Reported brain coverage, scan acquisition time, and post-processing methods varied between studies (see Table 1). In general, three main deconvolution methods were used. Nine studies used delay and dispersion correction deconvolution models producing delay time (DT) maps with the post-processing software, MIStar (15, 16, 18–22, 48, 49). Six studies used standard or ‘delay sensitive’ (16, 23, 27, 28, 43, 44) deconvolution which does not correct for arrival delay of contrast and six studies used ‘delay-insensitive’ also known as ‘delay corrected’ which does correct for arrival delay of contrast (23, 28, 42, 45–47). Studies investigating the impact of post-processing methods showed thresholds to define core and penumbra varied depending on the deconvolution method used (16, 20, 27, 28).

Imaging reference was a combination of MRI and NCT (6 studies) (15, 28, 42, 47, 49, 50), MRI alone (16 studies) (16, 18–23, 26, 27, 41, 44–46, 48, 51), or NCT alone (2 studies) (24, 25). Two studies investigating thresholds for penumbra also used contemporaneous MR-PWI (43) or CTP or MRP at 24 h (15). Time interval to reference varied between study protocols. For studies investigating core, most did not record median or mean time to reference imaging. When reported, median interval to acute MRI was 27.5 (IQR 22–31.5) min (see Table 1; Supplementary material) (16, 21, 22, 26, 27, 41, 44, 46). Other studies used DWI 24 h after CTP in patients with major reperfusion (18–23), delayed MRI, at a median interval of 38 h in patients with complete recanalization (48) or at 5–7 days (45) to determine infarct core. For studies investigating penumbra, several studies did not report median or mean time to reference scan (see Table 2 in Supplementary material).

Seventeen studies included patients with evidence of large vessel occlusion on imaging using CTA, MRA, or DSA (16, 18, 23–25, 27, 28, 41–45, 47–51). When reported, recanalization was determined using three scales: the Mori (50), thrombolysis in myocardial infarction (TIMI) (23, 49), and modified thrombolysis in cerebral infarction (mTICI) scales (18, 42, 48). Reperfusion therapies included intravenous thrombolysis (15, 16, 18–26, 41, 43–46, 50), local intra-arterial fibrinolysis (24, 25, 50), and clot retrieval in combination with thrombolysis (16, 18, 26, 50). Since 2015, three studies examined perfusion parameters following treatment with endovascular clot retrieval alone (42, 47, 48).

Imaging analysis performed by studies included either a region of interest analysis (24, 25, 42–44, 47, 48, 50), or as defined by the authors a pixel (18–23, 41, 46, 49) or voxel-based analysis (15, 16, 26–28, 45, 51). The term ‘voxel’ is used to encompass both pixels and voxels. Eleven studies used semi-automatic or automatic delineation of infarct on MRI using signal intensity thresholds (16, 18–23, 27, 28, 46, 48). Others manually outlined the maximal visual extent of the infarct on follow-up imaging (MRI or NCT) (15, 24–26, 41–44, 47, 49–51). Voxels within both the hypoperfusion region on CTP and infarct on DWI or NCT were regarded as “true-positive,” and those within the region of hypoperfusion on the index test but not in the reference test was “false-positive.” Most studies analyzed only the ischemic hemisphere (15, 16, 18–23, 43, 44). One study delineated non-infarcted voxels in the whole brain parenchyma (41). Two studies only examined the section with the largest tissue which evolved to infarction (27, 28).

Two different approaches were employed to determine the ability of perfusion parameters to accurately identify the final infarct on the follow-up scan. The first, a volume analysis approach defined the optimal parameter as the one with the least volume difference between index and reference standard. The second used receiver operator characteristic (ROC) curve analysis to identify parameters with the highest sensitivity and specificity to predict the final infarction on the reference image. Most diagnostic accuracy studies used ROC curve analysis alone (27, 28, 41, 42, 45, 47, 50, 51) or in conjunction with a volume analysis approach (15, 16, 18–22, 24–26, 44, 46, 49). Only one study used a volume analysis approach alone (48). Sensitivity, specificity, area under the curve (AUC), and the Youden index for the optimal parameters for core and penumbra from each study are included in Tables 3 4, respectively. The parameter with the highest Youden index of 0.94 was a product of CBF and CBV which at a threshold of 31.3 differentiated core from penumbra in an early study (24). The Youden index for rCBF <31% to represent ischemic core was less than 0.5 in three out of five studies (median 0.60, IQR 0.48, 0.63) (15, 16, 44).

The most reported parameter was rCBF, described in a total of 16 studies. The median rCBF threshold was <30% (IQR 30–40%). Figure 2 shows that rCBF thresholds varied between studies even from the same research groups. A regression analysis evaluating the relationship between published rCBF thresholds over a 10-year period showed no statistically significant association (regression coefficient: 0.1; p = 0.21) between rCBF threshold values and the publication year (see Figure 2). Five studies constrained rCBF within the perfusion lesion (using TTP > 4 s or DT > 2 s or DT > 3 s) to reduce false positives from regions of leukoaraiosis (15, 20, 22, 23, 43). In the early time window, one study showed thresholds were dependent on the reperfusion method used. They reported a lower threshold of rCBF <20% in patients treated with clot retrieval due to faster reperfusion [see Figure 2; Bivard (18)].

The definition of ‘infarct core’ continued to vary between studies (8, 9), with some defining core using the acute DWI lesion (16, 21, 22, 26, 27, 41, 43, 44, 46, 51), and others using delayed imaging in patients with complete recanalization [TIMI 3 (23, 49), TICI 2b (42, 47), or TICI 3 (42, 47, 48)] and clinical improvement. Five studies defined core in patients with evidence of ‘major’ reperfusion following treatment with thrombolysis or clot retrieval. This was defined as more than 80% reduction in the perfusion lesion volume between acute CTP and 24-h MR imaging, using MTT >145% of the normal tissue (16, 19, 20, 22) or Tmax >6 s (15). Others did not specify the threshold for MTT (21). One study used TICI 3 recanalization and ‘complete’ reperfusion defined as more than 90% reduction from baseline CTP using DT > 3 s (18). One early study defined core as the infarcted tissue on delayed NCT or T2-weighted MRI with reduced CBV and CBF on initial CTP (50). Another early study defined penumbra as the tissue with CBF <25 mL/100 g/min, which did not infarct on delayed NCT and then applied logistic regression to differentiate between penumbra and core in the gray (24) and white (25) matter in recanalized patients.

Three studies reported Tmax >6 s as the most accurate parameter for penumbra (AUC 0.78–0.87) (16, 23, 43). Using the MIStar software, three studies found DT > 2 s was the most accurate parameter and threshold for penumbra (20, 22, 49). One study (n = 22) reported a higher threshold (DT > 3 s) in a Chinese population (15). Two older studies suggested that relative MTT (28, 41) with thresholds varying from 145 to 249% was optimal for penumbra.

Three different methods were used to define penumbra. One pivotal study (43) used MR-PWI within an hour of CTP and matched thresholds for penumbra to MR-PWI threshold of Tmax >6 s. Five studies defined penumbra using CTP parameters and thresholds to identify ‘no or minimal reperfusion’ (<20% reperfusion) at 24 h (15, 16, 19, 20, 22). Studies defined minimal or no reperfusion using MTT >145% of the normal tissue (16, 19, 20, 22) or Tmax>6 s (15). Three studies used infarct growth between the area of hypoperfusion on CTP and the infarcted region found on 24-h MRI (23, 41) alone or combined with delayed NCT (49) in patients with persistent symptomatic occlusion on follow-up CTA or MRA (23, 41, 49). One study assumed persistent occlusion by including patients with vessel occlusion who in the absence of thrombolysis had no late clinical or radiological confirmation of reperfusion (28). Only two studies investigated thresholds for differentiating penumbra or ‘tissue at risk’ from ‘oligemia.’ (28, 50) Oligemia or ‘penumbra that recovers’ was defined as hypoperfusion on CTP (28) or reduced CBF and normal CBV (50), which did not infarct on delayed imaging.

The relationship between thresholds and time to reperfusion was investigated in three studies (42, 47, 48). Two of these studies demonstrated a statistically significant relationship between Tmax and aCBF and time to reperfusion from CTP (42, 47). If reperfusion occurred within 90 min, the optimal Tmax thresholds for the core was higher, and aCBF threshold lower than if reperfusion occurred between 90 and 180 min (42, 47). Optimal threshold values reported for rCBF were also lower if reperfusion occurred within 90 min (42) or when recanalization occurred within 4.5 h of stroke onset (see Figure 2; Laredo 2020) (48).

Eight studies investigated thresholds for the gray matter (GM) and the white matter (WM) (16, 24, 25, 28, 42, 45, 47, 50). Two early studies reported higher threshold values for the product of CBF and CBV in GM compared to WM (24, 25). For identifying penumbra, studies reported higher thresholds in WM than GM for aMTT (n = 23) (28) and Tmax (n = 31) (16) and lower thresholds in WM than GM for parameters, aCBF, and aCBV (n = 14) (50). However, thresholds for DT were constant (n = 31) (16). For identifying core, no significant difference in Tmax thresholds was reported between GM and WM (42, 47). Two studies reported consistently lower thresholds in WM than GM for CBF (16, 47), but one study (n = 132) did not (42). Reported thresholds for CBV were also generally lower in WM than GM (42, 47). One study reported higher thresholds for rCBF in highly vulnerable brain regions including the caudate body, putamen, and insular ribbon (51).