Ischemia and Perfusion

When the Blood Stops: Imaging a Race Against the Clock

Acute ischemic stroke is the clinical situation in which MRI most directly changes whether a patient walks out of the hospital. The brain consumes roughly 20 percent of the body's oxygen on about 2 percent of its mass, and it holds essentially no energy reserve. Within seconds of arterial occlusion the affected tissue begins to fail, and within minutes a cascade of ionic and osmotic events writes itself into the diffusion signal. The radiologist's task is not merely to detect the infarct but to read, from the interplay of diffusion and perfusion, how much brain is already dead, how much is dying but salvageable, and therefore whether reopening the artery will help or harm. This lesson follows that physiology from the synapse to the treatment decision.

Two companion lessons developed the physics in isolation: diffusion derived the Stejskal-Tanner $b$-value and the apparent diffusion coefficient, and perfusion and angiography derived the tracer-kinetic basis of dynamic susceptibility contrast. Here we fuse them into the clinical reasoning chain that defines modern stroke care, because in ischemia diffusion and perfusion are not two separate measurements but two halves of a single hemodynamic story.

The Ischemic Cascade and the Origin of Cytotoxic Edema

Cerebral blood flow in normal gray matter is about 50 to 60 mL per 100 g per minute. As flow falls, function fails before structure does. Below roughly 20 mL per 100 g per minute the electroencephalogram flattens and the neuron is electrically silent but still viable: this is the classic penumbra, defined by Astrup and Symon as tissue between the threshold for electrical failure and the lower threshold for membrane (ionic) failure. Below about 10 to 15 mL per 100 g per minute the membrane pumps themselves fail and the tissue proceeds rapidly toward infarction. The penumbra is therefore a flow-defined, time-dependent band, not a fixed anatomic structure.

The molecular engine of infarction is energy failure. Oxidative phosphorylation halts, adenosine triphosphate (ATP) is exhausted within minutes, and the sodium-potassium ATPase can no longer maintain the transmembrane ionic gradients. Sodium and calcium pour into the cell while potassium leaks out; the loss of the osmotic gradient draws water from the extracellular space into the cells, which swell. This is cytotoxic edema (more precisely, cellular or oncotic edema). Critically, in pure cytotoxic edema the total tissue water content does not yet change; water has merely shifted compartments. That distinction is the entire reason diffusion imaging, not T2 imaging, detects the hyperacute infarct.

1. Energy failure: occlusion halts oxidative phosphorylation; ATP falls within minutes.
2. Ionic pump failure: the sodium-potassium ATPase stalls; sodium and calcium enter the cell, potassium exits.
3. Osmotic water shift: water follows the ions intracellularly; cells swell (cytotoxic edema) while total tissue water is initially unchanged.
4. Diffusion restriction: the extracellular space narrows and becomes tortuous, intracellular viscosity rises, and net water mobility falls, dropping the ADC.
5. Excitotoxicity and infarction: glutamate release and calcium overload trigger enzymatic cell death; later, blood-brain barrier breakdown adds vasogenic edema and the total water content finally rises (T2 prolongation).

Why does the compartment shift restrict diffusion? Three mechanisms act together. First, the extracellular space, normally the low-tortuosity highway for water self-diffusion, shrinks from about 20 percent of tissue volume toward perhaps 10 percent, so the dominant fast-diffusing pool becomes a narrow, tortuous maze. Second, water moving into the intracellular space encounters higher viscosity, abundant macromolecules, and organelle membranes that hinder its walk. Third, cell swelling and the collapse of transmembrane gradients reduce active cytoplasmic streaming and trans-membrane water exchange. The net apparent diffusion coefficient falls steeply, typically by 30 to 50 percent within the first hour, long before any change is visible on CT or on conventional T2-weighted MRI.

The ADC Time Course and Stroke Dating

Because the diffusion signature evolves in a stereotyped way, the diffusion-weighted image and the ADC map together act as a clock. The DWI signal is the product of genuine diffusion weighting and the underlying T2 of the tissue (recall that the $b = 0$ image carries heavy T2 weighting). The ADC map, having divided out that T2 term, is the quantitative ground truth. Reading the two together lets the radiologist place an infarct in time without any history.

The ADC trajectory has three phases. In the hyperacute and acute phase (minutes to a few days) cytotoxic edema dominates and ADC is markedly low (dark on the ADC map, bright on DWI). Around day 7 to 10 the ADC crosses back through normal: as cells lyse and vasogenic edema accumulates, restricted intracellular water is released and the rising T2 begins to offset the residual restriction. This is pseudonormalization, the single most important and most treacherous point on the curve, because the ADC map can look deceptively normal even though the tissue is densely infarcted. In the chronic phase (weeks to months) tissue breakdown, gliosis and cavitation leave a region of elevated ADC (bright on ADC) approaching cerebrospinal fluid values, with persistent T2 hyperintensity.

| Stage | Approx. age | DWI signal | ADC value / map | Dominant mechanism | | --- | --- | --- | --- | --- | | Hyperacute | 0 to 6 h | Bright (rising) | Low / dark | Cytotoxic edema, falling ADC | | Acute | 6 h to 4 days | Bright (peak) | Low / dark | Maximal cytotoxic edema | | Early subacute | 4 to 7 days | Bright | Low to normal / darkening less | Cell lysis begins, vasogenic edema accrues | | Pseudonormalization | about 7 to 10 days | Often still bright (T2 shine-through) | Pseudonormal / iso | Restriction and T2 effects cancel | | Chronic | weeks to months | Variable to dark | Elevated / bright | Encephalomalacia, gliosis, cavitation |

This clock has direct triage value. A patient who wakes with a deficit of unknown onset cannot be timed by history, yet a DWI-positive, FLAIR-negative mismatch (the infarct visible on diffusion but not yet on fluid-attenuated inversion recovery) implies a lesion younger than roughly 4.5 hours, because measurable T2/FLAIR change lags diffusion by several hours. The WAKE-UP trial used exactly this DWI-FLAIR mismatch to select wake-up stroke patients for intravenous thrombolysis, turning a physiologic time course into a treatment biomarker.

Perfusion: Sizing the Tissue at Risk

Diffusion sizes the core; perfusion sizes the larger volume of hypoperfused tissue. The clinical workhorse is dynamic susceptibility contrast (DSC): a compact bolus of gadolinium chelate is tracked with rapid T2*-weighted echo-planar imaging, one whole-brain volume every 1 to 2 seconds. While the contrast is confined to the vasculature, its susceptibility produces microscopic field gradients that dephase spins throughout the voxel, so even though blood is a small volume fraction the whole voxel transiently darkens. Because the T2* decay is exponential, the change in relaxation rate is proportional to the local tracer concentration, giving a concentration-time curve for every voxel.

Tracer-kinetic theory then converts those curves into hemodynamic maps governed by the central volume theorem, which ties together the three primary parameters.

Each parameter has a precise physical meaning. Cerebral blood volume (CBV) is the fraction of tissue occupied by blood, computed as the area under the tissue concentration curve normalized to the area under an arterial input function (AIF) sampled in a feeding artery; normal gray matter is about 4 to 5 mL per 100 g. Mean transit time (MTT) is the average time a tracer particle spends traversing the voxel's vasculature. Cerebral blood flow (CBF) is the delivery rate, recovered not by simple division but by deconvolution of the tissue curve by the AIF.

The deconvolution step is the crux. The measured tissue curve is the convolution of the AIF with the tissue residue function $R(t)$, the fraction of tracer still present at time $t$ after an idealized instantaneous input, scaled by flow.

Recovering the product $\mathrm{CBF} \cdot R(t)$ requires inverting the convolution, most commonly by singular value decomposition (SVD). CBF is then the peak height of that recovered function, and MTT follows from the central volume theorem. Crucially, SVD also yields the parameter that has come to dominate stroke triage: Tmax, the time to the maximum of the deconvolved residue function. Tmax conflates true delay in tracer arrival (from a proximal occlusion and slow collateral filling) with prolonged transit, and it is robust and easy to threshold, which is why automated software has standardized on it.

The Diffusion-Perfusion Mismatch and Modern Treatment Selection

Putting the two together yields the diffusion-perfusion mismatch, the imaging correlate of the ischemic penumbra. The DWI lesion (low ADC) approximates the irreversibly infarcted core. The perfusion lesion, conventionally the volume of tissue with Tmax greater than 6 seconds, marks the full territory of critical hypoperfusion. The arithmetic difference, perfusion lesion minus diffusion core, is the mismatch volume: hypoperfused but still-viable brain that thrombectomy or thrombolysis might salvage. A large mismatch means much to gain; a small or absent mismatch means the damage is largely done.

This physiology-based selection is the foundation of late-window reperfusion therapy. The landmark DAWN (2018) trial selected patients 6 to 24 hours from last-known-well using a clinical-core mismatch (a large deficit on the NIH Stroke Scale paired with a small core), while DEFUSE-3 (2018) selected patients 6 to 16 hours using an imaging mismatch: a core under 70 mL, a mismatch ratio of at least 1.8, and an absolute mismatch volume of at least 15 mL. Both trials showed dramatic benefit from thrombectomy far beyond the classic 6-hour window, and they rewrote the guidelines: in 2026, eligibility in the extended window is driven by tissue physiology, not by the clock alone.

Selection thresholds have since broadened. Trials of large established cores (for example SELECT2 and ANGEL-ASPECT, 2023) showed that even patients with sizable infarcts can benefit from thrombectomy, pushing the acceptable core volume upward and shrinking the population excluded for being too far along. The thrombolysis literature evolved in parallel: EXTEND used automated perfusion mismatch to extend intravenous alteplase to 4.5 to 9 hours and to wake-up strokes, and tenecteplase has largely supplanted alteplase as the lytic of choice for its single-bolus convenience and at least equivalent reperfusion. The common thread is that imaging-defined viable tissue, not elapsed time, now gates treatment.

Collaterals, Reperfusion Injury and Hemorrhagic Transformation

Why do two patients with the same occlusion at the same time have wildly different cores? The answer is collateral circulation. Leptomeningeal anastomoses and the circle of Willis can sustain penumbral flow above the infarction threshold for hours, slowing core growth. Good collaterals are the physiologic reason a large mismatch can persist into the late window; poor collaterals produce a fast progressor whose core has already consumed the penumbra by the time of imaging. On perfusion maps, robust collaterals manifest as prolonged Tmax with relatively preserved CBV (delay without volume loss), whereas failing collaterals show falling CBV that heralds imminent infarction. Collateral status, whether judged on CT angiography, MRA, or the perfusion maps themselves, is increasingly read as an independent modifier of who will benefit from reperfusion.

Restoring flow is the goal, but reperfusion is not free. Reperfusion injury describes the paradox that returning oxygen to ischemic tissue can accelerate damage: a burst of reactive oxygen species, calcium overload, mitochondrial permeability-transition-pore opening, and inflammatory infiltration compound the original insult. The vasculature suffers too. Ischemia degrades the basal lamina and tight junctions through matrix metalloproteinase activation, so when perfusion pressure returns, blood can extravasate. This is hemorrhagic transformation, ranging from petechial hemorrhagic infarction to a space-occupying parenchymal hematoma.

MRI is exquisitely suited to detect this because deoxyhemoglobin and its breakdown products are strongly paramagnetic. Susceptibility-weighted imaging (SWI) and T2*-weighted gradient-echo sequences reveal blooming hypointensity from microbleeds and early hemorrhagic transformation far more sensitively than CT, and a heavy burden of pre-existing cerebral microbleeds is itself a marker of fragile vasculature that raises the risk of symptomatic hemorrhage after thrombolysis. A very low CBV within the core, signifying profound, prolonged ischemia of the blood-brain barrier, likewise flags tissue prone to bleed once reopened.

Stroke Mimics: When Diffusion and Perfusion Lie

Restricted diffusion is sensitive for stroke but not perfectly specific, and several non-ischemic processes can lower ADC or distort perfusion, the so-called stroke mimics. The expert reasons in Bayesian terms: a lesion respecting an arterial territory in a patient with vascular risk factors and an abrupt deficit is overwhelmingly likely to be ischemic, whereas restriction that crosses territories, spares or follows the cortex atypically, or arises in the wrong clinical context should prompt the mimic differential before a thrombolytic is given.

Seizure and status epilepticus can produce cortical (and sometimes thalamic or hippocampal) restricted diffusion from the intense metabolic demand of sustained neuronal firing, typically with crossed or non-arterial distribution and accompanying hyperperfusion (shortened Tmap and increased CBF) rather than the hypoperfusion of ischemia. Hypoglycemia can cause symmetric restricted diffusion in the splenium, internal capsules, and cortex, reversible if glucose is corrected, and is the reason a fingerstick glucose precedes every stroke code. MELAS (mitochondrial encephalopathy, lactic acidosis and stroke-like episodes) produces stroke-like lesions that characteristically do not respect vascular territories, often spanning posterior cortex with elevated lactate on spectroscopy and, paradoxically, hyperperfusion during the acute episode. Cerebral abscess and dense, highly cellular tumors such as lymphoma also restrict diffusion but are mass lesions with enhancement and a clinical tempo quite unlike acute stroke.

| Mimic | Diffusion pattern | Perfusion | Discriminating clue | | --- | --- | --- | --- | | Acute infarct | Restricted, arterial territory | Hypoperfused, Tmax prolonged | Territory respected, abrupt deficit | | Seizure / status epilepticus | Cortical, crosses territories | Hyperperfusion | Ictal history, gyriform, often reversible | | Hypoglycemia | Symmetric splenium, capsules, cortex | Variable | Low serum glucose, reversible on correction | | MELAS | Posterior, non-territorial | Hyperperfusion acutely | Lactate peak on spectroscopy, maternal inheritance | | Abscess | Markedly restricted pus cavity | Peripheral hyperemia | Ring enhancement, mass effect, fever |

Failure Modes: How the Numbers Mislead

Because so consequential a decision rests on these images, the expert must know precisely how they fail. The errors are partly technical and partly cognitive, and both can cost brain.

• T2 shine-through: a bright DWI with a normal or bright ADC is not restriction; it is the long T2 of the lesion bleeding into the diffusion image. Mistaking shine-through (or a pseudonormalized infarct) for acute restriction misdates the lesion. The ADC map is the arbiter.
• EPI susceptibility and geometric distortion: single-shot echo-planar diffusion and DSC sequences distort and lose signal near air-tissue interfaces (skull base, sinuses, petrous bone), where off-resonance piles up signal along the phase-encode direction and can hide or fabricate a brainstem or inferior temporal lesion.
• Motion on EPI: bulk patient motion during the strong diffusion gradients produces gross phase errors, ghosting and spurious signal that can masquerade as a lesion or corrupt an ADC map; restless or aphasic stroke patients are precisely the ones who move.
• Tmax threshold sensitivity: the penumbra volume depends on the chosen Tmax cutoff. The 6-second threshold is a validated convention, not a law of nature; shifting it, or changing the SVD regularization, the AIF, or the vendor, resizes the mismatch and can move a patient across a treatment boundary.
• AIF and bolus errors: a partial-volume or poorly timed AIF, a fragmented bolus from poor cardiac output, or recirculation can distort every deconvolved map, inflating or collapsing the apparent mismatch.
• Cognitive bias: anchoring on the automated core volume, or premature closure on stroke before checking glucose and the seizure history, are the human failure modes that the territory-respecting versus crossing-territory heuristic is meant to guard against.

The discipline that ties this lesson together is to read diffusion and perfusion as a single hemodynamic narrative checked against the clinical story. Diffusion answers how much is already lost; perfusion answers how much is at risk; their mismatch, validated by the late-window trials and computed in minutes by automated software, answers whether reopening the artery will save brain or merely bleed it. Knowing the assumptions, the thresholds, and the failure modes is the difference between a colored map and a defensible decision to treat.