GE HealthCare is lifting the curtain on five cutting-edge research projects from its 2025 AI Innovation Lab, revealing early-stage innovations that demonstrate where healthcare AI is headed. The initiative accelerates early-concept AI technologies within the company, with this year’s portfolio focused heavily on agentic artificial intelligence for radiology and advanced foundation models for magnetic resonance imaging.
The research comes as healthcare organizations grapple with a stark reality: hospital systems currently harness only about 3 percent of their available data, according to GE HealthCare. With radiologist shortages intensifying—nearly 50 percent of radiologist job searches went unfilled in 2023 even as imaging volumes grow 5 percent annually—the opportunity to unlock transformative insights through AI has become a strategic imperative.
Agentic AI Targets Radiology Workflow Orchestration
GE HealthCare’s most ambitious project involves developing what the company describes as the first agentic AI diagnostic imaging assistant integrated directly into imaging devices and designed specifically for radiology. Unlike traditional radiology AI tools that simply analyze images, this technology aims to create a radiologist orchestration software platform that can process scans, enable natural language interactions, and create interactive reports.
According to the company, the agentic AI technology would go beyond analysis to actively streamline radiologist workflows, allowing clinicians to focus more on patient care than administrative tasks. The platform would leverage large language models, vision language models, and autonomous agents to orchestrate end-to-end workflows. GE HealthCare states it is exploring opportunities to integrate this capability into current clinical workflows through its enterprise imaging solutions.
The research arrives as agentic AI emerges as a paradigm shift in radiology, moving from passive, user-triggered tools to systems capable of autonomous workflow management, task planning, and clinical decision support. While competitors like Intelerad have partnered with RADPAIR to deliver agentic AI for radiology reporting, GE HealthCare’s approach focuses on device-level integration and comprehensive workflow orchestration rather than report generation alone.
MRI Foundation Model Fine-Tuning With Academic Partners
GE HealthCare announced collaborations with Mass General Brigham and University of Wisconsin-Madison to fine-tune its MRI research foundation model announced last year. The model was trained on a dataset of more than 200,000 MRI images from more than 20,000 studies, representing what GE HealthCare characterizes as a first-of-its-kind research foundation model for magnetic resonance imaging.
The collaboration represents a critical step in assessing the model’s adaptability for diverse operational and clinical use cases. Mass General Brigham AI will fine-tune the foundation model for prostate applications, including disease classification, lesion segmentation, and measurement, leveraging PI-RADS scores to adapt the model. University of Wisconsin-Madison will evaluate the model across operational and clinical scenarios including body region detection, image quality control, and contrast agent recognition, while benchmarking performance against other state-of-the-art foundation models.
The research comes as major vendors accelerate MRI foundation model development. Philips and NVIDIA announced a strategic partnership in May 2025 to co-develop a domain-specific foundation model for MRI, targeting what both companies describe as the next generation of intelligent radiology infrastructure. The proliferation of MRI foundation model initiatives reflects growing recognition that general-purpose models require significant domain-specific fine-tuning to deliver clinical value.
Addressing Incidental Findings With Autonomous AI
A third research initiative explores how agentic AI could detect and report incidental findings in computed tomography imaging. According to GE HealthCare, more than 47 percent of abdominal CTs reveal incidental findings—discoveries unrelated to the original imaging purpose, such as kidney cysts found during scans ordered for other reasons.
The company is developing an agentic AI solution designed to identify high-risk lesions, classify findings, and recommend follow-up imaging. According to GE HealthCare, a liver lesion flagged with 90 percent malignancy probability would prompt a recommendation for hepatic MRI follow-up. The AI would compare primary and prior studies to assess lesion progression, with the goal of integrating with existing radiology tools to enable streamlined workflows. Radiologists would retain full control, with the ability to approve findings before sharing them with clinicians. The initial research focuses on liver, spleen, kidney, adrenal gland, bone, and lung nodules.
Energy-Efficient Neural Networks for Sustainable Imaging
GE HealthCare is conducting research advancing sustainable AI in tomographic imaging—including computed tomography, positron emission tomography, and single photon emission computed tomography—by pioneering energy-efficient neural networks designed to consume less computational power.
In tomographic imaging, reconstruction refers to the process of turning data into cross-sectional images. Model-based iterative reconstruction provides impressive image quality but requires significant compute power for repeated calculation cycles. According to the company, its research team is developing an AI technique to reduce iterations, with early results reducing iterations from 40 to just six. Researchers are exploring how hierarchical AI models—similar to how human eyes judge depth—would focus on important details rather than processing entire images, potentially leading to reduced computing power usage.
Generative AI for Field Service Engineers
The fifth project applies generative AI to internal operations, piloting a custom-built multi-modal conversational agent that helps field service engineers—expert technicians who maintain and repair equipment on-site at hospitals—quickly find information and reduce equipment downtime.
According to GE HealthCare, these engineers previously had to manually search through thousands of pages of service manuals, logs, historical records, and knowledge articles. The system synthesizes data and recommends likely fixes while providing source material for quick verification. Building on early success, the company is developing a common generative AI-powered interface to accelerate conversational AI feature development across its product portfolio.
Strategic Context and Market Leadership
The AI Innovation Lab projects showcase groundbreaking work at GE HealthCare, a company applying a 125-year legacy of innovation with what it characterizes as startup energy to address healthcare’s most pressing challenges. The initiatives align with the company’s broader strategy to become cloud-first, AI-powered, and software-enabled, where data becomes a catalyst for discovery and progress in medicine.
GE HealthCare has invested in AI for years and has topped an FDA list of AI-enabled device authorizations for four consecutive years with 100 authorizations. The research projects reflect the company’s commitment to pioneering innovations that reimagine how AI can balance clinical workloads, enhance efficiency, and deliver measurable outcomes.
“At GE HealthCare, we are not just developing AI to address today’s most complex healthcare challenges—we are also investing in new research to anticipate tomorrow’s needs,” said Dr. Taha Kass-Hout, GE HealthCare’s Global Chief Science and Technology Officer. “Our AI Innovation Lab projects offer a behind-the-scenes look at areas that we believe hold real potential to transform care for both patients and clinicians. These projects reflect our commitment to pioneering research that reimagines how AI can be used to balance clinical workloads, enhance efficiency, and deliver measurable outcomes.”
The research underscores a pivotal moment in medical imaging AI, where the industry moves from narrow, task-specific algorithms toward foundation models and autonomous agentic systems capable of orchestrating complex workflows with minimal human intervention. As these technologies mature from research initiatives to commercial products, they promise to reshape how radiologists practice, how imaging departments operate, and ultimately how patients receive care.
— This original article was created with AI support.
