Diagnosis of Cerebral Amyloid Angiopathy: Evolution of the Boston Criteria

The history of how to diagnosis cerebral amyloid angiopathy (CAA) tells the story of the disease itself. CAA is defined by histopathology—deposition of β-amyloid in the cerebrovasculature—and through the 1980s the disorder was only diagnosed in patients with available brain tissue from hematoma evacuation, biopsy, or most commonly post-mortem exam.¹ Introduction of the imaging-based Boston Criteria for diagnosis of CAA in the 1990s^{2, 3} allowed a diagnosis of “probable CAA” in living patients with no available brain tissue and substantially moved the field from the pathologist’s realm to the clinician’s. The Boston Criteria for CAA have become the basis for clinical decision-making as well as a rapidly growing body of literature⁴ investigating the disease’s clinical manifestations, phenotypic spectrum, progression, and potential for disease-modifying therapy.

The history of CAA diagnostic criteria also illustrates broader issues for other major central nervous system diseases. If the brain were as accessible to direct tissue examination during life as the blood or even the liver, diagnosis and staging of brain disorders such as cerebral small-vessel or neurodegenerative disease would be relatively straightforward and the state of clinical trials would presumably be more advanced. Given the relative inaccessibility of brain tissue, however, diagnostic approaches have needed to rely on indirect but nonetheless powerful methods such as MRI. The current article will review the evolution and application of the Boston Criteria, how the criteria have contributed to the search for CAA biomarkers, and future directions in this still evolving field.

Development and Validation of the Boston Criteria for CAA

The Boston Criteria for diagnosing CAA arose from discussions between one of the authors (SMG), Drs. Carlos Kase, Daniel Kanter, and the late Michael Pessin. The criteria were first published in 1995 in the Methods section of an analysis of CAA and the apolipoprotein E ε4 allele² and in 1996 as a table in a clinical-pathologic case report.³ Using the category terminology applied to other brain disorders such as Alzheimer’s disease,^{5, 6} they defined definite CAA based on full autopsy, probable or possible CAA based on brain imaging plus clinical exclusions, and an additional category of probable CAA with supporting pathology based on clinical scenarios of having limited brain tissue from biopsy or hematoma evacuation (Table 1). Definite CAA requires high neuropathological severity (including features of advanced vasculopathy such as amyloid replacement and splitting of the blood vessel wall)^{7, 8} in order to avoid diagnosing the condition when the pathology is only mild and incidental. Lesser histopathological severity is required for probable CAA with supporting pathology to reflect the smaller amount of sampled tissue and consequent lower likelihood of identifying the most advanced foci of disease.⁹

Table 1.

The Modified Boston Criteria for Cerebral Amyloid Angiopathy

1. Definite CAA Full post-mortem examination demonstrating: Lobar, cortical, or cortical-subcortical hemorrhage Severe CAA with vasculopathy Absence of other diagnostic lesion

2. Probable CAA with supporting pathology Clinical data and pathologic tissue (evacuated hematoma or cortical biopsy) demonstrating: Lobar, cortical, or cortical-subcortical hemorrhage (including ICH, CMB, or cSS) Some degree of CAA in specimen Absence of other diagnostic lesion

3. Probable CAA Clinical data and MRI or CT demonstrating: Multiple hemorrhages (ICH, CMB) restricted to lobar, cortical, or cortical-subcortical regions (cerebellar hemorrhage allowed), OR single lobar, cortical, or cortical-subcortical hemorrhage and cSS (focal or disseminated) Age≥55 years Absence of other cause of hemorrhage*

4. Possible CAA Clinical data and MRI or CT demonstrating: Single lobar, cortical, or cortical-subcortical ICH, CMB, or cSS (focal or disseminated) Age ≥55 years Absence of other cause of hemorrhage*

^*Other causes of hemorrhage (differential diagnosis of lobar haemorrhages): antecedent head trauma, hemorrhagic transformation of an ischemic stroke, arteriovenous malformation, haemorrhagic tumour, warfarin therapy with international normalisation ratio > 3, vasculitis

The key diagnostic category for clinical practice and research is probable CAA, the highest level of diagnostic certainty currently achievable without obtaining brain tissue. As first formulated (“original Boston Criteria”), probable CAA entailed neuroimaging demonstration of multiple (i.e. two or more) hemorrhages restricted to lobar brain regions,^{2, 3} defined as cerebral cortex, the corticosubcortical (grey-white) junction, and subcortical white matter. A modification to count blood products in cortical sulci (cortical superficial siderosis, cSS) as one additional hemorrhagic lesion (“modified Boston Criteria”), was proposed and validated in 2010.¹⁰ The most recent version of the criteria are thus known as the modified Boston Criteria (Table 1). The requirement for multiple strictly lobar hemorrhages is based on the lobar predilection of CAA pathology and recurrent intracerebral hemorrhages (ICHs),¹ an anatomic distribution that is in near-perfect contrast to the deep hemispheric and brainstem locations favored by ICHs due to hypertensive arteriopathy.¹¹ Since CAA typically spares these deep territories, the presence of any hemorrhagic lesions in basal ganglia, thalamus, or pons preclude the probable CAA diagnosis. Cerebellar hemorrhages can result from either CAA or hypertensive arteriopathy and are therefore not counted by the Boston Criteria, either in favor or against a probable CAA diagnosis.

Several methodologic issues arise in applying the Boston Criteria in practice. One is that all types of hemorrhagic lesions—ICHs, cerebral microbleeds (CMBs),¹² and (since publication of the modified Boston Criteria)¹⁰ acute convexity subarachnoid bleeds or cSS¹³—count towards the multiple lobar hemorrhages required for probable CAA, or alternatively preclude probable CAA if in deep territories. The rationale for counting all types of hemorrhagic lesions is that while different hemorrhage sizes may arise by distinct pathogenic mechanisms,¹⁴ each presumably represents a distinct event of vessel leakage and therefore provides independent evidence for the underlying small vessel condition. The increasingly widespread use of blood-sensitive T2*-weighted MRI methods has greatly influenced detection of CMBs (Fig. 1) and cSS (Fig. 2) and thus the diagnosis of CAA as discussed below. Conversely hemorrhagic lesions that may be part of a larger hemorrhage, such as smaller extensions in proximity to a larger hemorrhage or foci of cSS near or directly connected to ICHs that have ruptured into the subarachnoid space, are considered as part of a single bleeding event and thus not counted as separate hemorrhages (see Fig. 1C and 2C). A second practical issue is that a hemorrhagic lesion in the centrum semiovale can appear a long distance from the outer surface of the brain and still be quite close to the corticosubcortical junction (Fig. 1D) because of the undulating curves of the cortical gyri. Hemorrhages are considered deep hemispheric only when clearly involving the basal ganglia, thalamus, or internal capsule.

Figure 1.

Patterns of cerebral microbleeds (CMB). (A) Multiple strictly lobar CMB on T2*-weighted MRI of a 69-year-old woman who presented with a spontaneous lobar intracerebral hemorrhage. Brain autopsy showed advanced CAA. (B) Mixed CMB (arrowheads) affecting the right thalamus, a deep hemispheric territory, as well as lobar brain regions and therefore not fulfilling Boston criteria for probable CAA. (C) Subacute left frontoparietal lobar hemorrhage (asterisk), numerous strictly lobar CMB (white arrowheads), and a left frontal focus of cortical superficial siderosis (white arrow) on susceptibility-weighted imaging (SWI) MRI of a 78-year-old woman. The image additionally shows foci of CMB (orange arrowheads) and cortical superficial siderosis (red arrows, also seen in magnified image) that are immediately adjacent to the lobar hemorrhage and therefore not counted as separate lesions in determining the number of lobar hemorrhagic foci. (D) SWI from a 71-year-old man with memory loss and CAA on brain biopsy. The yellow arrows point to CMB in the right temporal and right occipital lobes that might appear distant from the brain surface, but would be counted as lobar microbleeds. The aligned T1-weighted slice shows that their positions (asterisks on the right panel) are within or very close to the cortical ribbon.

Figure 2.

Patterns of cortical superficial siderosis (cSS) in CAA patients. (A) Susceptibility-weighted imaging (SWI) MRI showing a single sulcus with cSS (arrow), classified as focal cSS. (B) SWI showing cSS affecting multiple cortical sulci, classified as disseminated (>3 affected sulci). (C) SWI images from a 68-year-old man with a right parasagittal CAA-related spontaneous lobar intracerebral hemorrhage (asterisk). The arrow points to an area of cSS close to the hematoma on the axial slice. The corresponding sagittal slice (right panel) shows that this cSS focus connects to the lobar hemorrhage and thus would not be counted as an independent hemorrhagic lesion. There are multiple lobar cerebral microbleeds in the left hemisphere.

MRI-histopathological studies to date^{10, 15–17} have provided validating evidence for the Boston Criteria probable CAA diagnosis (Table 2), with sensitivities that appear to depend in part on the clinical presentation of the patients examined. Among three hospital-based studies of patients presenting primarily with ICH who underwent T2*-weighted MRI,^{10, 15, 16} probable CAA by original Boston Criteria showed sensitivities ranging from 57.9% to 76.9% and specificities of 87.5% to 100%. One head-to-head comparison of original to modified criteria¹⁰ suggested that incorporation of cSS presence improved sensitivity without lowering specificity (Table 2). A fourth study¹⁷ analyzed a hospital-based cohort of non-ICH individuals with other clinical presentations such as cognitive impairment or transient focal neurological episodes, and found lower sensitivity of 42.4% and similar specificity of 90.9%. A regression model of these data showed that increasing lobar CMB counts predicted greater likelihood of CAA pathology. The same study¹⁷ also analyzed a community-based cohort of individuals with T2*-weighted MRI and subsequent autopsy, finding the probable CAA diagnosis to have low sensitivity of only 4.5% for pathological CAA with specificity of 88.0%.

Table 2.

Summary of validation studies of Boston Criteria probable CAA

Setting	CAA pathology + subjects (ICH+/ICH-)	CAA pathology -subjects (ICH+/ICH-)	Sensitivity	Specificity
MRI-neuropathology studies
Hospital-based15	11 (11/0)	4 (4/0)	72.7%	100%
Hospital-based10	38 (27/11)	22 (22/0)	57.9% 71.1%*	95.5% 95.5%*
Hospital-based16	14 (9/5)	10 (10/0)	76.9%	87.5%
Hospital-based Population-based17	33 (0/33) 22 (0/22)	22 (0/22) 25 (0/25)	42.4% 4.5%	90.9% 88.0%
MRI-genetic studies
	CAA mutation+ subjects
Hospital-based, Dutch-type CAA18	15 ICH+ 12 ICH-	NA	100% 16.7%	NA

Data only from individuals with both T2*-weighted MRI and histopathologic or genetic diagnosis.

^*Using modified Boston criteria. All other values use original Boston criteria.

The Boston Criteria have also been validated by MRI-genetic correlation in individuals carrying CAA-specific, fully penetrant amyloid precursor protein (APP) mutations (Table 2). Among carriers of the Dutch-type APP mutation, 15 of 15 with symptomatic hemorrhagic stroke had multiple strictly lobar hemorrhagic lesions meeting original probable CAA criteria (outside of the age≥55 requirement).¹⁸ Only 2 of 12 mutation carriers without symptomatic ICH met this definition, suggesting high sensitivity for symptomatic Dutch-type hereditary disease but low sensitivity for the pre-symptomatic phase. Specificity could not be estimated in this study, as mutation-negative individuals were not scanned. Another report identified five individuals carrying other CAA-associated APP mutations (Iowa-, Italian-, and Flemish-types) whose hemorrhagic lesions also met modified probable CAA criteria.¹⁹

The above validation studies have notable limitations: small sample sizes, restriction primarily to a single site and to white participants, and varying T2*-weighted MRI methods (discussed below). These concerns notwithstanding, the studies suggest that the modified probable CAA criteria have 1) reasonably high specificity for pathologic CAA in all settings, and 2) high sensitivity among patients presenting with symptomatic hemorrhages, possibly lower sensitivity for non-ICH presentations, and quite low sensitivity in the general population. The trend for sensitivity to increase with greater severity or later stage of CAA presumably reflects the long-recognized observation that CAA pathology needs to be substantially advanced before it is severe enough to trigger hemorrhages.^{7, 8} The regression analysis showing increasing specificity with higher CMB counts¹⁷ points to the additional possibility that likelihood of CAA follows a graded relationship with hemorrhage number rather than a sharp cutoff at ≥2 hemorrhages.

Variation in MRI hemorrhage detection adds a further layer of complexity. CMB detection in particular¹² is strongly influenced by a range of factors in MRI acquisition and processing, including magnetic field strength, echo time, scan resolution, incorporation of phase information (used in susceptibility-weighted imaging²⁰), or weighted averaging across multiple echo times (used in susceptibility weighted angiography²¹). Systematic comparisons of concurrently obtained MRIs indicate that these parameters substantially affect the number of CMBs counted by raters,^{22, 23} which means any study of CAA-associated CMB will be influenced by the precise MRI method used. Recent ex vivo MRI analysis of postmortem brains²⁴ suggests that the theoretical goal of close to 100% CMB detection might ultimately be achievable, but only with very high resolution (on the order of 200 μm isotropic voxels). Additional lesions detected on even further reductions in voxel size (to 75 μm isotropic) mostly represented CMB mimics such as small vessel occlusions or microaneurysms rather than bona fide microbleeds.

The possible CAA Boston Criteria category refers to individuals with exactly one hemorrhagic lesion,¹⁵ or (per modified Criteria) cSS only without ICH or CMB.¹⁰ Little MRI-pathological correlation has been performed in possible CAA. Of the 8 individuals in the initial validation study¹⁵ with only an isolated lobar ICH on T2*-weighted MRI, 3 had CAA pathology, supporting the interpretation that possible CAA carries less diagnostic certainty than probable CAA. The impact of MRI parameters on which individuals are diagnosed with probable vs possible CAA has not been systematically analyzed, but based on the above is likely considerable. Another commonly encountered pattern is “mixed” hemorrhagic lesions located in both lobar and deep territories (Fig. 1B). A recent analysis of 75 mixed-ICH patients²⁵ found 66 (88%) to be hypertensive and have other markers of hypertensive small vessel disease such as higher serum creatinine and abundant enlarged perivascular spaces in the basal ganglia relative to patients with probable CAA. These findings suggest contribution from hypertensive arteriopathy, but do not preclude overlapping CAA in at least a subset. Because of the high level of diagnostic uncertainty, patients in this mixed category pose substantial challenges to clinicians (see Future Directions below).

Role of the Boston Criteria in the Search for Biomarkers

The ability to diagnose CAA during life with good specificity is a prerequisite for identifying other biomarkers of the disease’s presence, severity, and future behavior. The probable CAA diagnosis—derived from the number and distribution of hemorrhagic lesions—has indeed been the basis for identifying a range of nonhemorrhagic biomarkers of CAA. Basing a biomarker on probable CAA rather than requiring histopathologic confirmation runs the risk that individuals wrongly diagnosed will yield spurious findings, either false-positives or more likely false-negatives due to misclassification of the “exposure” variable of CAA. The experience in CAA research, however, suggests this approach is preferable to the alternative of restricting studies solely to pathologically verified CAA. This latter approach carries its own major limitations in both sample size and generalizability, as brain tissue (whether from hematoma evacuation, biopsy, or autopsy) becomes available not at random but rather in select subgroups with particularly severe or atypical clinical courses.

Among the long (and likely still growing) list of CAA biomarkers made possible by the Boston Criteria are: 1) white matter T2 hyperintensities (WMH),²⁶ with tendency for posterior predominance²⁷ or a multiple subcortical spot pattern;²⁸ 2) altered diffusion-tensor imaging parameters such as global mean diffusion²⁹ or diffusion-tensor-derived global efficiency;³⁰ 3) vascular reactivity to functional stimulation;^{31, 32} 4) cortical thickness;³³ 5) punctate diffusion-weighted imaging hyperintensities suggestive of acute microinfarcts;^{34, 35} 6) enlarged perivascular spaces in the centrum semiovale (CSO-EPVS);^{36, 37} 7) positron-emission tomography (PET) labeling with the amyloid ligands Pittsburgh Compound B and Florbetapir;^38–42 and 8) reduced cerebrospinal fluid concentrations of β-amyloid.^43–45 These multimodal biomarkers serve as both important windows into the pathogenesis of CAA and candidate outcome markers for clinical trials.⁴⁶

The dependence of the Boston Criteria CAA diagnosis on hemorrhagic lesions limits analysis of any biomarkers that appear prior to or entirely without CAA-related bleeding. The major approach to circumventing this limitation has been to study pre-hemorrhage carriers of penetrant CAA-related APP mutations. Analysis of such carriers of the Dutch-type mutation suggest that reduced functional reactivity, microinfarcts, and WMH may precede CAA-related CMBs or ICH.^{47, 48}

One further application of the Boston Criteria has been as a starting point for formulating diagnostic criteria for the autoimmune syndrome of CAA-related inflammation (CAA-ri).⁴⁹ Being able to diagnose CAA-ri by clinical and imaging features alone is clinically important, as it allows patients to begin immunosuppressive treatment without the morbidity of brain biopsy. Proposed criteria for probable CAA-ri⁵⁰ require hemorrhagic lesions consistent with probable CAA as well as additional clinical and imaging features: presentation with headache, decreased consciousness, behavioral changes, focal signs or seizure, and MRI evidence of WMH lesions that are asymmetric and extend to the immediately subcortical white matter. In a validation study,⁵⁰ probable CAA-ri criteria were met by 14 of 17 individuals with pathologically confirmed CAA-ri vs 1 of 37 with pathologically confirmed non-inflammatory CAA, yielding a sensitivity of 82% and specificity of 97%. The high specificity is particularly relevant from a clinical standpoint, as it suggests brain biopsy can be safely avoided in patients meeting probable CAA-ri criteria. The subcortical WMH lesions in CAA-ri have similar appearance to edema-type amyloid-related imaging abnormalities (ARIA-E) observed in association with anti-amyloid immunotherapy trials,⁵¹ suggesting possible common mechanisms.

Future Directions in CAA Diagnosis

The history of the Boston Criteria highlights some broader issues in devising criteria for diseases where definitive tissue diagnosis is often not feasible. One is the inherent tradeoff between sensitivity and specificity, whereby highly sensitive criteria run the risk of false-positive diagnoses and highly specific criteria yield more false-negatives. There is no single correct balance, and indeed different applications may require different priorities. Diagnostic specificity may be the overriding consideration for determining eligibility for research trials, for example, whereas sensitivity may be key to assessing clinical risk for antithrombotic treatment. A second tension is how to balance ease of use versus complexity and comprehensiveness. The probable CAA definition is reasonably straightforward to apply with a single cutoff set at two or more strictly lobar hemorrhagic lesions, but more complex criteria incorporating other lesion categories and biomarkers might improve accuracy and be more useful for research.

A few aspects of probable CAA seem like promising opportunities for improvement. One is incorporating cSS severity as well as presence. Disseminated cSS, defined as involving more than three sulci (Fig. 2B), is associated with clinical markers of disease severity such as recurrent ICH^{52, 53} and post-ICH dementia⁵⁴ as well as other imaging biomarkers such as abundant CSO-EPVS,⁵⁵ suggesting that extent of cSS carries useful diagnostic information. Another area for potential improvement is diagnosing CAA in individuals with hemorrhagic lesions in mixed lobar and deep territories (Fig. 1B), particularly when the lobar CMB greatly outnumber the deep. A recent analysis, for example, defined a CMB ratio of lobar to deep CMB for individuals in this mixed category and found higher ratios to correlate with increasing Pittsburgh Compound B-PET signal,⁵⁶ suggesting likely CAA. Finally, one can imagine improvements to the Boston Criteria via incorporation of non-hemorrhagic imaging biomarkers. Counting severe CSO-EPVS as an additional lesion, for example, enhanced the Boston Criteria’s sensitivity without worsening specificity in a small series.¹⁶ Any incorporation of additional markers will need to account for the possibility noted above that their sensitivity/specificity may vary with CAA presentation (ICH, non-ICH symptoms, or asymptomatic).

As a next step towards updating and improving the diagnosis of CAA, a multicenter effort to update and externally validate the Boston Criteria has recently been undertaken by the International CAA Association. This project will analyze all available clinical and neuroimaging data from individuals age ≥50 with any of the potential CAA-related clinical presentations, MRI imaging, and histopathologic diagnoses. The goal is to produce and validate a data-driven “version 2.0” of the Boston Criteria that will meet the needs of clinicians and investigators and help maintain the rapid pace of progress towards better treatment of CAA.