When a new AI tool enters your workflow, a reasonable question follows immediately: can I trust this? AI is appearing in more clinical settings, but the terminology used to describe how it works, open loop, closed source, deep learning, can feel opaque. These are not marketing terms. They describe real design choices that affect how a system behaves, what it can and cannot do, and why it performs predictably.
What Is AI, and How Does It Work in a Medical Device?
Traditional clinical software follows explicit rules: if this value, then that action. AI works differently. Rather than being programmed with a fixed set of instructions, an AI system is trained on data. It recognizes patterns across many examples and applies those patterns to analyze new inputs.
Deep learning is a specific type of AI that uses layered neural networks to identify increasingly complex features within data. A deep learning model trained on images does not simply look for brightness or color, it learns to distinguish subtle characteristics that human observation might miss or apply inconsistently across physicians. This is the same underlying pattern-recognition approach discussed in Second-Degree Burn Healing Stages, which covers the visual assessment limitations this kind of tool is designed to address.
What Is Closed Source AI?
“Closed source” describes where the algorithm lives and who can access it. In a closed source AI system, the underlying code, model architecture, training pipeline, and model weights are proprietary. They are not publicly visible or modifiable by anyone outside the organization that built them.
In plain terms, no one outside Spectral AI can see or change how the AI works.
The FDA’s Software as a Medical Device (SaMD) pathway requires developers to document, validate, and control the AI system submitted for clearance. That means the algorithm is fixed, tested, and reviewed before use in clinical practice, rather than an open system where outside parties can introduce changes.
DeepView’s AI algorithm is closed source. The algorithm is trained on a defined dataset and is not accessible for external modification.
What Is Open Loop AI?
“Open loop” describes how an AI system handles feedback. In an open loop system, the AI takes an input, produces an output, and stops there. That output, and whatever happened after the physician acted on it, does not feed back into the model to change its future behaviour.
Consider the contrast: a continuous glucose monitor paired with an insulin pump that automatically adjusts delivery is a closed loop system. It measures, adjusts, and re-measures in a continuous cycle. A tool that generates an output and hands it to a physician, without that output altering the system’s future behaviour, is open loop.
Closed loop systems have shown performance gains in contexts where the AI controls an actuator directly. In anesthesia delivery, a meta-analysis found that closed loop systems increased time within target physiological ranges by 17.4 percent compared to manual control, with lower induction doses (Wingert et al., Anesthesiology Clinics, 2021). That gain depends on a direct, continuous link between sensor and actuator.
Diagnosis is a different context. The actuator is a physician, not a machine. Introducing automatic feedback cycles into a diagnostic AI raises safety, regulatory, and liability questions that remain unresolved industry-wide (Wingert et al., 2021). The consistency of open loop AI also supports validation: a model that behaves the same way every time can be measured, documented, and reviewed against a fixed standard.
For a point-of-care diagnostic tool, open loop is a design category well suited to current SaMD regulatory pathways.
How DeepView Is Designed: Open Loop, Closed Source, physician in Control
DeepView brings both principles together at the point of care.
The workflow: the physician scans the wound with the handheld device. The scan is stored locally on the device’s PC in an encrypted folder. The AI analyzes the captured images and generates a healing potential prediction, and that result is delivered to the physician. At the time of capture, the scan is not used to train or update the model. Scans collected through separate clinical studies may inform future model development.
The physician makes the final decision. DeepView is positioned as an adjunct that provides non-invasive tissue data to support clinical judgment. This reflects what Penn LDI describes as being “on the loop,” where the AI provides a data layer and the physician retains full decision authority (Meisel, 2025).
Why This Design Matters for Patient Safety and Clinical Trust
Predictable performance
Because a fixed model does not change behavior based on new inputs after deployment, its output remains consistent across uses.
Regulatory accountability
Closed source, FDA-regulated AI is documented and reviewed through a defined regulatory process rather than functioning as a self-modifying experimental system.
Physician remains central to the decision
Open loop design keeps the physician’s judgment central to the decision. The output from the AI system gives the user additional information rather than acting on it autonomously. Research from Penn LDI cites one Cedars-Sinai study in which AI-assisted recommendations were rated optimal 77 percent of the time compared to 67 percent for unaided physician decisions, an example of AI improving the information available to a physician rather than replacing physician judgment (Meisel, 2025).
Frequently Asked Questions
What is open loop AI?
Open loop AI takes an input, generates an output, and does not use feedback from the outcome to adjust future outputs. In a clinical context, this means the model produces an output that provides additional information for the physician to use in their decision making process. It does not update based on the previous inputs into the system or treatment decisions that follow.
What is closed source AI?
Closed source AI is a system where the underlying algorithm, training data, and model architecture are proprietary and not publicly accessible. In regulated medical AI, closed source is compatible with regulatory accountability: the algorithm is fixed and subject to review before clinical use.
Does closed source AI meet regulatory requirements for clinical use?
Closed source AI submitted through the FDA’s Software as a Medical Device pathway is documented and reviewed as part of that clearance process. Closed source describes who can access the algorithm; it is not, on its own, a safety or efficacy claim. Regulatory clearance is a separate, defined process.
Does DeepView learn from new patient scans?
New scans captured at the point of care are not added to the AI’s training database at the time of capture. Scans collected through clinical studies may be incorporated into future training datasets under separate research protocols.
Who makes the final clinical decision when using DeepView?
The physician . DeepView provides an AI prediction of non-healing burn tissue; the healthcare provider makes all treatment decisions. The device is positioned as a decision-support tool, not an autonomous actor.
How was DeepView’s AI algorithm trained?
DeepView’s AI algorithm was trained on a database of burn wound outcomes, using an established truthing process to label tissue characteristics, including oxygen saturation, thrombosis, and reticular collagen degeneration, associated with specific healing outcomes.
What is the difference between open loop and closed loop AI in healthcare?
In closed loop AI, the system measures an outcome, adjusts its output or an intervention, and re-measures in real time, as in automated insulin delivery systems. In open loop AI, the system generates an output and does not use that output or any resulting feedback to alter its future behaviour. For diagnostic and decision-support tools, open loop design is well suited because it keeps the physician central to every decision and supports validated, predictable AI performance.
Key Takeaways
- Open loop AI generates an output and does not use that output or any resulting feedback to alter future behaviour. This supports consistent, predictable performance.
- Closed source AI in regulated medical devices means the algorithm is proprietary and reviewed through the regulatory process, not that it is inaccessible or unaccountable.
- DeepView is positioned as an adjunct: it provides non-healing wound data, and the physician makes the final decision.
Learn more about DeepView or request a demo.
References
Meisel, Z. F. (2025, September 22). In the loop or on the loop: The conundrum of AI clinical decision support. Penn LDI. https://ldi.upenn.edu/our-work/research-updates/in-the-loop-or-on-the-loop-the-conundrum-of-ai-clinical-decision-support/
Wingert, T., Lee, C., and Cannesson, M. (2021). Machine learning, deep learning, and closed loop devices, anesthesia delivery. Anesthesiology Clinics, 39(3), 565 to 581. https://doi.org/10.1016/j.anclin.2021.03.012