Background

Patients with spinal pathology, especially lumbar spinal stenosis, are often given incon- sistent treatment recommendations, creating confusion and anxiety in trying to determine the best path that will optimize their outcome. Our goal is to investigate AI approaches, allowing patients to interpret their overall clinical condition, including their previous imaging studies, diagnostic tests, and other treatments and results, to determine and ana- lyze potential treatment strategies.

The objective is to develop a prototype of an AI-driven system that integrates evidence-based literature, scoring systems, imaging results, and clinical information to sup- port decision-making in the initial clinical context of lumbar spinal stenosis including lum- bar spondylolisthesis. This phase will involve a comprehensive literature review and user interviews with clinicians and patients to identify current gaps in decision-making

Current State

We aim to develop the Spine AI solution on top of the ‣ app. We aim to extend it with a sophisticated RAG-based functionality and support for spine surgery-related FHIR resources.

Future Directions

Our initial phase will focus on the core technology for interpreting radiology reports and other relevant diagnostic insights, while conducting an in-depth needs finding for patients and practitioners focusing on spine surgery. We aim to build a prototype using the Stanford Spezi software ecosystem to automate patient history intake for initial assessments and provide AI-driven recommendations based on the integrated data.

Similarly, we plan to conduct a small usability study with patients and clinical experts to evaluate the spine surgery-focused prototype and gather feedback. The team aims to produce a publication summarizing the findings, establish a robust backend for future mobile applications, and identify the most suitable AI model and system architecture for further development in subsequent phases.

Important Skills

1. Working on the application-side of the project: The SpineAI application and it’s foundations are all written in Swift and SwiftUI. The application is built based on Stanford Spezi; you will be using different Spezi modules and components. Please check out the LLMonFHIR application; familiarity with the components and structure used in the application can be helpful.
2. If you are interested in supporting the clinical trial: Familiarity with submitting, iterating, and communicating with the institutional review board (IRB) to facilitate an ideal study design. Familiarity with collecting patient feedback, running clinical trials, and obtaining structured data based on usability metrics and assessing the effectiveness of the LLM-based research projects.

First Steps

<aside> 📧 Our Choose Your Research Project page provides a great overview of how to engage with your supervisors. The following first steps will help you to get familiar with the project and demonstrate your existing skills.

</aside>

• We encourage you to do 2-3 smaller pull requests in LLMonFHIR, SpeziLLM, SpeziHealthKit, SpeziFHIR, or Spezi Questionnaire to get familiar with the project and demonstrate your ability to work with Swift and the Spezi modules. The Spezi Project board provides a great overview of open issues in Spezi and great ways to get started with good first issues and issues. The LLMonFHIR repository contains a good overview of open issues in the LLMonFHIR project.
• If you are interested in supporting us in conducting a usability study and evaluation of these applications share your existing experience in clinical trials, IRB processes, and data analysis and identify a proposed study design to demonstrate your skillset in supporting the evaluation work for our LLM-based EHR-integrated research projects.

After completing these initial steps, please reach out to Paul Schmiedmayer to discuss how you can become more actively involved in the project and contribute to the research.