Update from Dr. Jennifer E. Thorne on AI Birdshot research
Birdshot Uveitis Society of North America is currently funding a Birdshot Chorioretinitis &Artificial Intelligence study at Johns Hopkins Wilmer Eye Institute. The investigation is led by Dr. Jennifer E. Thorne, and today we are happy to share Dr. Thorne’s update on the first year of the 2-year project.
Special thanks to Dr. Thorne and her colleagues for their consistent support of the birdshot uveitis community. We are thrilled to be part of this important undertaking!
Members of the birdshot community are encouraged to help fund the study. Please see the link below.
Dear BUSNA Members,
I wanted to take the opportunity to write to all BUSNA members and supporters that have helped drive our research into birdshot chorioretinitis and give you an update specifically on the project in artificial intelligence (AI) in birdshot that is being generously funded by BUSNA. My team and I are so grateful for your continued support. This work would not have been possible without your generosity and engagement.
As many of you are aware, BUSNA started this project with an initial donation of $45,000 about one year ago and a pledge for a second year of funding at the same level. I’m happy to say that the work has been steadily moving forward and we have been able to stay within budget comfortably and without delays in workflow.
Aside from the typical 2 to 4-month period of setting up a study such as this, our group was able to get through the Johns Hopkins Institutional Review Board (IRB, a review to ensure the study meets strict ethical guidelines) processing and database design quite quickly, followed by the identification of 200 potential patients with birdshot chorioretinitis (BSCR) to form our original dataset. Of these, 193 patients had usable multimodal imaging for AI analysis. We divided the first project into 3 Phases. Phase 1 was to identify the visit date for each patient when at least 3 tests were performed and Phase 2 was to collect and upload each of these tests into a deidentified dataset for the purposes of piloting a foundational AI model. Phase 2 proved to be more challenging than originally anticipated as each imaging platform has different uploading requirements, so we prioritized the images on the Heidelberg system and will continue to work on processes to increase the use of imaging from other platforms going forward. Phase 3 is to collect the clinical data from the visit linked to the imaging in each patient. Each step in the project requires beta testing to optimize what data are collected and how.
Currently, we are downloading the imaging data of a smaller sample of patients to run 2-3 foundational AI models for model selection before building out the dataset further. My hope is that we can publish an analysis of a single visit for each patient (cross-sectional analysis) and then refine our model with longitudinal data as we have many patients with more than 10 years of imaging data. The AI Center has begun work with the current dataset and I expect we will have some fine-tuning to do once I look at the first analytic run-through, which I anticipate will be completed by the end of August 2026.
Usage of current funds are summarized here and I’ve used round figures. Approximately $5000 went to indirect costs charges by JHU enterprise for research projects (indirect costs are variable by project but for donor funding the percentage is about 12% which is the lowest rate charged). $20,000 was committed to the AI center to cover the costs of analysts’ time, research of foundational AI models and the analysis of dataset. The remaining $20,000 is allotted to cover my time on the project and the time for data collection, which has been quite low because we have been able to utilize a talented and enthusiastic medical student. Of the $45,000, about $4000 (of the $20,000) has not been spent yet.
For the year ahead, I anticipate $5000 for indirects and $15,000 for the AI Center. Of the remaining, $25,000, I plan to earmark $3000-5000 for travel to present results and for publication (ideally making the paper Open Access). Approximately $10,000 will be necessary for data collection with the remainder going to cover my salary/effort on the project. As with the previous year, I can reduce my salary funding from this project to make sure there is adequate funding to cover the other costs because I am able to utilize other funds that I have, so it gives the project some safety from unanticipated added costs.
In our second year, I would like to have a model tested and validated and would love to have much of the longitudinal data collected. Uploading the additional imaging longitudinally will be the most labor-intensive part of the project and likely the part of the project that will take the most time; therefore, it is difficult to gauge how much of the year will be focused on this aspect. During this data collection phase, we will be working on reporting the results of the cross-sectional study in order to move all phases of the research forward as efficiently as possible.
Suffice to say, I’m very pleased with our progress given some of the initial difficulties with uploading images and I’m incredibly excited to continue this work and learn all that we can from our database. I have been speaking with other ophthalmologists interested in AI and birdshot and we are looking for ways to collaborate in this area of research. In the past, multi-center studies in birdshot have been hampered by data sharing and transfer agreements and the time-consuming bureaucracy that accompanies that. However, with AI modeling it may be possible to simply share the model (e.g., no sharing of patient data) and have colleagues test the model on their patient population, thus saving us a lot of time and costs!
Lastly, I want to express my deepest gratitude to BUSNA for your hard work and enthusiasm for my research. AI projects tend to be “high risk” projects as it is unknown what you might find; however, these projects can also be “high reward” as a foundational model can point to biomarkers of the disease that feasibly can be used for early diagnosis or for identifying who needs treatment or for understanding when/if treatment can be stopped. My hope is that we will learn a great deal from a foundational model in birdshot chorioretinitis and with your assistance, we are on our way!
Sincerely,
Jennifer E. Thorne, MD, PhD