SaMD: From Side Project to Central Strategy

Software as a medical device (SaMD) has moved from fringe to front-line for many device manufacturers. For medtech readers, the key questions in 2026 are no longer “What is SaMD?” but “How do we build, regulate, and integrate it with capital equipment in a way that scales and survives post-market scrutiny?”

Why This Is Important

Market analyses now estimate the global SaMD segment in the billions, with high-teens annual growth as artificial intelligence (AI)-based decision support and continuous monitoring mature.¹ The regulatory signal is even clearer: as of late 2025, FDA’s public list of AI-enabled medical devices is well over 1,200. The majority are software that either stand alone or operate in close conjunction with imaging, monitoring, or diagnostic capital equipment.²

On the policy side, three themes matter most to device manufacturers:

1. Transparency and classification—FDA maintains a dedicated Artificial Intelligence-Enabled Medical Devices page, cataloging cleared and approved AI devices by date, manufacturer, and intended use. This list includes many SaMD modules that are either embedded in or interoperate with larger device systems.

2. Post-market monitoring and model change—CDRH’s Office of Science and Engineering Laboratories (OSEL) is actively developing methods and tools for post-market monitoring of AI-enabled devices, including techniques to detect input drift, monitor output quality, and understand performance changes over time. 

3. Alignment with broader quality and AI oversight—FDA has finalized the Quality Management System Regulation to harmonize with ISO 13485, directly impacting SaMD development and lifecycle management. Parallel efforts, such as bipartisan policy proposals on health-AI oversight, are exploring registries and tiered post-market requirements for higher-risk AI tools.

In that context, the most interesting SaMD stories for medtech are not pure apps, but intelligent software layers that make installed capital equipment and single-use devices more valuable.

Case 1: Sepsis Prediction SaMD on Top of Hospital Diagnostics

One of the most illustrative examples is Prenosis’ ImmunoScore,³ a machine-learning–based SaMD for sepsis risk prediction, cleared by FDA in 2024. Rather than being tied to a single device, ImmunoScore functions as an overlay on existing diagnostics and clinical data streams.

ImmunoScore analyzes a panel of 22 clinical and laboratory parameters, including standard blood tests and vital signs, to estimate a patient’s risk of progressing to sepsis within 24 hours. Operationally, inputs come from in-hospital analyzers and monitors (e.g., hematology, chemistry, vital-sign monitors) and the EHR. The SaMD runs its model and returns a risk score that is surfaced in the clinical workflow, typically within the EHR. The software converts a heterogeneous fleet of commodity devices and routine labs into an integrated early-warning system. As a result, several strategic points emerge.

Value without new capital equipment—ImmunoScore illustrates how SaMD can turn an installed base of analyzers and monitors into a differentiated sepsis-management solution. For IVD and monitoring manufacturers, similar approaches can increase stickiness and ASP without redesigning capital equipment.

Evidence and endpoints—The authorization hinged on clinically meaningful endpoints (e.g., prediction of organ dysfunction and mortality) rather than pure algorithm performance metrics, aligning the SaMD more tightly with outcome-driven value propositions.

Lifecycle and drift—Because sepsis incidence, case mix, and practice patterns can change, ImmunoScore is exactly the type of AI SaMD that will need ongoing post-market performance monitoring, the scenario FDA’s OSEL work is explicitly targeting. That has implications for how manufacturers design data pipelines, maintenance plans, and predetermined change control plans (PCCPs) for similar products.

The lesson is that SaMD can economically re-platform routine diagnostics, but only if the product is designed from the start for robust data integration and continuous performance oversight.

Case 2: AI-Enhanced Imaging Suites—SaMD and Capital Equipment

Radiology remains the flagship SaMD category. The AI-enabled device list shows a steady stream of clearances where the primary innovation is software layered on MRI, CT, and PET/CT systems. Recent entries include:

• GE Medical Systems’ SIGNA MAGNUS MRI platform—Cleared in October 2024, this system incorporates AI-based software features as part of its radiology suite to assist with image processing and reconstruction.
• Siemens Healthineers’ syngo.via and related applications—Multiple radiology software packages listed in 2025 use AI to support image interpretation, quantification, and workflow automation.

In each case, the scanner capital equipment is the regulated device, but the differentiating function is the SaMD layer. It reconstructs or denoises images to maintain quality at a lower dose or shorter acquisition time. It also automatically detects and flags critical findings for triage. In addition, the SaMD component generates quantitative outputs (e.g., volumes, perfusion maps) that would be infeasible to produce manually.

In a meta-analysis, AI concurrent assistance reduced radiologist reading time by 27.2%. The reading quantity decreased by 44.47% and 61.72% when AI served as the second reader and pre-screening, respectively.⁴ For imaging companies and their suppliers, three points are clear:

1. System-plus-SaMD economics—Capital equipment is increasingly sold with tiered software bundles as a subscription or license model. That shifts revenue toward recurring SaMD and changes how R&D prioritizes features.

2. Tighter regulatory integration—Since many of these SaMD functions are cleared as part of the system, quality and regulatory processes for software and capital equipment must be deeply intertwined. Model changes, software updates, and cybersecurity patches all become regulated events.

3. Post-market responsibility—The AHA, commenting on AI-enabled medical devices, has emphasized the need for ongoing real-world performance monitoring and transparency, particularly for high-impact diagnostic aids. Imaging companies offering AI SaMD will likely face mounting expectations from regulators and hospital customers to demonstrate performance in the field matches pivotal-trial results.

For medtech professionals used to capital equipment-centric roadmaps, these imaging suites exemplify how SaMD now defines competitive differentiation, pricing, and post-market obligations.

Case 3: Foundation-Model CT Triage SaMD—One Model, Many Devices

The third example pushes SaMD into new territory: foundation models. In early 2025, Aidoc announced FDA clearance of a rib fracture triage application built on its CARE1 foundation model, the first FDA-cleared clinical AI device based on a foundation model rather than a narrow, single-task network. Unlike the company-specific examples, Aidoc’s SaMD is capital equipment-agnostic: it works across CT scanners and PACS systems from multiple manufacturers.

Aidoc’s platform sits between CT scanners and radiologist workstations. The CT scanners produce DICOM images as usual. Those images are routed to the Aidoc SaMD, where CARE1 performs automated analysis. Suspected rib fractures are flagged and prioritized within the radiologist’s normal reader environment.

The significance is not just the clinical use case but the model architecture. CARE1 is designed as a general imaging foundation model, pre-trained on diverse datasets and then adapted to specific tasks. Aidoc argues this approach allows new applications to be developed “from years to weeks,” as the model can be fine-tuned rather than trained from scratch.

For medtech stakeholders, foundation-model SaMD raises several strategic and regulatory questions:

1. Cross-device interoperability as table stakes—Since foundation-model SaMD is designed to ingest images from multiple companies and technologies, it naturally sits on top of diverse scanner fleets. That creates both a competitive threat to company-specific solutions and an opportunity for organizations to partner or co-develop.

2. Regulatory complexity—Existing AI SaMD policies are largely written around static or narrowly scoped models. Foundation models, especially if updated frequently, fit awkwardly into that paradigm. FDA and external experts are now debating how to adapt PCCPs, documentation, and post-market monitoring requirements for such systems, including options like public registries and risk-tiered oversight.

3. New post-market obligations—As FDA’s AI regulatory science program makes clear, high-impact AI SaMD will be expected to have robust mechanisms for input drift detection, performance surveillance, and explainability. For foundation models serving multiple indications and device types, satisfying those expectations will be significantly more complex than for a single-task tool.

In short, a foundation-model SaMD like CARE1 signals a shift from “one model, one device, one use case” to platform AI spanning devices and indications.

The Medi-Vantage Perspective

From these cases, several themes recur:

1. SaMD is a primary value driver for capital equipment. Whether it’s turning standard labs into sepsis-predictors or making scanners faster and more informative, the software layer is increasingly where differentiation and margin reside.

2. Lifecycle management is the new battleground. With FDA investing in post-market monitoring science and calls for risk-based AI oversight, SaMD cannot be treated as a “ship once” feature. Manufacturers need plans for data collection, drift detection, and evidence refresh baked into design and business models.

3. Integration and workflow often matter more than raw AI performance. The most impactful SaMD products integrate seamlessly with existing technologies, minimizing friction for clinicians. For many hospital customers, usability and interoperability outweigh small differences in areas under the curve or sensitivity.

4. Platform decisions can’t be deferred. As foundation-model SaMD matures, device manufacturers must choose whether to build proprietary AI platforms, partner with cross-device SaMD vendors, or focus on being “AI-ready” with open interfaces and robust data pathways.

For medtech’s capital equipment-savvy audience, SaMD is no longer just a digital adjunct. It has become the intelligence layer that will shape competitive positioning, regulatory exposure, and post-market performance expectations for the next generation of medical devices.

References

1. tinyurl.com/mpo260501
2. tinyurl.com/mpo260502
3. tinyurl.com/mpo260503
4. tinyurl.com/mpo260504

MORE FROM THIS AUTHOR—The Selective Capex Comeback: Medtech Capital Sales in 2026

Maria Shepherd has more than 20 years of experience in marketing in small startups and top-tier companies. She founded Medi-Vantage, which provides marketing and business strategy for the medtech industry. She can be reached at [email protected]. Visit her website at www.medi-vantage.com.