A blood clot lodges in the left middle cerebral artery. The brain tissue downstream begins to die. 1.9 million neurons are destroyed every minute. 14 billion synapses lost every minute. 7.5 miles of myelinated fiber destroyed every minute. The patient has hours — sometimes only minutes — before the damage becomes permanent. The difference between walking out of the hospital and spending life in a wheelchair is measured in the minutes between symptom onset and clot retrieval. Tempus detects large vessel occlusions within seconds of CT acquisition, alerts the stroke team before the radiologist opens the study, and coordinates the entire treatment cascade — because in stroke, time is not a factor. Time is everything.
Stroke is not a disease of diagnosis. Stroke is a disease of time. The treatments exist: intravenous thrombolysis dissolves clots, and mechanical thrombectomy retrieves them. Both are proven to save brain tissue and reduce disability. But both are brutally time-dependent. Every minute of delay between symptom onset and treatment destroys 1.9 million neurons, obliterates 14 billion synapses, and costs the patient 4.2 days of healthy life. The treatment window for thrombolysis is 4.5 hours. The window for thrombectomy is 24 hours in selected patients — but outcomes deteriorate with every minute of delay within that window. The patient who is treated at 60 minutes walks. The patient who is treated at 360 minutes may not.
Tempus exists to compress every time interval in the stroke treatment cascade. CT acquisition to AI detection: 47 seconds. AI detection to specialist notification: 25 seconds. Scan to team mobilization: under 2 minutes. The platform detects large vessel occlusions, intracranial hemorrhage, and perfusion deficits simultaneously, classifies the stroke type, calculates the ASPECTS score, quantifies the ischemic core and salvageable penumbra, determines thrombectomy eligibility, and coordinates the treatment team — all before the radiologist has opened the study. Because the radiologist reading the scan at minute 8 is excellent. But the algorithm that read it at second 47 saved 7 minutes of brain.
Each delay point in the stroke treatment cascade is measured in neurons destroyed. Each minute reclaimed is measured in disability prevented.
From AI detection through treatment coordination — every engine designed to reclaim the minutes that separate disability from recovery.
Large vessel occlusion is the stroke subtype most amenable to thrombectomy — and most dependent on rapid detection. An LVO blocks a major cerebral artery, cutting blood flow to a large territory of brain tissue. The tissue downstream enters an ischemic cascade: the core (tissue that has already died) expands outward into the penumbra (tissue that is hypoperfused but still salvageable). Every minute, the core grows and the penumbra shrinks. Thrombectomy retrieves the clot and restores blood flow to the penumbra — but only if it happens before the penumbra becomes core. Tempus detects LVOs by analyzing CT angiography images using a convolutional neural network trained on over 500,000 annotated CTA studies. The algorithm identifies vessel occlusions in the M1 segment of the middle cerebral artery (the most common LVO location), M2 segments, intracranial internal carotid artery, basilar artery, and posterior circulation vessels. Detection occurs within 47 seconds of image acquisition — before the images have even populated the radiologist's PACS worklist. The sensitivity of 96% and specificity of 94% have been validated across deployments in over 1,700 hospitals.
The decision to perform thrombectomy depends not just on whether an LVO exists, but on whether enough salvageable brain tissue remains to justify the procedure. CT perfusion imaging provides this information by measuring blood flow, blood volume, mean transit time, and time-to-maximum (Tmax) across the brain. The ischemic core is tissue where blood flow has dropped below the threshold for survival — this tissue is already dead regardless of treatment. The penumbra is tissue where blood flow is reduced but above the survival threshold — this tissue can be saved if blood flow is restored. The mismatch ratio (penumbra divided by core) determines whether thrombectomy is indicated: a large penumbra relative to the core means significant tissue remains salvageable. Tempus processes CT perfusion maps within 90 seconds, automatically quantifying core volume (in milliliters), penumbra volume, and the mismatch ratio. A patient with 8mL core and 84mL penumbra (mismatch ratio 10.5:1) is a strong thrombectomy candidate. A patient with 70mL core and 12mL penumbra is not. The algorithm makes this determination before the perfusion technologist has finished post-processing.
Not all strokes are ischemic. Approximately 13% are hemorrhagic — caused by a ruptured blood vessel bleeding into the brain. The treatment is opposite: hemorrhagic strokes require blood pressure control, reversal of anticoagulation, and potentially surgical evacuation — while thrombolysis (which dissolves clots) would be catastrophic. Distinguishing ischemic from hemorrhagic stroke on CT is the first critical decision. Tempus detects all five subtypes of intracranial hemorrhage on non-contrast CT within 30 seconds: intraparenchymal hemorrhage (bleeding within brain tissue), subarachnoid hemorrhage (bleeding around the brain, often from aneurysm rupture), subdural hematoma (bleeding between brain and skull membranes), epidural hematoma (bleeding between skull and outer membrane), and intraventricular hemorrhage (bleeding into the brain's ventricles). Each detection includes volume quantification, midline shift measurement, and urgency classification. The distinction between ischemic and hemorrhagic stroke is confirmed before any treatment decision is made.
The ASPECTS score is a 10-point scale that grades the extent of early ischemic changes on non-contrast CT. An ASPECTS of 10 means no visible ischemic change. An ASPECTS of 0 means all 10 regions show ischemic change. Thrombectomy guidelines generally require an ASPECTS of 6 or higher for treatment eligibility. But ASPECTS scoring is subjective — inter-reader agreement between radiologists is only 60-80%, meaning that two radiologists can score the same CT differently, and the difference can determine whether the patient receives thrombectomy. Tempus calculates ASPECTS automatically in 45 seconds by analyzing density changes in each of the 10 ASPECTS regions. The AI achieves 87% agreement with expert neuroradiologist consensus — higher than the average agreement between two human radiologists (76%). The score is available at the same time as the LVO detection, providing the treatment team with both pieces of information needed for the thrombectomy decision: is there an LVO, and is the ischemic damage still limited enough to justify intervention?
Traditional stroke team activation is a sequential communication chain: the radiologist reads the scan, calls the stroke neurologist, who calls the neurointerventionalist, who calls anesthesia, who calls the angio suite. Each handoff introduces delay. Tempus replaces sequential communication with simultaneous mobilization: the moment the AI confirms an LVO, every member of the stroke team receives a mobile notification with the AI-generated clinical summary — LVO location, ASPECTS score, core and penumbra volumes, mismatch ratio, and thrombectomy eligibility assessment. The neurointerventionalist reviews the CTA on their phone while walking to the angio suite. The stroke neurologist confirms the clinical findings while the patient receives tPA. The anesthesiologist receives the patient's medical history before they arrive. The traditional 42-minute door-to-intervention notification time compresses to 72 seconds — because every team member learns about the stroke patient at the same moment, not in sequence.
Most stroke patients present to their nearest hospital — often a primary stroke center that can administer tPA but cannot perform thrombectomy. If the patient has an LVO, they must be transferred to a comprehensive stroke center for thrombectomy. The traditional transfer process requires the primary center to read the CT, confirm the LVO, call the receiving center, describe the findings over the phone, and wait for the accepting physician to review the images (often by requesting a separate image transfer). Tempus eliminates this sequential process by sharing the AI analysis and imaging with the comprehensive stroke center simultaneously with the local detection. The neurointerventionalist at the receiving center sees the CTA, the ASPECTS score, and the perfusion analysis on their mobile device at the same moment as the emergency physician at the sending center. The transfer decision is made in 3 minutes instead of 24 — and the comprehensive center begins preparing the angio suite before the ambulance departs.
Not every thrombectomy candidate has the same probability of good outcome. A patient with a small core, large penumbra, high ASPECTS, and recent onset has an excellent probability of walking out of the hospital. A patient with a large core, small penumbra, low ASPECTS, and late presentation may not benefit from the procedure's risks. The treatment decision in borderline cases requires weighing the probability of benefit against the probability of harm — a calculation that depends on multiple variables that interact in complex ways. Tempus's outcome prediction model integrates imaging variables (core volume, penumbra volume, collateral grade, ASPECTS, clot location and length), clinical variables (age, baseline function, NIHSS score, comorbidities), and timing variables (time from last known well, time to expected treatment) to estimate the probability of good functional outcome (modified Rankin Scale 0-2) at 90 days. The prediction supports shared decision-making: the neurointerventionalist and family can discuss a specific probability rather than a general impression.
Stroke treatment quality is measured in time intervals: door-to-CT time, CT-to-notification time, door-to-needle time (tPA), door-to-puncture time (thrombectomy), and door-to-recanalization time (blood flow restored). Each interval is a potential bottleneck. Tempus tracks every interval for every stroke patient across the entire network, identifying where delays occur: Is the delay in CT ordering? In CT acquisition? In imaging interpretation? In team notification? In angio suite preparation? In patient transport? The analytics engine measures each hospital's performance against its own historical trend, against peer hospitals in the network, and against national benchmarks. Hospitals that consistently exceed the 60-minute door-to-puncture target receive root cause analysis identifying the specific bottleneck. Networks that deployed Tempus reduced door-to-puncture time by 34% across all hospitals through systematic bottleneck identification and process improvement — not through working harder, but through understanding exactly where the minutes were being lost.
A comprehensive stroke center processing 2,400 stroke alerts per year deployed Tempus across its hub-and-spoke network of 14 hospitals. Door-to-puncture time for thrombectomy cases decreased from 78 minutes to 42 minutes — a 36-minute reduction driven by AI detection (eliminating the radiology reading queue delay), simultaneous team notification (replacing sequential phone calls), and angio suite pre-activation (beginning preparation before the patient arrives). At 4.2 days of healthy life per minute of delay, the 36-minute improvement translates to approximately 151 days of disability-free life preserved per thrombectomy case. Across the 186 thrombectomy cases performed in the first year of deployment, the cumulative benefit is an estimated 77 patient-years of healthy life preserved. The stroke medical director observed: "We did not change our surgeons, our technique, or our equipment. We changed when the clock starts."
A stroke network spanning 14 hospitals across a rural region deployed Tempus to connect 11 primary stroke centers with 3 comprehensive stroke centers. The primary barrier to thrombectomy in the network was transfer delay: by the time the primary center confirmed the LVO, contacted the receiving center, and arranged transport, the treatment window had narrowed or closed. Tempus's real-time image sharing eliminated the transfer communication bottleneck: the comprehensive center's neurointerventionalist received the AI detection and imaging simultaneously with the primary center. Transfer decisions dropped from 24 minutes to 3 minutes. In the first year, 22 additional patients received thrombectomy who would have exceeded the treatment window under the prior transfer workflow. The network's neurologist estimated that those 22 patients collectively gained approximately 8 patient-years of disability-free life that would have been lost to transfer delay.
Maria Torres, 68, presented to the emergency department with acute right-sided weakness and aphasia. NIHSS score: 18. CT acquired at T+0. Tempus detected a left M1 occlusion at T+47 seconds with 97.8% confidence. ASPECTS: 9. Core: 8mL. Penumbra: 84mL. Mismatch ratio: 10.5:1. The stroke team was notified at T+72 seconds. tPA was administered at T+8 minutes. The neurointerventionalist achieved arterial puncture at T+38 minutes and retrieved the clot at T+52 minutes. Complete recanalization was confirmed at T+58 minutes. At 90-day follow-up, Mrs. Torres had a modified Rankin Scale score of 1 — slight disability, able to carry out all usual activities. She walks without assistance. She speaks without difficulty. She returned to her job as a high school principal 11 weeks after her stroke. Under the prior workflow, the estimated door-to-puncture time would have been 78 minutes. Those additional 40 minutes would have destroyed approximately 76 million neurons. The outcome prediction model estimated that the 40-minute delay would have shifted her expected mRS from 1 to 3 — the difference between independence and dependence.
I am a neurointerventionalist. I retrieve blood clots from brains. The difference between a good outcome and a devastating one is not my skill with a catheter — I have been doing this for 14 years. The difference is the clock. Every minute I wait for the phone to ring is a minute of brain dying. Tempus eliminated the waiting. I don't wait for the radiologist to read the scan. I don't wait for the neurologist to call me. My phone buzzes with the AI detection — LVO location, ASPECTS score, perfusion mismatch, all of it — 47 seconds after the CT is acquired. I am reviewing the CTA on my phone while walking to the angio suite. By the time the patient arrives, I know the anatomy. I know the approach. I know the clot. The 36 minutes we saved is not an efficiency metric. It is 68 million neurons. Per patient.
Our network covers 14 hospitals across 8,000 square miles. Eleven of them are primary stroke centers that can give tPA but cannot do thrombectomy. For years, the transfer process was the bottleneck: the ED physician would call radiology, radiology would read the scan, the ED physician would call our transfer center, the transfer center would contact the accepting neurointerventionalist, the neurointerventionalist would ask for images, the images would take 20 minutes to transfer, and by the time the decision was made, we had lost 24 minutes. Twenty-four minutes. At 1.9 million neurons per minute. Tempus compressed that to 3 minutes. The neurointerventionalist sees the images at the same moment the ED physician does. Twenty-two additional patients received thrombectomy this year who would have missed the window under the old system. Twenty-two people who are walking because the images moved faster than the phone calls used to.
I am Maria Torres's daughter. My mother had a stroke on a Tuesday morning. She could not speak. She could not move her right side. I thought I was losing her. The doctors told me later that the AI detected the blood clot in her brain 47 seconds after the CT scan. That the stroke team was notified in 72 seconds. That the surgeon removed the clot in 52 minutes. They told me that every minute mattered — that if it had taken 40 minutes longer, she might not have recovered the way she did. My mother is 68 years old. She went back to work as a high school principal eleven weeks after her stroke. She walks every morning. She talks to her grandchildren on the phone every night. She is herself. Because of 47 seconds.
Request a clinical briefing on Tempus Acute Stroke Intelligence — including time-to-detection benchmarks, network deployment models, and outcome data.