What Made It Into Hospital AI Workflows in 2025

Hospitals in 2025 are using AI in many small, focused ways. The FDA’s list of cleared AI/ML medical devices now exceeds 950 products, with 295 new clearances just in 2025.

Most of these are radiology image analysis tools, but regulators caution against hype. Former FDA Commissioner Robert Califf warns that healthcare AI is still “overhyped,” noting that “I hear way too much about the money… not a lot of human values” in discussions about AI.

In practice, hospitals emphasize narrow AI assistants integrated into existing processes. For example, Kaiser Permanente’s chief digital officer notes that ambient AI scribes (transcription assistants) let physicians focus on patients rather than typing. Despite the surge in cleared devices, experts point out that most AI in hospitals is still limited to specific tasks (imaging triage, risk alerts, documentation, etc.) rather than general-purpose diagnostics.

Indeed, industry reports note that 71% of cleared AI devices are in radiology, and hospital leaders stress that each tool must fit tightly into the workflow. Regulators and clinicians alike call for careful validation, noting evidence gaps for many AI tools, so 2025’s hospital AI mostly consists of targeted analytic aids already shown safe and useful in real settings.

1. Imaging AI for urgent findings

Where it’s used: Advocate Aurora Health (IL/WI), among others.

Advocate Health, a large Midwest health system, embeds FDA-cleared AI algorithms in its radiology workflow. When imaging scans (CT, MRI, X-ray) are acquired, AI software immediately flags critical findings like intracranial hemorrhage or pulmonary embolism.

For example, Advocate is rolling out Aidoc’s platform at dozens of hospitals: radiologists receive instant AI alerts on urgent cases, so dangerous conditions can be reviewed and acted on faster. A radiologist still reads every scan, but AI adds a “second set of eyes” that triages and brings priority cases to attention earlier. This helps shorten emergency response time without replacing the doctor. Advocate’s CDO says these tools, when “deployed at scale… don’t just make health care more efficient, they make it more accurate”

2. Ambient AI scribes in clinics

Where it’s used: Kaiser Permanente (The Permanente Medical Group).

Kaiser Permanente’s Oakland-based Permanente Medical Group uses ambient-recording AI to transcribe doctor-patient visits. Microphones pick up the conversation (with patient consent), and an AI system (Microsoft’s DAX/CoPilot) drafts the medical note in real time. The physician reviews and edits the draft afterwards, so the AI handles most typing.

In practice, this means doctors spend more face-to-face time with patients and less time documenting. The Permanente Medical Group reports that after one year and 2.5 million encounters, physicians saved thousands of hours on notes. The AI doesn’t make clinical decisions (doctors verify every note) but it automates the routine task of note-taking. Hospital leaders say it has reduced physicians’ clerical burden and helped combat burnout.

3. AI-driven sepsis alerts

Where it’s used: Cleveland Clinic (Cleveland, OH).

Cleveland Clinic expanded an AI-based sepsis early-warning system across its hospitals in 2025. The system (Bayesian Health’s Sepsis Watch platform) continuously analyzes patient data – vitals, labs, notes – in real time. It runs inside the electronic health record and flags patients showing early sepsis signs. When the AI detects risk patterns, it immediately alerts clinicians so they can evaluate and start treatment sooner. In a pilot at two hospitals, the alerts helped catch dozens of additional sepsis cases and reduced false alarms.

Nurses and doctors still review each alert (the AI does not order treatments autonomously) but integrates into the ICU workflow to prioritize evaluations. The health system reports that using this AI “improves patient outcomes” by accelerating care for those at ris.

Caveat: Evidence so far is from Cleveland Clinic’s own pilots. Its broader impact across the entire health system is still being assessed.

4. AI in lung cancer screening program

Where it’s used: Emory Healthcare Winship Cancer Institute (Atlanta, GA).

In late 2025 Emory launched a Lung Screening and Nodule Program that incorporates AI imaging. All low-dose chest CT scans from lung cancer screenings are reviewed by AI software for any detected pulmonary nodules. When the AI finds an incidental nodule, it flags the finding and routes the case to a specialized lung nodule clinic.

Thoracic specialists then coordinate prompt follow-up. Emory reports that AI-assisted imaging is one of the advanced techniques (along with robot-assisted bronchoscopy) used to detect lung cancer as early as possible. Patients with small nodules receive faster care coordination because the AI helps ensure these findings aren’t overlooked. The lung screening program integrates this AI step into its clinical workflow, but all final diagnostic and treatment decisions remain with the multidisciplinary cancer team.

5. Digital pathology image review

Where it’s used: Moffitt Cancer Center (Tampa, FL).

Moffitt Cancer Center rolled out a cloud-based digital pathology platform (AISight Dx) to modernize its pathology labs. All glass slides are digitized and managed through this system, which also integrates AI analysis tools. Pathologists use AISight Dx as their primary workspace (viewing high-resolution slide images on-screen) and can launch AI models on-demand. For example, algorithms can highlight regions of interest or quantify tumor markers. The hospital’s physician-in-chief says deploying AISight Dx “enables faster, more accurate pathologist review”. Importantly, all final diagnoses are still made by pathologists; the AI simply assists with flagging features or measuring findings. This system is FDA-cleared for primary diagnosis tasks, meaning the AI meets regulatory standards for clinical use. By unifying image management with AI, Moffitt aims to speed workflow and support collaboration among specialists.

6. AI‑augmented virtual nursing for fall prevention

Where it’s used: Emory Healthcare (Atlanta, GA).

Emory University Hospital Midtown (Atlanta) piloted an AI-powered “virtual nursing” system in 2025. Every patient room on one floor was equipped with LIDAR motion sensors and a telehealth station. These sensors use AI to monitor patient movement patterns. If a patient is about to fall (e.g. trying to leave bed), the system detects it up to 30 seconds in advance and sounds an in-room voice alert, while simultaneously sending a notification to the remote nursing control center.

Meanwhile, off-site “virtual nurses” follow up via video to coach the patient or call a staff member. Emory’s clinical informatics officer explains that this spatial-intelligence solution “helps to enhance patient care…preventing falls and notifying nurses of any concerns”. Nurses still make the final call on each alert, but the AI+telehealth setup allows nurses to supervise many more rooms simultaneously. Emory plans to expand this system to multiple units after evaluating its impact on safety and staffing.

Caveat: So far this is an initial rollout on one floor (32 rooms). The effectiveness and patient outcomes are still under study, and expanded deployment depends on results.

7. AI-assisted colonoscopy polyp detection

Where it’s used: Memorial Healthcare System (Hollywood, FL).

Memorial Healthcare’s colorectal program introduced GI Genius, an FDA-cleared AI module for colonoscopy procedures. During every colonoscopy, the AI runs on the video feed and highlights potential polyps in real time (for example, circling suspicious tissue on the screen). It serves as a “second set of eyes” for the endoscopist, flagging flat or hidden polyps that might otherwise be missed. According to Memorial’s colorectal surgeon, this enhances what the doctors already do by identifying abnormalities during screening. Any flagged polyp is still reviewed by the physician, who decides whether to biopsy or remove it. By integrating seamlessly into routine colonoscopy workflow, GI Genius helps increase the adenoma detection rate. The hospital expects this will improve patient outcomes by catching precancerous lesions earlier.

8. Stroke CT triage with AI

Where it’s used: Henry Ford Health (Detroit, MI).

Henry Ford Health System uses RapidAI, a CT image analysis platform, for acute stroke care. When a suspected stroke patient gets a CT scan, RapidAI instantly analyzes the image for large-vessel occlusion and other critical findings. The AI then pushes the images and alerts to neurologists’ phones and stroke teams, enabling a parallel response.

Previously, the scan had to be read by a radiologist, then a neurologist, delaying treatment. With AI triage, the stroke team can start planning interventions (like thrombectomy) as soon as the scan finishes. Henry Ford reports that using RapidAI has substantially reduced door-to-treatment times and improved coordination among specialists. Neurologists still confirm the diagnosis, but the AI workflow shave critical minutes, which is known to improve patient outcomes.

In 2025 most hospital AI systems remain purpose-built tools that augment clinicians.

They focus on narrow tasks: image analysis, alerting, or documentation, where algorithms can be validated on concrete metrics. Regulators and researchers emphasize the evidence gap in AI: real-world validation is still emerging, and many FDA summaries lack robust trial data. In practice, the AI tools that persist in routine care will be those with clear performance data from clinical pilots.

For example, AI for imaging triage, risk alerts, and paperwork automation have shown measurable workflow benefits. Other promised applications, like predictive analytics broadly in EHRs, are still under study or early in deployment.