Perfusion magnetic resonance imaging in the differentiation between glioma recurrence and pseudoprogression: a systematic review, meta-analysis and meta-regression.

Background: Tumor recurrence and pseudoprogression (PsP) have similar imaging manifestations in conventional magnetic resonance imaging (MRI), although the subsequent treatments are completely different. This study aimed to evaluate the value of perfusion-weighted imaging (PWI) in differentiating PsP from glioma recurrence.
Methods: A comprehensive literature search was performed to evaluate clinical studies focused on differentiating recurrent glioma from PsP using PWI, including dynamic susceptibility contrast MRI (DSC-MRI), dynamic contrast enhanced MRI (DCE-MRI), and arterial spin labeling (ASL). Study selection and data extraction were independently completed by two reviewers. The Quality Assessment of Diagnostic Accuracy Studies 2 (QUADAS-2) tool was applied to evaluate the quality of the included studies. The software Stata 16.0 and Meta-Disc 1.4 were used for the meta-analysis. Meta-regression and subgroup analyses were applied to identify the sources of heterogeneity in the studies. This study was registered in the International Prospective Register of Systematic Reviews (PROSPERO) prior to initiation (CRD42022304404).
Results: A total of 40 studies were included, including 27 English studies and 13 Chinese studies. There were 1,341 patients with glioma recurrence and 876 patients with PsP. The pooled sensitivity and specificity of DSC-MRI for differentiating glioma recurrence from PsP were 0.82 [95% confidence interval (CI): 0.78 to 0.86] and 0.87 (95% CI: 0.80 to 0.92), respectively. The pooled sensitivity and specificity of DCE-MRI were 0.83 (95% CI: 0.76 to 0.89) and 0.83 (95% CI: 0.78 to 0.87), respectively. The pooled sensitivity and specificity of ASL were 0.80 (95% CI: 0.73 to 0.86) and 0.86 (95% CI: 0.76 to 0.92), respectively. Discussion: The DSC-MRI, DCE-MRI, and ASL perfusion techniques displayed high accuracy in distinguishing glioma recurrence from PsP, and DSC-MRI had a higher diagnostic performance than the other two techniques. However, due to the diversity of the parameters and threshold differences, further investigation and standardization are needed. 2022 Quantitative Imaging in Medicine and Surgery. All rights reserved.

Introduction

Glioma is the most common primary intracerebral tumor of the central nervous system (1). According to the 2021 World Health Organization (WHO) classification of tumors of the central nervous system, adult-type diffuse gliomas are divided into astrocytoma, oligodendroglioma and glioblastoma, accounting for approximately 22% of all central nervous system tumors (1,2). Currently, radiotherapy combined with adjuvant temozolomide chemotherapy has become the standard treatment for newly diagnosed glioma in adults (3,4). The higher the tumor grade, the higher the risk of recurrence and death (5). Regular follow-up and early detection of tumor recurrence have important clinical significance (6). According to the recommendations of the Response Assessment in Neuro-Oncology (RANO) Working Group, magnetic resonance imaging (MRI) examination is the main method for follow-up after treatment; however, in conventional MRI, tumor recurrence and pseudoprogression (PsP) have similar imaging manifestations, making them difficult to making them difficult to differentiate (7). Furthermore, the subsequent treatments for tumor recurrence and PsP are completely different (8).

At present, magnetic resonance (MR) perfusion-weighted imaging (PWI) is a hot research topic for many researchers in China and internationally. The most commonly used PWI techniques include dynamic susceptibility contrast MRI (DSC-MRI), dynamic contrast enhanced MRI (DCE-MRI), and arterial spin labeling (ASL). Among these methods, DSC-MRI is usually used to evaluate the distribution of microcirculation, the degree of microvascular proliferation, and blood perfusion (9); DCE-MRI is mainly applied to calculate functional parameters related to tissue flow and leakage of contrast agent from the intravascular space (10); and ASL can noninvasively reflect tissue blood perfusion information without contrast agents (11). The PWI findings of gliomas are shown in , and the MRI scan protocols and parameters are shown in Appendix 1 (Table S1). However, the previous meta-analysis (12-14) based on the above studies have involved small sample sizes, limited glioma grading and short time spans, which affected the stability and reliability of the results, and evaluation of the diagnostic value of MR perfusion imaging has remained incomplete. Therefore, this study attempted to perform a meta-analysis of published studies to evaluate the accuracy of MR perfusion studies in the differentiation of glioma recurrence from PsP, which may assist with future clinical treatment selection and improve the prognosis of patients. We present the following article in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses Diagnostic Test Accuracy (PRISMA-DTA) reporting checklist (15) (available at https://qims.amegroups.com/article/view/10.21037/qims-22-32/rc).

Figure 1

MRI findings of glioma recurrence and PsP. (A-D) Recurrent IDH-wildtype glioblastoma in the right temporal lobe in a 54-year-old man. (A) Axial T2-weighted imaging shows ill-defined lesion with heterogeneous hyperintensity. (B) Axial post-contrast T1-weighted imaging shows heterogeneous enhancement. (C) ASL image shows iso-perfusion mixed with spot-like hyper-perfusion. (D) The CBV map of DSC-MRI shows hyper-perfusion in most of the lesion. (E-H) PsP in IDH-mutant astrocytoma after surgery and radiotherapy in the right frontal lobe in a 53-year-old woman. (E) Axial T2-weighted imaging shows well-defined lesion with iso-intensity. (F) Axial post-contrast T1-weighted imaging shows ring enhancement. (G) ASL image shows hypo-perfusion. (H) The CBV map of DSC also shows hypo-perfusion. MRI, magnetic resonance imaging; PsP, pseudoprogression; ASL, arterial spin labeling; CBV, cerebral blood volume; DSC-MRI, dynamic susceptibility contrast magnetic resonance imaging.

Methods

The meta-analysis was registered on the International Prospective Register of Systematic Reviews (PROSPERO) with the registration number of CRD42022304404.

Literature search strategy

A systematic search in 4 international databases (PubMed, Embase, Web of Science and Cochrane Library) and 4 Chinese local academic databases [China National Knowledge Infrastructure (CNKI), Wanfang Med Online, Sinomed, and Chinese Medical Journal of Database (CMJD)] was performed up to 31 October 2021. The search terms were a combination of Medical Subject Headings (MeSH) terms and text words representing (I) glioma, (II) MR PWI, (III) tumor recurrence, and (IV) PsP. Details of the literature search strategy are provided in Appendix 2. Two reviewers independently screened paper titles, abstracts, and full text. For any difference of opinion that emerged during data extraction, consensus was reached between the two reviewers by discussion or consultation with a third reviewer. Articles that cited the included articles were also checked to see if any studies were omitted after the initial search.

Selection criteria

The inclusion criteria were as follows: (I) clinical studies using MRI perfusion imaging to differentiate between glioma recurrence and PsP, in Chinese or English; (II) studies in which local or whole brain radiotherapy was performed after surgery, and an abnormal enhanced lesion appeared in the operative area; (III) studies in which recurrence was defined as the pathological results of a second operation or combined follow-up examination, and the standard for radiation brain injury was mainly evidence from MRI follow-up; (IV) studies in which diagnostic 2×2 tables could be extracted directly or indirectly.

Studies were excluded if (I) the study type was a case report or review or they were published in the Chinese literature but not included in the Institute of Scientific and Technical Information of China (ISTIC); (II) they contained a sample size ≤30; (III) they included patients aged ≤18 years; (IV) they demonstrated incomplete reporting of essential data, such as a lack of sensitivity and specificity and incomplete and informally published studies.

Data extraction and quality assessment

Data extraction and quality evaluation included the following: (I) basic information: first author, publication year, country, study type, number of cases, age, WHO classification, treatment, and diagnostic criteria; (II) MR information, including MR equipment, field intensity, and perfusion imaging methods and parameters; (III) Data from 2×2 tables of diagnostic tests requiring evaluation, including true positive (TP), true negative (TN), false positive (FP), and false negative (FN) test results. The Quality Assessment of diagnostic Accuracy Studies 2 (QUADAS-2) tool was used to evaluate the quality of the included studies, including the risk of bias and applicability concerns.

Statistical analysis

Statistical analysis was performed using Stata 16.0 (Stata Corp, College Station, TX, USA) and Meta-Disc 1.4 (http://www.hrc.es/investigacion/metadisc_en.htm). The pooled sensitivity, specificity and diagnostic odds ratio (DOR) were calculated and the summary receiver operating characteristic (SROC) was plotted. The Spearman correlation coefficient was used to test the threshold effect. A Fagan plot was drawn to calculate prior probability and posterior probability. Cochran’s Q test was applied to determine whether there was heterogeneity, and I2 was adopted to measure the heterogeneity. If homogeneity among the results was good (Cochran’s Q test P>0.1; I2≤50%), a fixed-effects model was adopted; otherwise, a random-effect model was used along with an attempt to identify the source of heterogeneity through meta-regression and subgroup analysis. A P value of <0.05 was considered to indicate a statistically significant difference. Funnel plots were drawn to determine whether publication bias existed.

Results

Literature search process and study selection

A total of 1,342 studies were preliminarily identified through electronic database searches. After removing duplicate studies, they were assessed for eligibility for inclusion. A total of 40 studies were finally selected, including 27 English studies and 13 Chinese studies. There were 2,217 patients, including 1,341 cases of tumor recurrence and 876 cases of PsP. Of the patients in all of the included studies, 39.5% [95% confidence interval (CI): 0.37 to 0.42] displayed PsP due to treatment effects. A total of 60.5% (95% CI: 0.58 to 0.63) of the patients with progression were diagnosed with true progression. A flow chart of the study selection process is shown in .

Figure 2

Flow chart of the study selection process.

The characteristics of the included studies are shown in . Of the 40 studies, 30 were retrospective studies, 5 were prospective studies (24,28,35,41,42), and 5 were not described (43,46,47,52,53). Tumor types were classified into glioma (WHO grade II–IV) (31,35,37,40,44,46,52), high-grade glioma (WHO grade III–IV) (27,33,34,36,42,43,45,47-51,54,55) and glioblastoma (WHO grade IV) (16-26,28-30,32,38,39,41,53). All glioma grades were based on the WHO classification prior to 2021. As the “gold standard” for diagnosis, 28 studies used pathological diagnosis combined with follow-up, 6 studies used pathological diagnosis (20,26,33,35,55), and 6 studies were confirmed by follow-up alone (23,24,36,39,41,44). Among the 40 studies, DSC-MRI was used in 28 studies to distinguish glioma recurrence from PsP, most of which used relative cerebral blood volume (rCBV) as the best parameter (approximately 50%), and the other parameters included relative peak height (rPH), relative cerebral blood flow (rCBF), and 90% normalized cerebral blood volume (nCBV). The DCE-MRI technique was performed in 14 studies (20-22,25,30,32,36,38,39,41,42,44,45,51), most of which used transfer constant (Ktrans) as the best parameter; ASL was applied in 12 studies (19,24,29,31,35,36,40,43,48,49,53,55), and rCBF was the most commonly used parameter. For the quality assessment of the literature, the QUADAS-2 tool showed low-risk bias and good clinical applicability. The risk of bias and applicability concerns graph of the included studies is shown in . The methodology for quality assessment in this study was consistent with that of the previous meta-analysis (12), which included the risk of bias and applicability concerns. There was potential introduction of bias in patient selection, index test, reference standard, and flow and timing.

Table 1

Characteristics of the included studies

First author	Year	Nation	Study design	Cases	Age (y)	WHO grade	PWI	Best parameter	Field strength	Reference standard	Follow-up interval	TP	FP	TN	FN
Baek (16)	2012	Korea	R	79	50.6 [19–83]	IV	DSC-MRI	nCBV	3.0-T Philips	Both	Within 4 w	36	4	33	6
Barajas (17)	2009	USA	R	66	54.2	IV	DSC-MRI	rPH	1.5-T GE	Both	1.7–50.2 m	41	4	16	5
Cha (18)	2014	Korea	R	35	49 [24–70]	IV	DSC-MRI	Mean rCBV	3.0-T Philips	Both	Within 6 m	9	4	20	2
Choi (19)	2013	Korea	R	62	49.3 [22–79]	IV	DSC-MRI	nCBV	3.0-T Philips	Both	Within 4 w	28	9	19	6
							ASL	ASLmaps				27	10	18	7
Chung (20)	2013	Korea	R	57	52.1 [25–69]	IV	DCE-MRI	mAUCR	3.0-T Philips	Path	39.6 w	30	3	22	2
Elshafeey (21)	2019	USA	R	98	UR	IV	DCE-MRI	Ktrans	1.5-T and 3.0-T UR	Both	≤24 m	70	2	20	6
							DSC-MRI	rCBV				70	0	22	6
Heo (22)	2015	Korea	R	45	53.9 [27–73]	IV	DCE-MRI	90th% IAUC	3.0-T Philips	Both	14 w	16	7	18	4
Hu (23)	2011	USA	R	31	UR	IV	DSC-MRI	nCBV	UR	Follow up	Every 2–3 m	13	1	15	2
Jovanovic (24)	2017	Serbia	P	31	49±13.84	IV	ASL	CBF	3.0-T Siemens	Follow up	>3 m	18	1	10	2
							DSC-MRI	nCBV				20	0	11	0
Kim (25)	2014	Korea	R	169	52.2 [25–69]	IV	DSC-MRI	90th% nCBV	3.0-T Philips	Both	46.5 w	73	4	78	14
							DCE-MRI	30th% IAUC				79	11	71	8
Kim (26)	2014	Korea	R	51	51.5 [25–72]	IV	DSC-MRI	90th% nCBV	3.0-T Philips	Path	46.5 w	26	1	19	5
Kim (27)	2017	Korea	R	51	52.9±11.6	III–IV	DSC-MRI	90th% nCBV	3.0-T GE and Siemens	Both	≥6 m	28	5	14	4
Kong (28)	2011	Korea	P	59	25–74	IV	DSC-MRI	rCBV	3.0-T Philips	Both	≥3–4 m	27	6	20	6
Manning (29)	2020	USA	R	32	56±13	IV	ASL	nCBF	3.0-T GE	Both	≥6 m	23	1	6	2
							DSC-MRI	nrCBF				22	1	6	3
Nael (30)	2018	USA	R	46	32–78	IV	DSC-MRI	rCBV	3.0-T Siemens	Both	9–13 m	27	1	11	7
							DCE-MRI	Ktrans				23	2	10	11
Ozsunar (31)	2010	Turkey	R	32	42±11	II–IV	DSC-MRI	CBV	1.5-T GE	Both	1 d–4 w	19	3	7	3
							ASL	ASLmaps				18	1	11	3
Park (32)	2015	Korea	R	54	49.1±10.5	IV	DSC-MRI	90th% nCBV	UR	Both	Within 12 w	18	5	26	5
							DCE-MRI	90th% IAUC				19	6	25	4
Prager (33)	2015	USA	R	68	54.9 [22.6–79.4]	III–IV	DSC-MRI	rCBV	1.5-T and 3.0-T GE	Path	6.1 m (0.4–40.4 m)	50	2	8	8
Qiao (34)	2019	China	R	42	UR	III–IV	DSC-MRI	rCBVmean	3.0-T Siemens	Both	Interval >3 m	22	2	7	11
Razek (35)	2018	Egypt	P	42	UR	II–IV	ASL	CBF	1.5-T Philips	Path	11 m	23	1	17	1
Seeger (36)	2013	Germany	R	40	53.6	III–IV	DSC-MRI	CBV	1.5-T Siemens	Follow up	10 m (6–15 m)	19	4	13	4
				40	53.6	III–IV	DCE-MRI	Ktrans				14	3	14	9
				26	53.6	III–IV	ASL	rCBF				8	2	10	6
Steidl (37)	2021	Germany	R	104	52 [20–78]	II–IV	DSC-MRI	rCBV	1.5-T Philips and 3.0-T Siemens	Both	WHO III–IV 6 m, WHO II 12 m	45	0	21	38
Suh (38)	2013	Korea	R	79	51.2 [25–69]	IV	DCE-MRI	mAUCRH	3.0-T Philips	Both	Within 5 w	39	6	30	4
Thomas (39)	2015	USA	R	37	37–87	IV	DCE-MRI	90th% nVp	1.5- and 3.0-T GE	Follow up	UR	22	2	11	2
Wang (40)	2018	China	R	69	41.6 [18–77]	II–IV	ASL	nCBF	3.0-T GE	Both	Every 2–3 m/17 m (6–96 m)	24	3	31	11
				69	41.6 [18–77]	II–IV	DSC-MRI	nrCBF	3.0-T GE	Both	Every 2–3 m/17 m (6–96 m)	26	3	31	9
Yun (41)	2015	Korea	P	33	54.6 [28–82]	IV	DCE-MRI	5th% Ve	3.0-T Siemens	Follow up	UR	13	2	14	4
Zakhari (42)	2019	Canada	P	65	54.1 [50.9–57.3]	III–IV	DSC-MRI	CBV	3.0-T Siemens	Both	1–3 m	24	6	22	13
							DCE-MRI	Ktrans				19	9	19	18
Hu (43)	2019	China	UR	32	47 [28–69]	III–IV	ASL	rCBF	3.0-T Siemens	Both	11 m (6–26 m)	16	2	5	9
Qian (44)	2016	China	R	32	52.2±9.1	II–IV	DCE-MRI	Ktrans	3.0-T GE	Follow up	≥6 m	13	2	12	5
Ren (45)	2019	China	R	32	48.0±14.1	III–IV	DCE-MRI	Ktrans	UR	Both	≥6 m	21	1	9	1
Sha (46)	2013	China	UR	52	50.4±18.8	II–IV	DSC-MRI	rCBVmax	3.0-T Siemens	Both	2–4 m	21	0	22	9
Shan (47)	2020	China	UR	32	49.4	III–IV	DSC-MRI	rCBV	3.0-T Siemens	Both	≥6 m	16	1	9	6
Shi (48)	2020	China	R	40	55±11	III–IV	ASL	rCBF	3.0-T GE	Both	≥6 m	17	2	18	3
							DSC-MRI	rCBV				18	6	14	2
Wang (49)	2016	China	R	36	50 [19–70]	III–IV	ASL	rCBF	1.5-T GE	Both	≥10 m	6	5	23	2
Wang (50)	2017	China	R	56	56.4 [14.5–67.8]	III–IV	DSC-MRI	rCBV	3.0-T Siemens	Both	UR	25	0	26	5
Xie (51)	2019	China	R	86	45.2±5.6	III–IV	DCE-MRI	Ktrans	UR	Both	UR	40	7	31	8
Xing (52)	2016	China	UR	54	47 [13–74]	II–IV	DSC-MRI	rCBV	3.0-T Siemens	Both	≥6 m	26	1	19	8
Xu (53)	2018	China	UR	31	UR	IV	ASL	CBF	3.0-T Siemens	Both	≥6 m	11	9	8	3
							DSC-MRI	rCBV				9	11	6	5
Yin (54)	2015	China	R	96	43 [24–55]	III–IV	DSC-MRI	rCBV	UR	Both	2 m–2 y	67	18	6	5
Zhang (55)	2019	China	R	58	58 [18–65]	III–IV	ASL	rCBF	3.0-T GE	Path	UR	29	0	20	9

R, retrospective; P, prospective; UR, unreported; y, years; WHO, World Health Organization; PWI, perfusion-weighted imaging; DSC-MRI, dynamic susceptibility contrast magnetic resonance imaging; ASL, arterial spin labeling; DCE-MRI, dynamic contrast enhanced magnetic resonance imaging; CBV, cerebral blood volume; nCBV, normalized CBV; rPH, relative peak height; rCBV, relative CBV; AUC, area under the curve; mAUCR, mean AUC ratio; Ktrans, transfer constant; IAUC, initial AUC; CBF, cerebral blood flow; nCBF, normalized CBF; nrCBF, normalized relative CBF; rCBF, relative CBF; mAUCRH, mean AUC RH; Vp, volumetric plasma volume; nVp, normalized Vp; Ve, volume fraction of extracellular extravascular space; Path, pathology; d, days; w, weeks; m, months; TP, true positive; TN, true negative; FP, false positive; FN, false negative.

Figure 3

Risk of bias and applicability concerns graph for each included study. High risk (−), unclear risk (?) and low risk (+).

Results of the meta-analysis of the DSC-MRI studies

A total of 28 studies were included in the meta-analysis, including 21 English studies and 7 Chinese studies. The threshold effect test results showed that Spearman’s correlation coefficient was −0.3 (P=0.09). After drawing the SROC diagram, no obvious “shoulder-arm shape” emerged, indicating that there was no heterogeneity caused by the threshold effect in this study. The results of the forest plots showed that the Q test of sensitivity was P<0.01 with I2=68.33%, and the Q test of specificity was P<0.01 with I2=81.00%, indicating that there was significant heterogeneity among the included studies. Therefore, a random-effects model was used to analyze the pooled sensitivity and specificity in the DSC-MRI studies. The results showed that the pooled sensitivity, specificity, positive likelihood ratio (PLR), negative likelihood ratio (NLR), and DOR were 0.82 (95% CI: 0.78 to 0.86), 0.87 (95% CI: 0.80 to 0.92), 6.5 (95% CI: 4.1 to 10.3), 0.20 (95% CI: 0.17 to 0.25) and 32 (95% CI: 18 to 55), respectively. The area under the curve (AUC) was 0.89 (95% CI: 0.86 to 0.92). Fagan plots displayed a prior probability of 0.5 and a posterior probability of 0.87 and 0.17 for the PLR and NLR, respectively. The above results for the DSC-MRI studies are shown in and .

Table 2

Diagnostic results of PWI for differentiating glioma recurrence from PsP

PWI	Studies	Cases	Se (95% CI)	Sp (95% CI)	PLR (95% CI)	NLR (95% CI)	DOR [95% CI]	AUC (95% CI)
DSC-MRI	28	1,645	0.82 (0.78–0.86)	0.87 (0.80–0.92)	6.5 (4.1–10.3)	0.20 (0.17–0.25)	32 [18–55]	0.89 (0.86–0.92)
DCE-MRI	14	873	0.83 (0.76–0.89)	0.83 (0.78–0.87)	4.9 (3.6–6.6)	0.20 (0.13–0.30)	24 [12–47]	0.88 (0.85–0.91)
ASL	12	492	0.80 (0.73–0.86)	0.86 (0.76–0.92)	5.7 (3.1–10.3)	0.23 (0.16–0.33)	24 [10–57]	0.88 (0.85–0.91)

PWI, perfusion-weighted imaging; PsP, pseudoprogression; DSC-MRI, dynamic susceptibility contrast magnetic resonance imaging; DCE-MRI, dynamic contrast enhanced magnetic resonance imaging; ASL, arterial spin labeling; Se, sensitivity; CI, confidence interval; Sp, specificity; PLR, positive likelihood ratio; NLR, negative likelihood ratio; DOR, diagnostic odds ratio; AUC, area under the curve.

Figure 4

Forest plots of sensitivity and specificity in the included studies [(A) DSC-MRI, (B) DCE-MRI, (C) ASL]. CI, confidence interval; DSC-MRI, dynamic susceptibility contrast magnetic resonance imaging; DCE-MRI, dynamic contrast enhanced magnetic resonance imaging; ASL, arterial spin labeling.

Figure 5

SROC curves of three PWI techniques to distinguish glioma recurrence from PsP [(A) DSC-MRI, (B) DCE-MRI, (C) ASL]. DSC-MRI, dynamic susceptibility contrast magnetic resonance imaging; SENS, sensitivity; SPEC, specificity; SROC, the summary receiver operating characteristic; AUC, area under the curve; PWI, perfusion-weighted imaging; PsP, pseudoprogression; DCE-MRI, dynamic contrast enhanced magnetic resonance imaging; ASL, arterial spin labeling.

Figure 6

Fagan plots of three PWI techniques to distinguish glioma recurrence from PsP [(A) DSC-MRI, (B) DCE-MRI, (C) ASL]. LR, likelihood ratio; PWI, perfusion-weighted imaging; PsP, pseudoprogression; DSC-MRI, dynamic susceptibility contrast magnetic resonance imaging; DCE-MRI, dynamic contrast enhanced magnetic resonance imaging; ASL, arterial spin labeling.

Results of the meta-analysis of the DCE-MRI studies

A total of 14 studies were included, including 11 English studies and 3 Chinese studies. The threshold effect test results showed that Spearman’s correlation coefficient was 1.0 (P=1.00). After drawing the SROC diagram, no obvious “shoulder-arm shape” emerged, indicating that there was no heterogeneity caused by the threshold effect in this study. The results of forest plots showed that the Q test of sensitivity was P<0.01 with I2=77.64%, and the Q test of specificity was P>0.1 with I2=0.00%, indicating that there was significant heterogeneity in the sensitivity among the included studies. Therefore, a random-effects model was used to analyze the pooled sensitivity and specificity in the DCE-MRI studies. The results showed that the pooled sensitivity, specificity, PLR, NLR, and DOR were 0.83 (95% CI: 0.76 to 0.89), 0.83 (95% CI: 0.78 to 0.87), 4.9 (95% CI: 3.6 to 6.6), 0.20 (95% CI: 0.13 to 0.30) and 24 (95% CI: 12 to 47), respectively. The AUC was 0.88 (95% CI 0.85 to 0.91). Fagan plots showed a prior probability of 0.5 and a posterior probability of 0.83 and 0.17 for PLR and NLR, respectively. The above results for the DCE studies are shown in and .

Results of the meta-analysis of the ASL studies

A total of 12 studies were included, including 7 English studies and 5 Chinese studies. The threshold effect test results showed that Spearman’s correlation coefficient was 0.4 (P=0.16). After drawing the SROC diagram, no obvious “shoulder-arm shape” emerged, indicating that there was no heterogeneity caused by the threshold effect in this study. The results of forest plots showed that the Q test of sensitivity was P>0.1 with I2=44.70%, and the Q test of specificity was P<0.01 with I2=68.29%, indicating that there was significant heterogeneity in the sensitivity among the included studies. Therefore, a random-effects model was used to analyze the pooled sensitivity and specificity in the ASL studies. The results showed that the pooled sensitivity, specificity, PLR, NLR, and DOR were 0.80 (95% CI: 0.73 to 0.86), 0.86 (95% CI: 0.76 to 0.92), 5.7 (95% CI: 3.1 to 10.3), 0.23 (95% CI: 0.16 to 0.33) and 24 (95% CI: 10 to 57), respectively. The AUC was 0.88 (95% CI: 0.85 to 0.91). Fagan plots showed a prior probability of 0.5 and a posterior probability of 0.85 and 0.19 for PLR and NLR, respectively. The above results for the ASL studies are shown in and .

Meta regression analysis and subgroup analysis

Univariate regression analysis was applied to identify the sources of study heterogeneity, including study type, tumor type, diagnostic criteria, field strength, and MRI parameter. The results demonstrated that tumor type was the main factor leading to the heterogeneity of the sensitivity in the DSC-MRI studies and specificity in the DSC-MRI studies, and the difference was statistically significant. In the DCE-MRI studies, study type was the main reason leading to the heterogeneity of the sensitivity. The results of the meta-regression analysis are shown in Appendix 3 (Table S2).

Subgroup analysis further clarified the impact of the above factors on the heterogeneity of the results. In the DSC-MRI studies, in which 14 studies were conducted on WHO grade IV gliomas, the pooled sensitivity and specificity were 0.85 and 0.88, respectively, and the AUC was 0.90. Nine studies were conducted on WHO grade III–IV gliomas, with pooled sensitivity and specificity of 0.79 and 0.87, respectively, and an AUC of 0.87. The remaining 5 studies involved WHO grade II–IV gliomas with pooled sensitivity and specificity of 0.82 and 0.87, respectively, and an AUC of 0.87. In the DCE-MRI studies, 12 studies were retrospective studies, with pooled sensitivity and specificity of 0.86 and 0.84, respectively, and an AUC of 0.89. Two studies were prospective, with pooled sensitivity and specificity of 0.60 and 0.75, respectively. In the studies of ASL, 3 involved WHO grade II–IV gliomas, and their pooled sensitivity and specificity were 0.81 and 0.92, respectively, with an AUC of 0.98. Five studies involved WHO grade III–IV gliomas, with pooled sensitivity and specificity values of 0.72 and 0.88, respectively, and an AUC of 0.84. Four studies focused on WHO grade IV gliomas. The pooled sensitivity and specificity were 0.85 and 0.67, respectively, with an AUC of 0.95. See Appendix 4 (Table S3).

Publication bias

The Deeks’ funnel plot showed that the data were symmetrically distributed (DSC-MRI: P=0.216; DCE-MRI: P=0.381; ASL: P=0.735), suggesting no significant publication bias, as shown in . In view of the possibility of publication bias as expected in the studies from local China databases, we performed subgroup analyses and drew funnel plots to determine whether publication bias existed. The results showed that there was no publication bias in DSC-MRI, DCE-MRI, and ASL studies from Chinese and English databases (P>0.05) [Appendix 5 (Table S4, Figures S1-S3)].

Figure 7

Funnel plots of the included studies [(A) DSC-MRI, (B) DCE-MRI, (C) ASL]. ESS, effective sample size; DSC-MRI, dynamic susceptibility contrast magnetic resonance imaging; DCE-MRI, dynamic contrast enhanced magnetic resonance imaging; ASL, arterial spin labeling.

Discussion

In the process of postoperative radiation therapy for glioma, it is easy to cause brain damage (56). As it usually occurs weeks to months after radiotherapy (8,57), PsP is difficult to distinguish from tumor recurrence (58). Histopathological diagnosis is the gold standard for differentiating between the two; however, it is difficult for patients to undergo biopsy or a second operation before receiving the final diagnosis. At present, some MRI imaging methods are used to determine whether there is tumor progression, such as diffusion weighted imaging (DWI), PWI, and MR spectrum imaging and so on. Of these, PWI is one of the most reliable imaging techniques (24,29). In this study, three MRI perfusion imaging techniques (DSC-MRI, DCE-MRI, and ASL) were included to systematically evaluate and statistically analyze studies focused on differentiating between tumor recurrence and PsP. The incidence of PsP found in our study was comparable with what is known from the literature, 39.5% in the included studies versus 37% in a previous meta-analysis by Abbasi et al. (59).

The results demonstrated that all three perfusion imaging methods displayed a high pooled diagnostic performance. Among the three, DSC-MRI performed the best, with a pooled sensitivity and specificity of 0.82 and 0.87, respectively, a DOR of 24, and an area under the SROC curve of 0.89. However, heterogeneity analysis indicated obvious heterogeneity among the studies. The threshold effect test combined with the SROC curve results suggested that heterogeneity was not caused by the threshold effect; hence, meta regression analysis and subgroup analysis were applied to find the source of heterogeneity. After analysis, the sensitivity of the tumor type group was significantly higher than that of the other groups. Further subgroup analysis showed that the sensitivity of WHO grade IV glioma patients was significantly higher than that of WHO grades III–IV and WHO grades II–IV glioma patients, with sensitivities of 0.85, 0.79, and 0.72, respectively.

The DSC-MRI is a functional MR imaging technique that reflects the distribution of tissue microvessels and blood perfusion. Due to the increased expression of vascular endothelial growth factor, high neovascular density, and immature vascular structure in the recurrence area of grade IV glioma, the recurrence area shows hyperperfusion, while the area of PsP displays hypoperfusion due to vascular endothelial cell apoptosis (60). For lower-grade gliomas (such as WHO grade II), there was no significant difference in changes after radiotherapy due to relatively little neovascularization. The results in our study were similar to those of the previous meta-analysis by Wang et al. (61), suggesting that DSC-MRI is the perfusion imaging technique with the highest accuracy for differentiating glioma recurrence from PsP; however, the results of the subgroup analysis were different. In the study by Wang et al., field strength and tumor type specificity were the sources of heterogeneity among DSC-MRI studies. The reason for the difference may be related to the exclusion of small sample (≤30 participants) studies as well as the inclusion of only adult glioma patients in our study, and the addition of Chinese glioma-related studies.

Compared with traditional DSC-MRI, DCE-MRI has higher spatial resolution, which not only provides tumor perfusion information, but also reflects vascular permeability (62). However, DCE needs to select an appropriate pharmacokinetic model, and the parameters obtained are relatively complex and diverse, leading to relatively few studies having focused on this technique and less frequent clinical application compared with DSC-MRI. Despite its limitations, our study shows that DCE also displays a high accuracy for differentiating recurrent glioma from PsP, with a sensitivity and specificity of 0.83 and 0.83, respectively. As a complete noncontrast agent perfusion imaging technology, ASL applies water molecules in endogenous arterial blood as tracers, which are not affected by the integrity of the blood-brain barrier and are able to more truly reflect tissue perfusion (63). Our results are similar to the meta-analysis results of Du et al. (14), demonstrating that ASL has high sensitivity and specificity in distinguishing glioma recurrence and PsP, with a sensitivity and specificity of 0.80 and 0.86 respectively.

The studies included in this meta-analysis were based on the 2016 or earlier WHO classification. In 2021, the WHO classification was updated, emphasizing the important role of genetics in the development and subsequent treatment of glioma (2). For adult gliomas, changes in glioblastoma have greater clinical significance. It has been verified that IDH-mutant and IDH-wildtype have distinct biological behaviors and prognosis (64-66). In the new classification, glioblastoma represents only IDH-wildtype glioma. Alternatively, tumors that contain one or more of three genes [TERT promoter mutation, EGFR gene amplification, or copy number changes on chromosome 7/10 (+7/−10)] into the classification of glioblastoma (2). These changes contribute to a more homogeneous study population in clinical trials. Other molecular alterations, such as CDKN2A/B homozygous deletion in IDH-mutant gliomas, tends to predict worse prognosis (67,68).

There were some limitations to this study. First, the inclusion criteria of this study did not entirely depend on histopathological diagnosis, and differences in follow-up time and diagnostic criteria may have caused bias in the study results. Second, this study included WHO grade II–IV tumors according to the WHO Classification of central nervous system tumors prior to 2021. Although most of the tumors were WHO grade IV tumors, the results of the analysis may have been biased by treatment differences due to different tumor grades. In addition, most of the included studies were retrospective studies, MRI perfusion imaging parameters were more complicated, and the selection of parameters and threshold values lacked uniform standards, which may have aggravated the heterogeneity of the studies. Finally, in the quality assessment of the included studies, it was found that some of the studies did not report blinding in detail, and there may have been risk bias in measurements and subsequent results.

To sum up, our meta-analysis demonstrated that DSC-MRI, DCE-MRI, and ASL, as advanced MR perfusion imaging techniques, could accurately differentiate postoperative glioma recurrence from PsP. Among them, DSC-MRI had a higher diagnostic performance than the other two techniques. Therefore, MRI perfusion imaging could be used as a feasible and quantitative examination method for postoperative follow-up after radiotherapy and chemotherapy, providing strong evidence to support the subsequent clinical treatment.

The article’s supplementary files as