Page 140 - AIH-2-4

Artificial Intelligence in Health Artificial intelligence app for EVD navigation

Figure 5. Example of a dodecahedron attached to an external ventricular drain stylet. Sample training data/validation data are shown with red segmentations;
testing data are shown with blue bounding boxes. The model successfully tracked and localized the different markings on the dodecahedron.

technique. 41,50 While experienced providers may not EVD placement can cause parenchymal hemorrhage,
require such assistance in straightforward cases, AI-based intracranial hypertension from occlusion, or delayed
navigation could improve safety and efficacy for trainees CSF drainage—all potentially preventable with better
and in patients with challenging anatomy. By improving visualization.
first-pass accuracy, AI-assisted systems like ours have the Moreover, once adopted, this technology could also
potential to significantly reduce downstream complications serve as a real-time procedural documentation tool. By
and associated costs. capturing trajectory data, timestamps, and alignment
Equally important is the role of AI in medical education. metrics, the application could offer medico-legal
Systems equipped with explainable AI features can serve protection for providers and support quality improvement
not only as navigational tools but also as digital mentors— initiatives. It could further contribute to a growing body
offering real-time procedural feedback, recording attempts of procedural analytics that may be mined for insights
for later analysis, and integrating with curricula to track into improving technique, developing personalized risk
skill acquisition longitudinally. This capacity aligns profiles, and enabling population-level outcome modeling
with emerging research on competency-based training through federated learning frameworks. In the future,
frameworks that leverage AI for both assessment and real-time procedural metrics could be incorporated into
remediation. credentialing, maintenance of certification, and residency
milestone assessments.
Looking ahead, our findings support the argument for
more democratized, hardware-agnostic AI integration Although current navigation systems enhance surgical
into surgical care. Unlike legacy stereotactic systems that safety, they also have significant limitations. These systems
cost hundreds of thousands of dollars, require sterilized are large and bulky, occupying valuable space in operating
hardware, and demand specialized personnel, our mobile or procedural rooms. They often rely on rigidly attached
solution operates on a standard smartphone at no additional reference arrays that are registered only at the beginning
cost once deployed. This approach not only reduces barriers of a case, making them prone to errors if anatomical
to implementation in large hospital systems but may also shifts occur or if the patient’s position changes relative to
4
revolutionize emergency neurosurgical interventions in the reference array. In addition, traditional navigation
remote or battlefield environments. systems require substantial user interaction, which can
potentially introduce operator error and inconsistencies
A major design goal of this project was to maximize during critical steps. In contrast, the automated and
6
ease of use, accuracy, and portability. By leveraging standardized nature of this study’s custom iOS application
commercially available iOS hardware, the system enables minimizes user-dependent variability and enables more
real-time, continuous patient registration and a full consistent, accurate navigation through continuously
navigation experience without the need for complex, updated registration of non-immobilized subjects. The iOS
expensive, or proprietary hardware. The integration application also reduces the need for additional personnel
of AI-based navigational tools into bedside workflows or large, costly equipment, making navigation feasible
therefore supports not only improved accuracy but also in bedside and space-constrained settings where it was
enhanced procedural safety. For instance, misdirected previously impractical or impossible.

Volume 2 Issue 4 (2025) 134 doi: 10.36922/aih.8195