve for ADC value + ER. AUC is 0.96.

Table 3.

Univariate and multivariate models for differentiation of glioblastoma and PCL.

Model 1 parameters	Glioblastoma	PCL	AUC	Model 2 parameters	Glioblastoma	PCL	AUC
ADC values	71.4% (10/14)	83.3% (15/18)	0.86	ADC value + ADC ratio	13/14 (92.9%)	17/18 (94.4%)	0.95
ADC ratio	85.7% (12/14)	100% (18/18)	0.93	ADC value + ER	12/14 (85.7%)	18/18 (100%)	0.96
ER	71.4% (10/14)	88.9% (16/18)	0.92	ER + ADC ratio	14/14 (100%)	18/18 (100%)	1.0

Discussion

This study found that patients with PCL exhibit lower ADC values and greater enhancement than patients with glioblastoma. Our results demonstrate that ADC values and enhancement obtained in solid enhancing areas of glioblastoma and PCL, as well as that of contralateral NAWM, correlate well with previous reports.^5,10,15–17 Although we were unable to show a statistically significant difference in ERs between the two patient populations, a statistically significant difference (p < 0.05) in ADC values was obtained. Previous studies have shown that ADC values can be used to differentiate between PCL and glioblastomas, and our findings support this.¹⁶

Lymphomas appear relatively hyper-intense to grey matter on DWI and isointense to hypo-intense on ADC maps. These are features consistent with restricted water diffusion. Glioblastomas, on the other hand, appear hyper-intense on both DWIs and ADC maps, signifying increased diffusion.⁵ Abul-Kasim et al. have previously worked on perfusion-weighted MRI as a distinguishing modality of primary central nervous system lymphoma from other homogenously enhancing brain tumours. They found that relative regional cerebral blood volume (rCBV) was a significant (p < 0.001) differentiating modality between PLC and glioblastomas or metastases.²⁰ Studies by Goyal et al., who also examined rCBV, You et al., who examined arterial spin labelling MRI, and Ko et al., who used ADC and AUC in the tumour and surrounding areas for comparison, all found significant differences between PCL and other brain tumours, mostly glioblastomas.^21–23

ADC values measured from DWI have been shown to be inversely related to tumour cellularity.¹⁶ ADC values in PCL are reported to be significantly lower than those in glioblastomas. Similarly, PCL have a higher tumour cellularity than glioblastomas, as proven by a higher nuclear-to-cytoplasm ratio. This indicates that the higher tumour cellularity in PCL leads to restricted mobility and diffusion of water molecules in tissue, hence leading to lower ADC scores. Authors of numerous studies have corroborated this.^{5,10,16–19,24}

As evident by our univariate and ROC curve analysis, ADC values and ratios have satisfactory discriminative capacity with sufficient sensitivity and specificity to distinguish between the two pathologies. These results are in line with what has been reported by previous studies.^{5,16,18,19,21,23} What is interesting, however, are the ER. Enhancement patterns depend on the integrity of the blood– barrier (BBB). Lesions that disturb the BBB present as contrast-enhancing lesions. Due to a significant overlap between the enhancement patterns of PCL and glioblastoma, conventional MRIs are not considered a significant modality to distinguish between the two. However, ERs are a new variable that have not been studied before to the best of our knowledge. If proven to be an accurate and reliable predictor, ERs and hence conventional MRIs have the capacity to play a greater role as a diagnostic modality. ERs are easily calculated and may be useful in resource-constrained settings where the luxury of DWI is not available.

According to our results, ERs managed to predict 71.4% of patients with glioblastoma and 88.9% patients with PCL. Its discriminative index of an AUC of 0.92 is similar to that of ADC ratios. Best results were achieved by combining ADC ratios and ERs which correctly predicted tumour type for all patients (p < 0.05).

Our study was limited by the small sample size. It was also limited by the fact that our recordings were made on a conventional T1 post-contrast sequence acquired at one point in time. Dynamic studies are time-consuming, expensive and not part of our standard departmental protocol. Therefore, they were not done for this study. T1 mapping could not be acquired due to lack of software and hence is a limitation to our study.

Conclusions

PCL shows an increased ER compared with glioblastoma. The ADC value and ADC ratio of PLC is lower than those of glioblastoma. The combination of ADC ratio and ER yielded the best results in the identification of both patient subgroups.

A future prospective study with a larger cohort is needed to strengthen the implication of our results further and to establish ADC and ER as a differentiating diagnostic modality between PCL and glioblastoma.

Funding

The authors received no financial support for the research, authorship, and/or publication of this article.

Conflict of interest

The authors declared no potential conflicts of interest with respect to the research, authorship and/or publication of this article.